WIP

Cellulite integration

crates/dump/src/lib.rs

@@ -158,6 +158,10 @@ pub enum KindDump {
     UpgradeDatabase {
         from: (u32, u32, u32),
     },
+    NetworkTopologyChange {
+        network: Option<meilisearch_types::enterprise_edition::network::Network>,
+        origin: Option<meilisearch_types::tasks::Origin>,
+    },
 }
 
 impl From<Task> for TaskDump {
@@ -240,6 +244,9 @@ impl From<KindWithContent> for KindDump {
             KindWithContent::UpgradeDatabase { from: version } => {
                 KindDump::UpgradeDatabase { from: version }
             }
+            KindWithContent::NetworkTopologyChange { network, origin } => {
+                KindDump::NetworkTopologyChange { network, origin }
+            }
         }
     }
 }
@@ -253,7 +260,7 @@ pub(crate) mod test {
     use big_s::S;
     use maplit::{btreemap, btreeset};
     use meilisearch_types::batches::{Batch, BatchEnqueuedAt, BatchStats};
-    use meilisearch_types::enterprise_edition::network::{Network, Remote};
+    use meilisearch_types::enterprise_edition::network::{DbNetwork, DbRemote};
     use meilisearch_types::facet_values_sort::FacetValuesSort;
     use meilisearch_types::features::RuntimeTogglableFeatures;
     use meilisearch_types::index_uid_pattern::IndexUidPattern;
@@ -544,10 +551,10 @@ pub(crate) mod test {
         RuntimeTogglableFeatures::default()
     }
 
-    fn create_test_network() -> Network {
-        Network {
+    fn create_test_network() -> DbNetwork {
+        DbNetwork {
             local: Some("myself".to_string()),
-            remotes: maplit::btreemap! {"other".to_string() => Remote { url: "http://test".to_string(), search_api_key: Some("apiKey".to_string()), write_api_key: Some("docApiKey".to_string()) }},
+            remotes: maplit::btreemap! {"other".to_string() => DbRemote { url: "http://test".to_string(), search_api_key: Some("apiKey".to_string()), write_api_key: Some("docApiKey".to_string()) }},
             sharding: false,
         }
     }

crates/dump/src/reader/v6/mod.rs

@@ -24,7 +24,7 @@ pub type Batch = meilisearch_types::batches::Batch;
 pub type Key = meilisearch_types::keys::Key;
 pub type ChatCompletionSettings = meilisearch_types::features::ChatCompletionSettings;
 pub type RuntimeTogglableFeatures = meilisearch_types::features::RuntimeTogglableFeatures;
-pub type Network = meilisearch_types::enterprise_edition::network::Network;
+pub type Network = meilisearch_types::enterprise_edition::network::DbNetwork;
 pub type Webhooks = meilisearch_types::webhooks::WebhooksDumpView;
 
 // ===== Other types to clarify the code of the compat module

crates/dump/src/writer.rs

@@ -5,7 +5,7 @@ use std::path::PathBuf;
 use flate2::write::GzEncoder;
 use flate2::Compression;
 use meilisearch_types::batches::Batch;
-use meilisearch_types::enterprise_edition::network::Network;
+use meilisearch_types::enterprise_edition::network::DbNetwork;
 use meilisearch_types::features::{ChatCompletionSettings, RuntimeTogglableFeatures};
 use meilisearch_types::keys::Key;
 use meilisearch_types::settings::{Checked, Settings};
@@ -72,7 +72,7 @@ impl DumpWriter {
         )?)
     }
 
-    pub fn create_network(&self, network: Network) -> Result<()> {
+    pub fn create_network(&self, network: DbNetwork) -> Result<()> {
         Ok(std::fs::write(self.dir.path().join("network.json"), serde_json::to_string(&network)?)?)
     }
 

crates/filter-parser/src/condition.rs

@@ -7,23 +7,14 @@
 
 use nom::branch::alt;
 use nom::bytes::complete::tag;
-use nom::character::complete::char;
-use nom::character::complete::multispace0;
-use nom::character::complete::multispace1;
-use nom::combinator::cut;
-use nom::combinator::map;
-use nom::combinator::value;
-use nom::sequence::preceded;
-use nom::sequence::{terminated, tuple};
+use nom::character::complete::{char, multispace0, multispace1};
+use nom::combinator::{cut, map, value};
+use nom::sequence::{preceded, terminated, tuple};
 use Condition::*;
 
 use crate::error::IResultExt;
-use crate::value::parse_vector_value;
-use crate::value::parse_vector_value_cut;
-use crate::Error;
-use crate::ErrorKind;
-use crate::VectorFilter;
-use crate::{parse_value, FilterCondition, IResult, Span, Token};
+use crate::value::{parse_vector_value, parse_vector_value_cut};
+use crate::{parse_value, Error, ErrorKind, FilterCondition, IResult, Span, Token, VectorFilter};
 
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub enum Condition<'a> {

crates/filter-parser/src/error.rs

@@ -75,7 +75,11 @@ pub enum ExpectedValueKind {
 pub enum ErrorKind<'a> {
     ReservedGeo(&'a str),
     GeoRadius,
+    GeoRadiusArgumentCount(usize),
     GeoBoundingBox,
+    GeoPolygon,
+    GeoPolygonNotEnoughPoints(usize),
+    GeoCoordinatesNotPair(usize),
     MisusedGeoRadius,
     MisusedGeoBoundingBox,
     VectorFilterLeftover,
@@ -189,7 +193,7 @@ impl Display for Error<'_> {
             }
             ErrorKind::InvalidPrimary => {
                 let text = if input.trim().is_empty() { "but instead got nothing.".to_string() } else { format!("at `{}`.", escaped_input) };
-                writeln!(f, "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` {}", text)?
+                writeln!(f, "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` {text}")?
             }
             ErrorKind::InvalidEscapedNumber => {
                 writeln!(f, "Found an invalid escaped sequence number: `{}`.", escaped_input)?
@@ -198,11 +202,23 @@ impl Display for Error<'_> {
                 writeln!(f, "Found unexpected characters at the end of the filter: `{}`. You probably forgot an `OR` or an `AND` rule.", escaped_input)?
             }
             ErrorKind::GeoRadius => {
-                writeln!(f, "The `_geoRadius` filter expects three arguments: `_geoRadius(latitude, longitude, radius)`.")?
+                writeln!(f, "The `_geoRadius` filter must be in the form: `_geoRadius(latitude, longitude, radius, optionalResolution)`.")?
+            }
+            ErrorKind::GeoRadiusArgumentCount(count) => {
+                writeln!(f, "Was expecting 3 or 4 arguments for `_geoRadius`, but instead found {count}.")?
             }
             ErrorKind::GeoBoundingBox => {
                 writeln!(f, "The `_geoBoundingBox` filter expects two pairs of arguments: `_geoBoundingBox([latitude, longitude], [latitude, longitude])`.")?
             }
+            ErrorKind::GeoPolygon => {
+                writeln!(f, "The `_geoPolygon` filter doesn't match the expected format: `_geoPolygon([latitude, longitude], [latitude, longitude])`.")?
+            }
+            ErrorKind::GeoPolygonNotEnoughPoints(n) => {
+                writeln!(f, "The `_geoPolygon` filter expects at least 3 points but only {n} were specified")?;
+            }
+            ErrorKind::GeoCoordinatesNotPair(number) => {
+                writeln!(f, "Was expecting 2 coordinates but instead found {number}.")?
+            }
             ErrorKind::ReservedGeo(name) => {
                 writeln!(f, "`{}` is a reserved keyword and thus can't be used as a filter expression. Use the `_geoRadius(latitude, longitude, distance)` or `_geoBoundingBox([latitude, longitude], [latitude, longitude])` built-in rules to filter on `_geo` coordinates.", name.escape_debug())?
             }

crates/filter-parser/src/lib.rs

@@ -19,6 +19,7 @@
 //! word           = (alphanumeric | _ | - | .)+
 //! geoRadius      = "_geoRadius(" WS* float WS* "," WS* float WS* "," float WS* ")"
 //! geoBoundingBox = "_geoBoundingBox([" WS * float WS* "," WS* float WS* "], [" WS* float WS* "," WS* float WS* "]")
+//! geoPolygon     = "_geoPolygon([[" WS* float WS* "," WS* float WS* "],+])"
 //! ```
 //!
 //! Other BNF grammar used to handle some specific errors:
@@ -156,8 +157,9 @@ pub enum FilterCondition<'a> {
     Or(Vec<Self>),
     And(Vec<Self>),
     VectorExists { fid: Token<'a>, embedder: Option<Token<'a>>, filter: VectorFilter<'a> },
-    GeoLowerThan { point: [Token<'a>; 2], radius: Token<'a> },
+    GeoLowerThan { point: [Token<'a>; 2], radius: Token<'a>, resolution: Option<Token<'a>> },
     GeoBoundingBox { top_right_point: [Token<'a>; 2], bottom_left_point: [Token<'a>; 2] },
+    GeoPolygon { points: Vec<[Token<'a>; 2]> },
 }
 
 pub enum TraversedElement<'a> {
@@ -189,6 +191,7 @@ impl<'a> FilterCondition<'a> {
             FilterCondition::VectorExists { .. }
             | FilterCondition::GeoLowerThan { .. }
             | FilterCondition::GeoBoundingBox { .. }
+            | FilterCondition::GeoPolygon { .. }
             | FilterCondition::In { .. } => None,
         }
     }
@@ -202,6 +205,7 @@ impl<'a> FilterCondition<'a> {
             }
             FilterCondition::GeoLowerThan { .. }
             | FilterCondition::GeoBoundingBox { .. }
+            | FilterCondition::GeoPolygon { .. }
             | FilterCondition::In { .. } => None,
             FilterCondition::VectorExists { fid, .. } => Some(fid),
         }
@@ -396,23 +400,27 @@ fn parse_not(input: Span, depth: usize) -> IResult<FilterCondition> {
 /// If we parse `_geoRadius` we MUST parse the rest of the expression.
 fn parse_geo_radius(input: Span) -> IResult<FilterCondition> {
     // we want to allow space BEFORE the _geoRadius but not after
-    let parsed = preceded(
-        tuple((multispace0, word_exact("_geoRadius"))),
-        // if we were able to parse `_geoRadius` and can't parse the rest of the input we return a failure
-        cut(delimited(char('('), separated_list1(tag(","), ws(recognize_float)), char(')'))),
-    )(input)
-    .map_err(|e| e.map(|_| Error::new_from_kind(input, ErrorKind::GeoRadius)));
+
+    let (input, _) = tuple((multispace0, word_exact("_geoRadius")))(input)?;
+
+    // if we were able to parse `_geoRadius` and can't parse the rest of the input we return a failure
+
+    let parsed =
+        delimited(char('('), separated_list1(tag(","), ws(recognize_float)), char(')'))(input)
+            .map_cut(ErrorKind::GeoRadius);
 
     let (input, args) = parsed?;
 
-    if args.len() != 3 {
-        return Err(nom::Err::Failure(Error::new_from_kind(input, ErrorKind::GeoRadius)));
+    if !(3..=4).contains(&args.len()) {
+        return Err(Error::failure_from_kind(input, ErrorKind::GeoRadiusArgumentCount(args.len())));
     }
 
     let res = FilterCondition::GeoLowerThan {
         point: [args[0].into(), args[1].into()],
         radius: args[2].into(),
+        resolution: args.get(3).cloned().map(Token::from),
     };
+
     Ok((input, res))
 }
 
@@ -420,26 +428,33 @@ fn parse_geo_radius(input: Span) -> IResult<FilterCondition> {
 /// If we parse `_geoBoundingBox` we MUST parse the rest of the expression.
 fn parse_geo_bounding_box(input: Span) -> IResult<FilterCondition> {
     // we want to allow space BEFORE the _geoBoundingBox but not after
-    let parsed = preceded(
-        tuple((multispace0, word_exact("_geoBoundingBox"))),
-        // if we were able to parse `_geoBoundingBox` and can't parse the rest of the input we return a failure
-        cut(delimited(
-            char('('),
-            separated_list1(
-                tag(","),
-                ws(delimited(char('['), separated_list1(tag(","), ws(recognize_float)), char(']'))),
-            ),
-            char(')'),
-        )),
+
+    let (input, _) = tuple((multispace0, word_exact("_geoBoundingBox")))(input)?;
+
+    // if we were able to parse `_geoBoundingBox` and can't parse the rest of the input we return a failure
+
+    let (input, args) = delimited(
+        char('('),
+        separated_list1(
+            tag(","),
+            ws(delimited(char('['), separated_list1(tag(","), ws(recognize_float)), char(']'))),
+        ),
+        char(')'),
     )(input)
-    .map_err(|e| e.map(|_| Error::new_from_kind(input, ErrorKind::GeoBoundingBox)));
+    .map_cut(ErrorKind::GeoBoundingBox)?;
 
-    let (input, args) = parsed?;
-
-    if args.len() != 2 || args[0].len() != 2 || args[1].len() != 2 {
+    if args.len() != 2 {
         return Err(Error::failure_from_kind(input, ErrorKind::GeoBoundingBox));
     }
 
+    if let Some(offending) = args.iter().find(|a| a.len() != 2) {
+        let context = offending.first().unwrap_or(&input);
+        return Err(Error::failure_from_kind(
+            *context,
+            ErrorKind::GeoCoordinatesNotPair(offending.len()),
+        ));
+    }
+
     let res = FilterCondition::GeoBoundingBox {
         top_right_point: [args[0][0].into(), args[0][1].into()],
         bottom_left_point: [args[1][0].into(), args[1][1].into()],
@@ -447,6 +462,47 @@ fn parse_geo_bounding_box(input: Span) -> IResult<FilterCondition> {
     Ok((input, res))
 }
 
+/// geoPolygon     = "_geoPolygon([[" WS* float WS* "," WS* float WS* "],+])"
+/// If we parse `_geoPolygon` we MUST parse the rest of the expression.
+fn parse_geo_polygon(input: Span) -> IResult<FilterCondition> {
+    // we want to allow space BEFORE the _geoPolygon but not after
+
+    let (input, _) = tuple((multispace0, word_exact("_geoPolygon")))(input)?;
+
+    // if we were able to parse `_geoPolygon` and can't parse the rest of the input we return a failure
+
+    let (input, args): (_, Vec<Vec<LocatedSpan<_, _>>>) = delimited(
+        char('('),
+        separated_list1(
+            tag(","),
+            ws(delimited(char('['), separated_list1(tag(","), ws(recognize_float)), char(']'))),
+        ),
+        preceded(opt(ws(char(','))), char(')')), // Tolerate trailing comma
+    )(input)
+    .map_cut(ErrorKind::GeoPolygon)?;
+
+    if args.len() < 3 {
+        let context = args.last().and_then(|a| a.last()).unwrap_or(&input);
+        return Err(Error::failure_from_kind(
+            *context,
+            ErrorKind::GeoPolygonNotEnoughPoints(args.len()),
+        ));
+    }
+
+    if let Some(offending) = args.iter().find(|a| a.len() != 2) {
+        let context = offending.first().unwrap_or(&input);
+        return Err(Error::failure_from_kind(
+            *context,
+            ErrorKind::GeoCoordinatesNotPair(offending.len()),
+        ));
+    }
+
+    let res = FilterCondition::GeoPolygon {
+        points: args.into_iter().map(|a| [a[0].into(), a[1].into()]).collect(),
+    };
+    Ok((input, res))
+}
+
 /// geoPoint      = WS* "_geoPoint(float WS* "," WS* float WS* "," WS* float)
 fn parse_geo_point(input: Span) -> IResult<FilterCondition> {
     // we want to forbid space BEFORE the _geoPoint but not after
@@ -516,8 +572,8 @@ fn parse_primary(input: Span, depth: usize) -> IResult<FilterCondition> {
                 Error::new_from_kind(input, ErrorKind::MissingClosingDelimiter(c.char()))
             }),
         ),
-        parse_geo_radius,
-        parse_geo_bounding_box,
+        // Made a random block of functions because we reached the maximum number of elements per alt
+        alt((parse_geo_radius, parse_geo_bounding_box, parse_geo_polygon)),
         parse_in,
         parse_not_in,
         parse_condition,
@@ -597,9 +653,12 @@ impl std::fmt::Display for FilterCondition<'_> {
                 }
                 write!(f, " EXISTS")
             }
-            FilterCondition::GeoLowerThan { point, radius } => {
+            FilterCondition::GeoLowerThan { point, radius, resolution: None } => {
                 write!(f, "_geoRadius({}, {}, {})", point[0], point[1], radius)
             }
+            FilterCondition::GeoLowerThan { point, radius, resolution: Some(resolution) } => {
+                write!(f, "_geoRadius({}, {}, {}, {})", point[0], point[1], radius, resolution)
+            }
             FilterCondition::GeoBoundingBox {
                 top_right_point: top_left_point,
                 bottom_left_point: bottom_right_point,
@@ -613,6 +672,13 @@ impl std::fmt::Display for FilterCondition<'_> {
                     bottom_right_point[1]
                 )
             }
+            FilterCondition::GeoPolygon { points } => {
+                write!(f, "_geoPolygon([")?;
+                for point in points {
+                    write!(f, "[{}, {}], ", point[0], point[1])?;
+                }
+                write!(f, "])")
+            }
         }
     }
 }
@@ -776,12 +842,17 @@ pub mod tests {
         insta::assert_snapshot!(p("_geoRadius(12, 13, 14)"), @"_geoRadius({12}, {13}, {14})");
         insta::assert_snapshot!(p("NOT _geoRadius(12, 13, 14)"), @"NOT (_geoRadius({12}, {13}, {14}))");
         insta::assert_snapshot!(p("_geoRadius(12,13,14)"), @"_geoRadius({12}, {13}, {14})");
+        insta::assert_snapshot!(p("_geoRadius(12,13,14,1000)"), @"_geoRadius({12}, {13}, {14}, {1000})");
 
         // Test geo bounding box
         insta::assert_snapshot!(p("_geoBoundingBox([12, 13], [14, 15])"), @"_geoBoundingBox([{12}, {13}], [{14}, {15}])");
         insta::assert_snapshot!(p("NOT _geoBoundingBox([12, 13], [14, 15])"), @"NOT (_geoBoundingBox([{12}, {13}], [{14}, {15}]))");
         insta::assert_snapshot!(p("_geoBoundingBox([12,13],[14,15])"), @"_geoBoundingBox([{12}, {13}], [{14}, {15}])");
 
+        // Test geo polygon
+        insta::assert_snapshot!(p("_geoPolygon([12, 13], [14, 15], [16, 17])"), @"_geoPolygon([[{12}, {13}], [{14}, {15}], [{16}, {17}], ])");
+        insta::assert_snapshot!(p("_geoPolygon([12, 13], [14, 15], [-1.2,2939.2], [1,1])"), @"_geoPolygon([[{12}, {13}], [{14}, {15}], [{-1.2}, {2939.2}], [{1}, {1}], ])");
+
         // Test OR + AND
         insta::assert_snapshot!(p("channel = ponce AND 'dog race' != 'bernese mountain'"), @"AND[{channel} = {ponce}, {dog race} != {bernese mountain}, ]");
         insta::assert_snapshot!(p("channel = ponce OR 'dog race' != 'bernese mountain'"), @"OR[{channel} = {ponce}, {dog race} != {bernese mountain}, ]");
@@ -838,50 +909,80 @@ pub mod tests {
         11:12 channel = 🐻 AND followers < 100
         "###);
 
-        insta::assert_snapshot!(p("'OR'"), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `\'OR\'`.
+        insta::assert_snapshot!(p("'OR'"), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `\'OR\'`.
         1:5 'OR'
-        "###);
+        ");
 
         insta::assert_snapshot!(p("OR"), @r###"
         Was expecting a value but instead got `OR`, which is a reserved keyword. To use `OR` as a field name or a value, surround it by quotes.
         1:3 OR
         "###);
 
-        insta::assert_snapshot!(p("channel Ponce"), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `channel Ponce`.
+        insta::assert_snapshot!(p("channel Ponce"), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `channel Ponce`.
         1:14 channel Ponce
-        "###);
+        ");
 
-        insta::assert_snapshot!(p("channel = Ponce OR"), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` but instead got nothing.
+        insta::assert_snapshot!(p("channel = Ponce OR"), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` but instead got nothing.
         19:19 channel = Ponce OR
-        "###);
+        ");
 
-        insta::assert_snapshot!(p("_geoRadius"), @r###"
-        The `_geoRadius` filter expects three arguments: `_geoRadius(latitude, longitude, radius)`.
-        1:11 _geoRadius
-        "###);
+        insta::assert_snapshot!(p("_geoRadius"), @r"
+        The `_geoRadius` filter must be in the form: `_geoRadius(latitude, longitude, radius, optionalResolution)`.
+        11:11 _geoRadius
+        ");
 
-        insta::assert_snapshot!(p("_geoRadius = 12"), @r###"
-        The `_geoRadius` filter expects three arguments: `_geoRadius(latitude, longitude, radius)`.
-        1:16 _geoRadius = 12
-        "###);
+        insta::assert_snapshot!(p("_geoRadius = 12"), @r"
+        The `_geoRadius` filter must be in the form: `_geoRadius(latitude, longitude, radius, optionalResolution)`.
+        11:16 _geoRadius = 12
+        ");
 
-        insta::assert_snapshot!(p("_geoBoundingBox"), @r###"
+        insta::assert_snapshot!(p("_geoBoundingBox"), @r"
         The `_geoBoundingBox` filter expects two pairs of arguments: `_geoBoundingBox([latitude, longitude], [latitude, longitude])`.
-        1:16 _geoBoundingBox
-        "###);
+        16:16 _geoBoundingBox
+        ");
 
-        insta::assert_snapshot!(p("_geoBoundingBox = 12"), @r###"
+        insta::assert_snapshot!(p("_geoBoundingBox = 12"), @r"
         The `_geoBoundingBox` filter expects two pairs of arguments: `_geoBoundingBox([latitude, longitude], [latitude, longitude])`.
-        1:21 _geoBoundingBox = 12
-        "###);
+        16:21 _geoBoundingBox = 12
+        ");
 
-        insta::assert_snapshot!(p("_geoBoundingBox(1.0, 1.0)"), @r###"
+        insta::assert_snapshot!(p("_geoBoundingBox(1.0, 1.0)"), @r"
         The `_geoBoundingBox` filter expects two pairs of arguments: `_geoBoundingBox([latitude, longitude], [latitude, longitude])`.
-        1:26 _geoBoundingBox(1.0, 1.0)
-        "###);
+        17:26 _geoBoundingBox(1.0, 1.0)
+        ");
+
+        insta::assert_snapshot!(p("_geoPolygon([1,2,3])"), @r"
+        The `_geoPolygon` filter expects at least 3 points but only 1 were specified
+        18:19 _geoPolygon([1,2,3])
+        ");
+
+        insta::assert_snapshot!(p("_geoPolygon(1,2,3)"), @r"
+        The `_geoPolygon` filter doesn't match the expected format: `_geoPolygon([latitude, longitude], [latitude, longitude])`.
+        13:19 _geoPolygon(1,2,3)
+        ");
+
+        insta::assert_snapshot!(p("_geoPolygon([1,2],[1,2],[1,2,3])"), @r"
+        Was expecting 2 coordinates but instead found 3.
+        26:27 _geoPolygon([1,2],[1,2],[1,2,3])
+        ");
+
+        insta::assert_snapshot!(p("_geoPolygon([1,2],[1,2,3])"), @r"
+        The `_geoPolygon` filter expects at least 3 points but only 2 were specified
+        24:25 _geoPolygon([1,2],[1,2,3])
+        ");
+
+        insta::assert_snapshot!(p("_geoPolygon(1)"), @r"
+        The `_geoPolygon` filter doesn't match the expected format: `_geoPolygon([latitude, longitude], [latitude, longitude])`.
+        13:15 _geoPolygon(1)
+        ");
+
+        insta::assert_snapshot!(p("_geoPolygon([1,2)"), @r"
+        The `_geoPolygon` filter doesn't match the expected format: `_geoPolygon([latitude, longitude], [latitude, longitude])`.
+        17:18 _geoPolygon([1,2)
+        ");
 
         insta::assert_snapshot!(p("_geoPoint(12, 13, 14)"), @r###"
         `_geoPoint` is a reserved keyword and thus can't be used as a filter expression. Use the `_geoRadius(latitude, longitude, distance)` or `_geoBoundingBox([latitude, longitude], [latitude, longitude])` built-in rules to filter on `_geo` coordinates.
@@ -938,15 +1039,15 @@ pub mod tests {
         34:35 channel = mv OR followers >= 1000)
         "###);
 
-        insta::assert_snapshot!(p("colour NOT EXIST"), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `colour NOT EXIST`.
+        insta::assert_snapshot!(p("colour NOT EXIST"), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `colour NOT EXIST`.
         1:17 colour NOT EXIST
-        "###);
+        ");
 
-        insta::assert_snapshot!(p("subscribers 100 TO1000"), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `subscribers 100 TO1000`.
+        insta::assert_snapshot!(p("subscribers 100 TO1000"), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `subscribers 100 TO1000`.
         1:23 subscribers 100 TO1000
-        "###);
+        ");
 
         insta::assert_snapshot!(p("channel = ponce ORdog != 'bernese mountain'"), @r###"
         Found unexpected characters at the end of the filter: `ORdog != \'bernese mountain\'`. You probably forgot an `OR` or an `AND` rule.
@@ -1071,38 +1172,38 @@ pub mod tests {
         5:7 NOT OR EXISTS AND EXISTS NOT EXISTS
         "###);
 
-        insta::assert_snapshot!(p(r#"value NULL"#), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `value NULL`.
+        insta::assert_snapshot!(p(r#"value NULL"#), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `value NULL`.
         1:11 value NULL
-        "###);
-        insta::assert_snapshot!(p(r#"value NOT NULL"#), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `value NOT NULL`.
+        ");
+        insta::assert_snapshot!(p(r#"value NOT NULL"#), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `value NOT NULL`.
         1:15 value NOT NULL
-        "###);
-        insta::assert_snapshot!(p(r#"value EMPTY"#), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `value EMPTY`.
+        ");
+        insta::assert_snapshot!(p(r#"value EMPTY"#), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `value EMPTY`.
         1:12 value EMPTY
-        "###);
-        insta::assert_snapshot!(p(r#"value NOT EMPTY"#), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `value NOT EMPTY`.
+        ");
+        insta::assert_snapshot!(p(r#"value NOT EMPTY"#), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `value NOT EMPTY`.
         1:16 value NOT EMPTY
-        "###);
-        insta::assert_snapshot!(p(r#"value IS"#), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `value IS`.
+        ");
+        insta::assert_snapshot!(p(r#"value IS"#), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `value IS`.
         1:9 value IS
-        "###);
-        insta::assert_snapshot!(p(r#"value IS NOT"#), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `value IS NOT`.
+        ");
+        insta::assert_snapshot!(p(r#"value IS NOT"#), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `value IS NOT`.
         1:13 value IS NOT
-        "###);
-        insta::assert_snapshot!(p(r#"value IS EXISTS"#), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `value IS EXISTS`.
+        ");
+        insta::assert_snapshot!(p(r#"value IS EXISTS"#), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `value IS EXISTS`.
         1:16 value IS EXISTS
-        "###);
-        insta::assert_snapshot!(p(r#"value IS NOT EXISTS"#), @r###"
-        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `value IS NOT EXISTS`.
+        ");
+        insta::assert_snapshot!(p(r#"value IS NOT EXISTS"#), @r"
+        Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `value IS NOT EXISTS`.
         1:20 value IS NOT EXISTS
-        "###);
+        ");
     }
 
     #[test]

crates/index-scheduler/Cargo.toml

@@ -23,6 +23,7 @@ dump = { path = "../dump" }
 enum-iterator = "2.1.0"
 file-store = { path = "../file-store" }
 flate2 = "1.1.2"
+hashbrown = "0.15.4"
 indexmap = "2.9.0"
 meilisearch-auth = { path = "../meilisearch-auth" }
 meilisearch-types = { path = "../meilisearch-types" }
@@ -45,6 +46,8 @@ tracing = "0.1.41"
 ureq = "2.12.1"
 uuid = { version = "1.17.0", features = ["serde", "v4"] }
 backoff = "0.4.0"
+itertools = "0.14.0"
+tokio = { version = "1.47.1", features = ["full"] }
 
 [dev-dependencies]
 big_s = "1.0.2"

crates/index-scheduler/src/dump.rs

@@ -234,6 +234,9 @@ impl<'a> Dump<'a> {
                     }
                 }
                 KindDump::UpgradeDatabase { from } => KindWithContent::UpgradeDatabase { from },
+                KindDump::NetworkTopologyChange { network: new_network, origin } => {
+                    KindWithContent::NetworkTopologyChange { network: new_network, origin }
+                }
             },
         };
 

crates/index-scheduler/src/error.rs

@@ -45,6 +45,7 @@ impl From<DateField> for Code {
     }
 }
 
+#[allow(clippy::large_enum_variant)]
 #[derive(Error, Debug)]
 pub enum Error {
     #[error("{1}")]

crates/index-scheduler/src/features.rs

@@ -1,6 +1,6 @@
 use std::sync::{Arc, RwLock};
 
-use meilisearch_types::enterprise_edition::network::Network;
+use meilisearch_types::enterprise_edition::network::DbNetwork;
 use meilisearch_types::features::{InstanceTogglableFeatures, RuntimeTogglableFeatures};
 use meilisearch_types::heed::types::{SerdeJson, Str};
 use meilisearch_types::heed::{Database, Env, RwTxn, WithoutTls};
@@ -24,7 +24,7 @@ mod db_keys {
 pub(crate) struct FeatureData {
     persisted: Database<Str, SerdeJson<RuntimeTogglableFeatures>>,
     runtime: Arc<RwLock<RuntimeTogglableFeatures>>,
-    network: Arc<RwLock<Network>>,
+    network: Arc<RwLock<DbNetwork>>,
 }
 
 #[derive(Debug, Clone, Copy)]
@@ -197,8 +197,8 @@ impl FeatureData {
         }));
 
         // Once this is stabilized, network should be stored along with webhooks in index-scheduler's persisted database
-        let network_db = runtime_features_db.remap_data_type::<SerdeJson<Network>>();
-        let network: Network = network_db.get(wtxn, db_keys::NETWORK)?.unwrap_or_default();
+        let network_db = runtime_features_db.remap_data_type::<SerdeJson<DbNetwork>>();
+        let network: DbNetwork = network_db.get(wtxn, db_keys::NETWORK)?.unwrap_or_default();
 
         Ok(Self {
             persisted: runtime_features_db,
@@ -234,8 +234,8 @@ impl FeatureData {
         RoFeatures::new(self)
     }
 
-    pub fn put_network(&self, mut wtxn: RwTxn, new_network: Network) -> Result<()> {
-        self.persisted.remap_data_type::<SerdeJson<Network>>().put(
+    pub fn put_network(&self, mut wtxn: RwTxn, new_network: DbNetwork) -> Result<()> {
+        self.persisted.remap_data_type::<SerdeJson<DbNetwork>>().put(
             &mut wtxn,
             db_keys::NETWORK,
             &new_network,
@@ -247,7 +247,7 @@ impl FeatureData {
         Ok(())
     }
 
-    pub fn network(&self) -> Network {
-        Network::clone(&*self.network.read().unwrap())
+    pub fn network(&self) -> DbNetwork {
+        DbNetwork::clone(&*self.network.read().unwrap())
     }
 }

crates/index-scheduler/src/insta_snapshot.rs

@@ -36,6 +36,7 @@ pub fn snapshot_index_scheduler(scheduler: &IndexScheduler) -> String {
         run_loop_iteration: _,
         embedders: _,
         chat_settings: _,
+        runtime: _,
     } = scheduler;
 
     let rtxn = env.read_txn().unwrap();
@@ -317,6 +318,9 @@ fn snapshot_details(d: &Details) -> String {
         Details::UpgradeDatabase { from, to } => {
             format!("{{ from: {from:?}, to: {to:?} }}")
         }
+        Details::NetworkTopologyChange { network: new_network } => {
+            format!("{{ new_network: {new_network:?} }}")
+        }
     }
 }
 

crates/index-scheduler/src/lib.rs

@@ -54,7 +54,7 @@ pub use features::RoFeatures;
 use flate2::bufread::GzEncoder;
 use flate2::Compression;
 use meilisearch_types::batches::Batch;
-use meilisearch_types::enterprise_edition::network::Network;
+use meilisearch_types::enterprise_edition::network::DbNetwork;
 use meilisearch_types::features::{
     ChatCompletionSettings, InstanceTogglableFeatures, RuntimeTogglableFeatures,
 };
@@ -216,6 +216,8 @@ pub struct IndexScheduler {
     /// A counter that is incremented before every call to [`tick`](IndexScheduler::tick)
     #[cfg(test)]
     run_loop_iteration: Arc<RwLock<usize>>,
+
+    runtime: Option<tokio::runtime::Handle>,
 }
 
 impl IndexScheduler {
@@ -242,6 +244,7 @@ impl IndexScheduler {
             run_loop_iteration: self.run_loop_iteration.clone(),
             features: self.features.clone(),
             chat_settings: self.chat_settings,
+            runtime: self.runtime.clone(),
         }
     }
 
@@ -260,6 +263,7 @@ impl IndexScheduler {
         options: IndexSchedulerOptions,
         auth_env: Env<WithoutTls>,
         from_db_version: (u32, u32, u32),
+        runtime: Option<tokio::runtime::Handle>,
         #[cfg(test)] test_breakpoint_sdr: crossbeam_channel::Sender<(test_utils::Breakpoint, bool)>,
         #[cfg(test)] planned_failures: Vec<(usize, test_utils::FailureLocation)>,
     ) -> Result<Self> {
@@ -341,6 +345,7 @@ impl IndexScheduler {
             run_loop_iteration: Arc::new(RwLock::new(0)),
             features,
             chat_settings,
+            runtime,
         };
 
         this.run();
@@ -892,13 +897,13 @@ impl IndexScheduler {
         Ok(())
     }
 
-    pub fn put_network(&self, network: Network) -> Result<()> {
+    pub fn put_network(&self, network: DbNetwork) -> Result<()> {
         let wtxn = self.env.write_txn().map_err(Error::HeedTransaction)?;
         self.features.put_network(wtxn, network)?;
         Ok(())
     }
 
-    pub fn network(&self) -> Network {
+    pub fn network(&self) -> DbNetwork {
         self.features.network()
     }
 
@@ -927,9 +932,10 @@ impl IndexScheduler {
 
     pub fn embedders(
         &self,
-        index_uid: String,
+        index_uid: &str,
         embedding_configs: Vec<IndexEmbeddingConfig>,
     ) -> Result<RuntimeEmbedders> {
+        let err = |err| Error::from_milli(err, Some(index_uid.to_owned()));
         let res: Result<_> = embedding_configs
             .into_iter()
             .map(
@@ -942,7 +948,7 @@ impl IndexScheduler {
                     let document_template = prompt
                         .try_into()
                         .map_err(meilisearch_types::milli::Error::from)
-                        .map_err(|err| Error::from_milli(err, Some(index_uid.clone())))?;
+                        .map_err(err)?;
 
                     let fragments = fragments
                         .into_inner()
@@ -972,9 +978,8 @@ impl IndexScheduler {
                     let embedder = Arc::new(
                         Embedder::new(embedder_options.clone(), self.scheduler.embedding_cache_cap)
                             .map_err(meilisearch_types::milli::vector::Error::from)
-                            .map_err(|err| {
-                                Error::from_milli(err.into(), Some(index_uid.clone()))
-                            })?,
+                            .map_err(milli::Error::from)
+                            .map_err(err)?,
                     );
                     {
                         let mut embedders = self.embedders.write().unwrap();

crates/index-scheduler/src/scheduler/autobatcher.rs

@@ -73,6 +73,7 @@ impl From<KindWithContent> for AutobatchKind {
             | KindWithContent::DumpCreation { .. }
             | KindWithContent::Export { .. }
             | KindWithContent::UpgradeDatabase { .. }
+            | KindWithContent::NetworkTopologyChange { .. }
             | KindWithContent::SnapshotCreation => {
                 panic!("The autobatcher should never be called with tasks that don't apply to an index.")
             }
@@ -287,7 +288,7 @@ impl BatchKind {
         };
 
         match (self, autobatch_kind) {
-            // We don't batch any of these operations  
+            // We don't batch any of these operations
             (this, K::IndexCreation | K::IndexUpdate | K::IndexSwap | K::DocumentEdition) => Break((this, BatchStopReason::TaskCannotBeBatched { kind, id })),
             // We must not batch tasks that don't have the same index creation rights if the index doesn't already exists.
             (this, kind) if !index_already_exists && this.allow_index_creation() == Some(false) && kind.allow_index_creation() == Some(true) => {

crates/index-scheduler/src/scheduler/create_batch.rs

@@ -55,6 +55,9 @@ pub(crate) enum Batch {
     UpgradeDatabase {
         tasks: Vec<Task>,
     },
+    NetworkTopologyChanges {
+        tasks: Vec<Task>,
+    },
 }
 
 #[derive(Debug)]
@@ -116,7 +119,8 @@ impl Batch {
             Batch::SnapshotCreation(tasks)
             | Batch::TaskDeletions(tasks)
             | Batch::UpgradeDatabase { tasks }
-            | Batch::IndexDeletion { tasks, .. } => {
+            | Batch::IndexDeletion { tasks, .. }
+            | Batch::NetworkTopologyChanges { tasks } => {
                 RoaringBitmap::from_iter(tasks.iter().map(|task| task.uid))
             }
             Batch::IndexOperation { op, .. } => match op {
@@ -151,6 +155,7 @@ impl Batch {
             | Dump(_)
             | Export { .. }
             | UpgradeDatabase { .. }
+            | NetworkTopologyChanges { .. }
             | IndexSwap { .. } => None,
             IndexOperation { op, .. } => Some(op.index_uid()),
             IndexCreation { index_uid, .. }
@@ -176,6 +181,7 @@ impl fmt::Display for Batch {
             Batch::IndexDeletion { .. } => f.write_str("IndexDeletion")?,
             Batch::IndexSwap { .. } => f.write_str("IndexSwap")?,
             Batch::Export { .. } => f.write_str("Export")?,
+            Batch::NetworkTopologyChanges { .. } => f.write_str("NetworkTopologyChange")?,
             Batch::UpgradeDatabase { .. } => f.write_str("UpgradeDatabase")?,
         };
         match index_uid {
@@ -545,7 +551,18 @@ impl IndexScheduler {
             return Ok(Some((Batch::Dump(task), current_batch)));
         }
 
-        // 6. We make a batch from the unprioritised tasks. Start by taking the next enqueued task.
+        // 6. We batch the network changes.
+        let to_network = self.queue.tasks.get_kind(rtxn, Kind::NetworkTopologyChange)? & enqueued;
+        if !to_network.is_empty() {
+            let mut tasks = self.queue.tasks.get_existing_tasks(rtxn, to_network)?;
+            current_batch.processing(&mut tasks);
+            current_batch.reason(BatchStopReason::TaskKindCannotBeBatched {
+                kind: Kind::NetworkTopologyChange,
+            });
+            return Ok(Some((Batch::NetworkTopologyChanges { tasks }, current_batch)));
+        }
+
+        // 7. We make a batch from the unprioritised tasks. Start by taking the next enqueued task.
         let task_id = if let Some(task_id) = enqueued.min() { task_id } else { return Ok(None) };
         let mut task =
             self.queue.tasks.get_task(rtxn, task_id)?.ok_or(Error::CorruptedTaskQueue)?;

crates/index-scheduler/src/scheduler/enterprise_edition/mod.rs

@@ -0,0 +1,6 @@
+// Copyright © 2025 Meilisearch Some Rights Reserved
+// This file is part of Meilisearch Enterprise Edition (EE).
+// Use of this source code is governed by the Business Source License 1.1,
+// as found in the LICENSE-EE file or at <https://mariadb.com/bsl11>
+
+mod process_network;

crates/index-scheduler/src/scheduler/enterprise_edition/process_network.rs

@@ -0,0 +1,362 @@
+// Copyright © 2025 Meilisearch Some Rights Reserved
+// This file is part of Meilisearch Enterprise Edition (EE).
+// Use of this source code is governed by the Business Source License 1.1,
+// as found in the LICENSE-EE file or at <https://mariadb.com/bsl11>
+
+use std::collections::BTreeMap;
+use std::time::Duration;
+
+use bumpalo::Bump;
+use itertools::{EitherOrBoth, Itertools};
+use meilisearch_types::enterprise_edition::network::{DbNetwork, DbRemote, Network, Remote};
+use meilisearch_types::milli::documents::PrimaryKey;
+use meilisearch_types::milli::progress::{EmbedderStats, Progress};
+use meilisearch_types::milli::update::new::indexer;
+use meilisearch_types::milli::update::Setting;
+use meilisearch_types::milli::{self};
+use meilisearch_types::tasks::{KindWithContent, Status, Task};
+use roaring::RoaringBitmap;
+
+use crate::scheduler::process_export::{ExportContext, ExportOptions, TargetInstance};
+use crate::{Error, IndexScheduler};
+
+impl IndexScheduler {
+    pub(crate) fn process_network_changes(
+        &self,
+        progress: Progress,
+        mut tasks: Vec<Task>,
+    ) -> crate::Result<Vec<Task>> {
+        let old_network = self.network();
+        let mut current_network = Some(old_network.clone());
+        for task in &tasks {
+            let KindWithContent::NetworkTopologyChange { network, origin } = &task.kind else {
+                continue;
+            };
+            current_network = match (current_network, network) {
+                (None, None) => None,
+                (None, Some(network)) => Some(accumulate(DbNetwork::default(), network.clone())?),
+                (Some(current_network), None) => Some(current_network),
+                (Some(current_network), Some(new_network)) => {
+                    Some(accumulate(current_network, new_network.clone())?)
+                }
+            };
+        }
+
+        'network: {
+            let mut new_network = current_network.unwrap_or_default();
+            if old_network == new_network {
+                // no change, exit
+                break 'network;
+            }
+
+            /// TODO: only do this if the task originates with an end-user
+            let must_replicate = old_network.sharding || new_network.sharding;
+
+            if !must_replicate {
+                self.put_network(new_network)?;
+                break 'network;
+            }
+
+            let must_stop_processing = &self.scheduler.must_stop_processing;
+
+            /// FIXME: make it mandatory for `self` to be part of the network
+            let old_this = old_network.local.as_deref();
+            /// FIXME: error here
+            let new_this = new_network.local.unwrap();
+
+            // in network replication, we need to tell old nodes that they are no longer part of the network.
+            // This is made difficult by "node aliasing": Meilisearch has no way of knowing if two nodes with different names
+            // or even different URLs actually refer to the same machine in two different versions of the network.
+            //
+            // This implementation ignores aliasing: a node is the same when it has the same name.
+            //
+            // To defeat aliasing, we iterate a first time to collect all deletions and additions, then we make sure to process the deletions
+            // first, rather than processing the tasks in the alphalexical order of remotes.
+            let mut node_deletions = Vec::new();
+            let mut node_additions = Vec::new();
+            for eob in old_network
+                .remotes
+                .iter()
+                .merge_join_by(new_network.remotes.iter(), |(left, _), (right, _)| left.cmp(right))
+            {
+                match eob {
+                    EitherOrBoth::Both((to_update_name, _), (_, new_node)) => {
+                        if to_update_name.as_str() == new_this {
+                            continue; // skip `self`
+                        }
+                        node_additions.push((to_update_name, new_node));
+                    }
+                    EitherOrBoth::Left((to_delete_name, to_delete_node)) => {
+                        if Some(to_delete_name.as_str()) == old_this {
+                            continue; // skip `self`
+                        }
+                        node_deletions.push((to_delete_name, to_delete_node));
+                    }
+                    EitherOrBoth::Right((to_add_name, to_add_node)) => {
+                        if to_add_name.as_str() == new_this {
+                            continue; // skip `self`
+                        }
+                        node_additions.push((to_add_name, to_add_node));
+                    }
+                }
+            }
+
+            let runtime = self.runtime.clone().unwrap();
+            let mut in_flight = Vec::new();
+            // process deletions
+            for (to_delete_name, to_delete) in node_deletions {
+                // set `self` to None so that this node is forgotten about
+                new_network.local = None;
+                in_flight.push(proxy_network(&runtime, to_delete.url.as_str(), &new_network)?);
+            }
+
+            runtime.block_on(async {
+                for task in in_flight.drain(..) {
+                    // TODO: log and ignore errors during deletion
+                    let res = task.await;
+                }
+            });
+
+            // process additions
+            for (to_add_name, to_add) in node_additions {
+                new_network.local = Some(to_add_name.clone());
+                in_flight.push(proxy_network(&runtime, to_add.url.as_str(), &new_network)?);
+            }
+
+            runtime.block_on(async {
+                for task in in_flight.drain(..) {
+                    // TODO: handle errors during addition
+                    let res = task.await;
+                }
+            });
+
+            // balance documents
+            new_network.local = Some(new_this);
+
+            self.balance_documents(&new_network, &progress, &must_stop_processing)?;
+
+            self.put_network(new_network)?;
+        }
+
+        for task in &mut tasks {
+            task.status = Status::Succeeded;
+        }
+        Ok(tasks)
+    }
+
+    fn balance_documents(
+        &self,
+        new_network: &DbNetwork,
+        progress: &Progress,
+        must_stop_processing: &crate::scheduler::MustStopProcessing,
+    ) -> crate::Result<()> {
+        /// FIXME unwrap
+        let new_shards = new_network.shards().unwrap();
+
+        // TECHDEBT: this spawns a `ureq` agent additionally to `reqwest`. We probably want to harmonize all of this.
+        let agent = ureq::AgentBuilder::new().timeout(Duration::from_secs(5)).build();
+
+        let mut indexer_alloc = Bump::new();
+
+        // process by batches of 20MiB. Allow for compression? Don't forget about embeddings
+        let _: Vec<()> = self.try_for_each_index(|index_uid, index| -> crate::Result<()> {
+            indexer_alloc.reset();
+            let err = |err| Error::from_milli(err, Some(index_uid.to_string()));
+            let index_rtxn = index.read_txn()?;
+            let all_docids = index.external_documents_ids();
+            let mut documents_to_move_to: hashbrown::HashMap<String, RoaringBitmap> =
+                hashbrown::HashMap::new();
+            let mut documents_to_delete = RoaringBitmap::new();
+
+            for res in all_docids.iter(&index_rtxn)? {
+                let (external_docid, docid) = res?;
+                match new_shards.processing_shard(external_docid) {
+                    Some(shard) if shard.is_own => continue,
+                    Some(shard) => {
+                        documents_to_move_to
+                            .entry_ref(shard.name.as_str())
+                            .or_default()
+                            .insert(docid);
+                    }
+                    None => {
+                        documents_to_delete.insert(docid);
+                    }
+                }
+            }
+
+            let fields_ids_map = index.fields_ids_map(&index_rtxn)?;
+
+            for (remote, documents_to_move) in documents_to_move_to {
+                /// TODO: justify the unwrap
+                let remote = new_network.remotes.get(&remote).unwrap();
+
+                let target = TargetInstance {
+                    base_url: &remote.url,
+                    api_key: remote.write_api_key.as_deref(),
+                };
+                let options = ExportOptions {
+                    index_uid,
+                    payload_size: None,
+                    override_settings: false,
+                    extra_headers: &Default::default(),
+                };
+                let ctx = ExportContext {
+                    index,
+                    index_rtxn: &index_rtxn,
+                    universe: &documents_to_move,
+                    progress,
+                    agent: &agent,
+                    must_stop_processing,
+                };
+
+                self.export_one_index(target, options, ctx)?;
+
+                documents_to_delete |= documents_to_move;
+            }
+
+            if documents_to_delete.is_empty() {
+                return Ok(());
+            }
+
+            let mut new_fields_ids_map = fields_ids_map.clone();
+
+            // candidates not empty => index not empty => a primary key is set
+            let primary_key = index.primary_key(&index_rtxn)?.unwrap();
+
+            let primary_key = PrimaryKey::new_or_insert(primary_key, &mut new_fields_ids_map)
+                .map_err(milli::Error::from)
+                .map_err(err)?;
+
+            let mut index_wtxn = index.write_txn()?;
+
+            let mut indexer = indexer::DocumentDeletion::new();
+            indexer.delete_documents_by_docids(documents_to_delete);
+            let document_changes = indexer.into_changes(&indexer_alloc, primary_key);
+            let embedders = index
+                .embedding_configs()
+                .embedding_configs(&index_wtxn)
+                .map_err(milli::Error::from)
+                .map_err(err)?;
+            let embedders = self.embedders(index_uid, embedders)?;
+            let indexer_config = self.index_mapper.indexer_config();
+            let pool = &indexer_config.thread_pool;
+
+            indexer::index(
+                &mut index_wtxn,
+                index,
+                pool,
+                indexer_config.grenad_parameters(),
+                &fields_ids_map,
+                new_fields_ids_map,
+                None, // document deletion never changes primary key
+                &document_changes,
+                embedders,
+                &|| must_stop_processing.get(),
+                &progress,
+                &EmbedderStats::default(),
+            )
+            .map_err(err)?;
+
+            index_wtxn.commit()?;
+
+            Ok(())
+        })?;
+        Ok(())
+    }
+}
+
+fn proxy_network(
+    runtime: &tokio::runtime::Handle,
+    url: &str,
+    network: &DbNetwork,
+) -> crate::Result<tokio::task::JoinHandle<()>> {
+    todo!()
+}
+
+fn accumulate(old_network: DbNetwork, new_network: Network) -> crate::Result<DbNetwork> {
+    let err = |err| Err(Error::from_milli(milli::Error::UserError(err), None));
+
+    let merged_local = match new_network.local {
+        Setting::Set(new_self) => Some(new_self),
+        Setting::Reset => None,
+        Setting::NotSet => old_network.local,
+    };
+
+    let merged_sharding = match new_network.sharding {
+        Setting::Set(new_sharding) => new_sharding,
+        Setting::Reset => false,
+        Setting::NotSet => old_network.sharding,
+    };
+
+    if merged_sharding && merged_local.is_none() {
+        return err(milli::UserError::NetworkShardingWithoutSelf);
+    }
+
+    let merged_remotes = match new_network.remotes {
+        Setting::Set(new_remotes) => {
+            let mut merged_remotes = BTreeMap::new();
+            for either_or_both in old_network
+                .remotes
+                .into_iter()
+                .merge_join_by(new_remotes.into_iter(), |left, right| left.0.cmp(&right.0))
+            {
+                match either_or_both {
+                    EitherOrBoth::Both((name, old), (_, Some(new))) => {
+                        let DbRemote {
+                            url: old_url,
+                            search_api_key: old_search_api_key,
+                            write_api_key: old_write_api_key,
+                        } = old;
+
+                        let Remote {
+                            url: new_url,
+                            search_api_key: new_search_api_key,
+                            write_api_key: new_write_api_key,
+                        } = new;
+
+                        let merged = DbRemote {
+                            url: match new_url {
+                                Setting::Set(new_url) => new_url,
+                                Setting::Reset => {
+                                    return err(milli::UserError::NetworkMissingUrl(name))
+                                }
+                                Setting::NotSet => old_url,
+                            },
+                            search_api_key: match new_search_api_key {
+                                Setting::Set(new_search_api_key) => Some(new_search_api_key),
+                                Setting::Reset => None,
+                                Setting::NotSet => old_search_api_key,
+                            },
+                            write_api_key: match new_write_api_key {
+                                Setting::Set(new_write_api_key) => Some(new_write_api_key),
+                                Setting::Reset => None,
+                                Setting::NotSet => old_write_api_key,
+                            },
+                        };
+                        merged_remotes.insert(name, merged);
+                    }
+                    EitherOrBoth::Both((_, _), (_, None)) | EitherOrBoth::Right((_, None)) => {}
+                    EitherOrBoth::Left((name, node)) => {
+                        merged_remotes.insert(name, node);
+                    }
+                    EitherOrBoth::Right((name, Some(node))) => {
+                        let Some(url) = node.url.set() else {
+                            return err(milli::UserError::NetworkMissingUrl(name));
+                        };
+                        let node = DbRemote {
+                            url,
+                            search_api_key: node.search_api_key.set(),
+                            write_api_key: node.write_api_key.set(),
+                        };
+                        merged_remotes.insert(name, node);
+                    }
+                }
+            }
+            merged_remotes
+        }
+        Setting::Reset => BTreeMap::new(),
+        Setting::NotSet => old_network.remotes,
+    };
+
+    Ok(DbNetwork { local: merged_local, remotes: merged_remotes, sharding: merged_sharding })
+}

crates/index-scheduler/src/scheduler/mod.rs

@@ -2,6 +2,7 @@ mod autobatcher;
 #[cfg(test)]
 mod autobatcher_test;
 mod create_batch;
+mod enterprise_edition;
 mod process_batch;
 mod process_dump_creation;
 mod process_export;

crates/index-scheduler/src/scheduler/process_batch.rs

@@ -135,6 +135,9 @@ impl IndexScheduler {
             Batch::Dump(task) => self
                 .process_dump_creation(progress, task)
                 .map(|tasks| (tasks, ProcessBatchInfo::default())),
+            Batch::NetworkTopologyChanges { tasks } => self
+                .process_network_changes(progress, tasks)
+                .map(|tasks| (tasks, ProcessBatchInfo::default())),
             Batch::IndexOperation { op, must_create_index } => {
                 let index_uid = op.index_uid().to_string();
                 let index = if must_create_index {

crates/index-scheduler/src/scheduler/process_export.rs

@@ -16,6 +16,7 @@ use meilisearch_types::milli::vector::parsed_vectors::{ExplicitVectors, VectorOr
 use meilisearch_types::milli::{self, obkv_to_json, Filter, InternalError};
 use meilisearch_types::settings::{self, SecretPolicy};
 use meilisearch_types::tasks::{DetailsExportIndexSettings, ExportIndexSettings};
+use roaring::RoaringBitmap;
 use serde::Deserialize;
 use ureq::{json, Response};
 
@@ -50,6 +51,7 @@ impl IndexScheduler {
         let agent = ureq::AgentBuilder::new().timeout(Duration::from_secs(5)).build();
         let must_stop_processing = self.scheduler.must_stop_processing.clone();
         for (i, (_pattern, uid, export_settings)) in indexes.iter().enumerate() {
+            let err = |err| Error::from_milli(err, Some(uid.to_string()));
             if must_stop_processing.get() {
                 return Err(Error::AbortedTask);
             }
@@ -61,104 +63,31 @@ impl IndexScheduler {
             ));
 
             let ExportIndexSettings { filter, override_settings } = export_settings;
+
             let index = self.index(uid)?;
             let index_rtxn = index.read_txn()?;
-            let bearer = api_key.map(|api_key| format!("Bearer {api_key}"));
-
-            // First, check if the index already exists
-            let url = format!("{base_url}/indexes/{uid}");
-            let response = retry(&must_stop_processing, || {
-                let mut request = agent.get(&url);
-                if let Some(bearer) = &bearer {
-                    request = request.set("Authorization", bearer);
-                }
-
-                request.send_bytes(Default::default()).map_err(into_backoff_error)
-            });
-            let index_exists = match response {
-                Ok(response) => response.status() == 200,
-                Err(Error::FromRemoteWhenExporting { code, .. }) if code == "index_not_found" => {
-                    false
-                }
-                Err(e) => return Err(e),
-            };
-
-            let primary_key = index
-                .primary_key(&index_rtxn)
-                .map_err(|e| Error::from_milli(e.into(), Some(uid.to_string())))?;
-
-            // Create the index
-            if !index_exists {
-                let url = format!("{base_url}/indexes");
-                retry(&must_stop_processing, || {
-                    let mut request = agent.post(&url);
-                    if let Some(bearer) = &bearer {
-                        request = request.set("Authorization", bearer);
-                    }
-                    let index_param = json!({ "uid": uid, "primaryKey": primary_key });
-                    request.send_json(&index_param).map_err(into_backoff_error)
-                })?;
-            }
-
-            // Patch the index primary key
-            if index_exists && *override_settings {
-                let url = format!("{base_url}/indexes/{uid}");
-                retry(&must_stop_processing, || {
-                    let mut request = agent.patch(&url);
-                    if let Some(bearer) = &bearer {
-                        request = request.set("Authorization", bearer);
-                    }
-                    let index_param = json!({ "primaryKey": primary_key });
-                    request.send_json(&index_param).map_err(into_backoff_error)
-                })?;
-            }
-
-            // Send the index settings
-            if !index_exists || *override_settings {
-                let mut settings =
-                    settings::settings(&index, &index_rtxn, SecretPolicy::RevealSecrets)
-                        .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
-                // Remove the experimental chat setting if not enabled
-                if self.features().check_chat_completions("exporting chat settings").is_err() {
-                    settings.chat = Setting::NotSet;
-                }
-                // Retry logic for sending settings
-                let url = format!("{base_url}/indexes/{uid}/settings");
-                retry(&must_stop_processing, || {
-                    let mut request = agent.patch(&url);
-                    if let Some(bearer) = bearer.as_ref() {
-                        request = request.set("Authorization", bearer);
-                    }
-                    request.send_json(settings.clone()).map_err(into_backoff_error)
-                })?;
-            }
-
-            let filter = filter
-                .as_ref()
-                .map(Filter::from_json)
-                .transpose()
-                .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?
-                .flatten();
-
-            let filter_universe = filter
-                .map(|f| f.evaluate(&index_rtxn, &index))
-                .transpose()
-                .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
-            let whole_universe = index
-                .documents_ids(&index_rtxn)
-                .map_err(|e| Error::from_milli(e.into(), Some(uid.to_string())))?;
+            let filter = filter.as_ref().map(Filter::from_json).transpose().map_err(err)?.flatten();
+            let filter_universe =
+                filter.map(|f| f.evaluate(&index_rtxn, &index)).transpose().map_err(err)?;
+            let whole_universe =
+                index.documents_ids(&index_rtxn).map_err(milli::Error::from).map_err(err)?;
             let universe = filter_universe.unwrap_or(whole_universe);
-
-            let fields_ids_map = index.fields_ids_map(&index_rtxn)?;
-            let all_fields: Vec<_> = fields_ids_map.iter().map(|(id, _)| id).collect();
-
-            // We don't need to keep this one alive as we will
-            // spawn many threads to process the documents
-            drop(index_rtxn);
-
-            let total_documents = universe.len() as u32;
-            let (step, progress_step) = AtomicDocumentStep::new(total_documents);
-            progress.update_progress(progress_step);
+            let target = TargetInstance { base_url, api_key };
+            let ctx = ExportContext {
+                index: &index,
+                index_rtxn: &index_rtxn,
+                universe: &universe,
+                progress: &progress,
+                agent: &agent,
+                must_stop_processing: &must_stop_processing,
+            };
+            let options = ExportOptions {
+                index_uid: uid,
+                payload_size,
+                override_settings: *override_settings,
+                extra_headers: &Default::default(),
+            };
+            let total_documents = self.export_one_index(target, options, ctx)?;
 
             output.insert(
                 IndexUidPattern::new_unchecked(uid.clone()),
@@ -167,155 +96,217 @@ impl IndexScheduler {
                     matched_documents: Some(total_documents as u64),
                 },
             );
-
-            let limit = payload_size.map(|ps| ps.as_u64() as usize).unwrap_or(20 * 1024 * 1024); // defaults to 20 MiB
-            let documents_url = format!("{base_url}/indexes/{uid}/documents");
-
-            let results = request_threads()
-                .broadcast(|ctx| {
-                    let index_rtxn = index
-                        .read_txn()
-                        .map_err(|e| Error::from_milli(e.into(), Some(uid.to_string())))?;
-
-                    let mut buffer = Vec::new();
-                    let mut tmp_buffer = Vec::new();
-                    let mut compressed_buffer = Vec::new();
-                    for (i, docid) in universe.iter().enumerate() {
-                        if i % ctx.num_threads() != ctx.index() {
-                            continue;
-                        }
-
-                        let document = index
-                            .document(&index_rtxn, docid)
-                            .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
-
-                        let mut document = obkv_to_json(&all_fields, &fields_ids_map, document)
-                            .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
-
-                        // TODO definitely factorize this code
-                        'inject_vectors: {
-                            let embeddings = index
-                                .embeddings(&index_rtxn, docid)
-                                .map_err(|e| Error::from_milli(e, Some(uid.to_string())))?;
-
-                            if embeddings.is_empty() {
-                                break 'inject_vectors;
-                            }
-
-                            let vectors = document
-                                .entry(RESERVED_VECTORS_FIELD_NAME)
-                                .or_insert(serde_json::Value::Object(Default::default()));
-
-                            let serde_json::Value::Object(vectors) = vectors else {
-                                return Err(Error::from_milli(
-                                    milli::Error::UserError(
-                                        milli::UserError::InvalidVectorsMapType {
-                                            document_id: {
-                                                if let Ok(Some(Ok(index))) = index
-                                                    .external_id_of(
-                                                        &index_rtxn,
-                                                        std::iter::once(docid),
-                                                    )
-                                                    .map(|it| it.into_iter().next())
-                                                {
-                                                    index
-                                                } else {
-                                                    format!("internal docid={docid}")
-                                                }
-                                            },
-                                            value: vectors.clone(),
-                                        },
-                                    ),
-                                    Some(uid.to_string()),
-                                ));
-                            };
-
-                            for (
-                                embedder_name,
-                                EmbeddingsWithMetadata { embeddings, regenerate, has_fragments },
-                            ) in embeddings
-                            {
-                                let embeddings = ExplicitVectors {
-                                    embeddings: Some(
-                                        VectorOrArrayOfVectors::from_array_of_vectors(embeddings),
-                                    ),
-                                    regenerate: regenerate &&
-                                    // Meilisearch does not handle well dumps with fragments, because as the fragments
-                                    // are marked as user-provided,
-                                    // all embeddings would be regenerated on any settings change or document update.
-                                    // To prevent this, we mark embeddings has non regenerate in this case.
-                                    !has_fragments,
-                                };
-                                vectors.insert(
-                                    embedder_name,
-                                    serde_json::to_value(embeddings).unwrap(),
-                                );
-                            }
-                        }
-
-                        tmp_buffer.clear();
-                        serde_json::to_writer(&mut tmp_buffer, &document)
-                            .map_err(milli::InternalError::from)
-                            .map_err(|e| Error::from_milli(e.into(), Some(uid.to_string())))?;
-
-                        // Make sure we put at least one document in the buffer even
-                        // though we might go above the buffer limit before sending
-                        if !buffer.is_empty() && buffer.len() + tmp_buffer.len() > limit {
-                            // We compress the documents before sending them
-                            let mut encoder =
-                                GzEncoder::new(&mut compressed_buffer, Compression::default());
-                            encoder
-                                .write_all(&buffer)
-                                .map_err(|e| Error::from_milli(e.into(), Some(uid.clone())))?;
-                            encoder
-                                .finish()
-                                .map_err(|e| Error::from_milli(e.into(), Some(uid.clone())))?;
-
-                            retry(&must_stop_processing, || {
-                                let mut request = agent.post(&documents_url);
-                                request = request.set("Content-Type", "application/x-ndjson");
-                                request = request.set("Content-Encoding", "gzip");
-                                if let Some(bearer) = &bearer {
-                                    request = request.set("Authorization", bearer);
-                                }
-                                request.send_bytes(&compressed_buffer).map_err(into_backoff_error)
-                            })?;
-                            buffer.clear();
-                            compressed_buffer.clear();
-                        }
-                        buffer.extend_from_slice(&tmp_buffer);
-
-                        if i > 0 && i % 100 == 0 {
-                            step.fetch_add(100, atomic::Ordering::Relaxed);
-                        }
-                    }
-
-                    retry(&must_stop_processing, || {
-                        let mut request = agent.post(&documents_url);
-                        request = request.set("Content-Type", "application/x-ndjson");
-                        if let Some(bearer) = &bearer {
-                            request = request.set("Authorization", bearer);
-                        }
-                        request.send_bytes(&buffer).map_err(into_backoff_error)
-                    })?;
-
-                    Ok(())
-                })
-                .map_err(|e| {
-                    Error::from_milli(
-                        milli::Error::InternalError(InternalError::PanicInThreadPool(e)),
-                        Some(uid.to_string()),
-                    )
-                })?;
-            for result in results {
-                result?;
-            }
-
-            step.store(total_documents, atomic::Ordering::Relaxed);
         }
 
         Ok(output)
     }
+
+    pub(super) fn export_one_index(
+        &self,
+        target: TargetInstance<'_>,
+        options: ExportOptions<'_>,
+        ctx: ExportContext<'_>,
+    ) -> Result<u64, Error> {
+        let err = |err| Error::from_milli(err, Some(options.index_uid.to_string()));
+
+        let bearer = target.api_key.map(|api_key| format!("Bearer {api_key}"));
+        let url = format!(
+            "{base_url}/indexes/{index_uid}",
+            base_url = target.base_url,
+            index_uid = options.index_uid
+        );
+        let response = retry(ctx.must_stop_processing, || {
+            let mut request = ctx.agent.get(&url);
+            if let Some(bearer) = &bearer {
+                request = request.set("Authorization", bearer);
+            }
+
+            request.send_bytes(Default::default()).map_err(into_backoff_error)
+        });
+        let index_exists = match response {
+            Ok(response) => response.status() == 200,
+            Err(Error::FromRemoteWhenExporting { code, .. }) if code == "index_not_found" => false,
+            Err(e) => return Err(e),
+        };
+        let primary_key =
+            ctx.index.primary_key(&ctx.index_rtxn).map_err(milli::Error::from).map_err(err)?;
+        if !index_exists {
+            let url = format!("{base_url}/indexes", base_url = target.base_url);
+            retry(ctx.must_stop_processing, || {
+                let mut request = ctx.agent.post(&url);
+                if let Some(bearer) = &bearer {
+                    request = request.set("Authorization", bearer);
+                }
+                let index_param = json!({ "uid": options.index_uid, "primaryKey": primary_key });
+                request.send_json(&index_param).map_err(into_backoff_error)
+            })?;
+        }
+        if index_exists && options.override_settings {
+            retry(ctx.must_stop_processing, || {
+                let mut request = ctx.agent.patch(&url);
+                if let Some(bearer) = &bearer {
+                    request = request.set("Authorization", bearer);
+                }
+                let index_param = json!({ "primaryKey": primary_key });
+                request.send_json(&index_param).map_err(into_backoff_error)
+            })?;
+        }
+        if !index_exists || options.override_settings {
+            let mut settings =
+                settings::settings(&ctx.index, &ctx.index_rtxn, SecretPolicy::RevealSecrets)
+                    .map_err(err)?;
+            // Remove the experimental chat setting if not enabled
+            if self.features().check_chat_completions("exporting chat settings").is_err() {
+                settings.chat = Setting::NotSet;
+            }
+            // Retry logic for sending settings
+            let url = format!(
+                "{base_url}/indexes/{index_uid}/settings",
+                base_url = target.base_url,
+                index_uid = options.index_uid
+            );
+            retry(ctx.must_stop_processing, || {
+                let mut request = ctx.agent.patch(&url);
+                if let Some(bearer) = bearer.as_ref() {
+                    request = request.set("Authorization", bearer);
+                }
+                request.send_json(settings.clone()).map_err(into_backoff_error)
+            })?;
+        }
+
+        let fields_ids_map = ctx.index.fields_ids_map(&ctx.index_rtxn)?;
+        let all_fields: Vec<_> = fields_ids_map.iter().map(|(id, _)| id).collect();
+        let total_documents = ctx.universe.len() as u32;
+        let (step, progress_step) = AtomicDocumentStep::new(total_documents);
+        ctx.progress.update_progress(progress_step);
+
+        let limit = options.payload_size.map(|ps| ps.as_u64() as usize).unwrap_or(20 * 1024 * 1024);
+        let documents_url = format!(
+            "{base_url}/indexes/{index_uid}/documents",
+            base_url = target.base_url,
+            index_uid = options.index_uid
+        );
+        let results = request_threads()
+            .broadcast(|broadcast| {
+                let index_rtxn = ctx.index.read_txn().map_err(milli::Error::from).map_err(err)?;
+
+                let mut buffer = Vec::new();
+                let mut tmp_buffer = Vec::new();
+                let mut compressed_buffer = Vec::new();
+                for (i, docid) in ctx.universe.iter().enumerate() {
+                    if i % broadcast.num_threads() != broadcast.index() {
+                        continue;
+                    }
+
+                    let document = ctx.index.document(&index_rtxn, docid).map_err(err)?;
+
+                    let mut document =
+                        obkv_to_json(&all_fields, &fields_ids_map, document).map_err(err)?;
+
+                    // TODO definitely factorize this code
+                    'inject_vectors: {
+                        let embeddings = ctx.index.embeddings(&index_rtxn, docid).map_err(err)?;
+
+                        if embeddings.is_empty() {
+                            break 'inject_vectors;
+                        }
+
+                        let vectors = document
+                            .entry(RESERVED_VECTORS_FIELD_NAME)
+                            .or_insert(serde_json::Value::Object(Default::default()));
+
+                        let serde_json::Value::Object(vectors) = vectors else {
+                            return Err(err(milli::Error::UserError(
+                                milli::UserError::InvalidVectorsMapType {
+                                    document_id: {
+                                        if let Ok(Some(Ok(index))) = ctx
+                                            .index
+                                            .external_id_of(&index_rtxn, std::iter::once(docid))
+                                            .map(|it| it.into_iter().next())
+                                        {
+                                            index
+                                        } else {
+                                            format!("internal docid={docid}")
+                                        }
+                                    },
+                                    value: vectors.clone(),
+                                },
+                            )));
+                        };
+
+                        for (
+                            embedder_name,
+                            EmbeddingsWithMetadata { embeddings, regenerate, has_fragments },
+                        ) in embeddings
+                        {
+                            let embeddings = ExplicitVectors {
+                                embeddings: Some(VectorOrArrayOfVectors::from_array_of_vectors(
+                                    embeddings,
+                                )),
+                                regenerate: regenerate &&
+                                // Meilisearch does not handle well dumps with fragments, because as the fragments
+                                // are marked as user-provided,
+                                // all embeddings would be regenerated on any settings change or document update.
+                                // To prevent this, we mark embeddings has non regenerate in this case.
+                                !has_fragments,
+                            };
+                            vectors
+                                .insert(embedder_name, serde_json::to_value(embeddings).unwrap());
+                        }
+                    }
+
+                    tmp_buffer.clear();
+                    serde_json::to_writer(&mut tmp_buffer, &document)
+                        .map_err(milli::InternalError::from)
+                        .map_err(milli::Error::from)
+                        .map_err(err)?;
+
+                    // Make sure we put at least one document in the buffer even
+                    // though we might go above the buffer limit before sending
+                    if !buffer.is_empty() && buffer.len() + tmp_buffer.len() > limit {
+                        // We compress the documents before sending them
+                        let mut encoder =
+                            GzEncoder::new(&mut compressed_buffer, Compression::default());
+                        encoder.write_all(&buffer).map_err(milli::Error::from).map_err(err)?;
+                        encoder.finish().map_err(milli::Error::from).map_err(err)?;
+
+                        retry(ctx.must_stop_processing, || {
+                            let mut request = ctx.agent.post(&documents_url);
+                            request = request.set("Content-Type", "application/x-ndjson");
+                            request = request.set("Content-Encoding", "gzip");
+                            if let Some(bearer) = &bearer {
+                                request = request.set("Authorization", bearer);
+                            }
+                            request.send_bytes(&compressed_buffer).map_err(into_backoff_error)
+                        })?;
+                        buffer.clear();
+                        compressed_buffer.clear();
+                    }
+                    buffer.extend_from_slice(&tmp_buffer);
+
+                    if i > 0 && i % 100 == 0 {
+                        step.fetch_add(100, atomic::Ordering::Relaxed);
+                    }
+                }
+
+                retry(ctx.must_stop_processing, || {
+                    let mut request = ctx.agent.post(&documents_url);
+                    request = request.set("Content-Type", "application/x-ndjson");
+                    if let Some(bearer) = &bearer {
+                        request = request.set("Authorization", bearer);
+                    }
+                    request.send_bytes(&buffer).map_err(into_backoff_error)
+                })?;
+
+                Ok(())
+            })
+            .map_err(|e| err(milli::Error::InternalError(InternalError::PanicInThreadPool(e))))?;
+        for result in results {
+            result?;
+        }
+        step.store(total_documents, atomic::Ordering::Relaxed);
+        Ok(total_documents as u64)
+    }
 }
 
 fn retry<F>(must_stop_processing: &MustStopProcessing, send_request: F) -> Result<ureq::Response>
@@ -374,4 +365,27 @@ fn ureq_error_into_error(error: ureq::Error) -> Error {
     }
 }
 
+// export_one_index arguments
+pub(super) struct TargetInstance<'a> {
+    pub(super) base_url: &'a str,
+    pub(super) api_key: Option<&'a str>,
+}
+
+pub(super) struct ExportOptions<'a> {
+    pub(super) index_uid: &'a str,
+    pub(super) payload_size: Option<&'a Byte>,
+    pub(super) override_settings: bool,
+    pub(super) extra_headers: &'a hashbrown::HashMap<String, String>,
+}
+
+pub(super) struct ExportContext<'a> {
+    pub(super) index: &'a meilisearch_types::milli::Index,
+    pub(super) index_rtxn: &'a milli::heed::RoTxn<'a>,
+    pub(super) universe: &'a RoaringBitmap,
+    pub(super) progress: &'a Progress,
+    pub(super) agent: &'a ureq::Agent,
+    pub(super) must_stop_processing: &'a MustStopProcessing,
+}
+
+// progress related
 enum ExportIndex {}

crates/index-scheduler/src/scheduler/process_index_operation.rs

@@ -97,7 +97,7 @@ impl IndexScheduler {
                     .embedding_configs()
                     .embedding_configs(index_wtxn)
                     .map_err(|e| Error::from_milli(e.into(), Some(index_uid.clone())))?;
-                let embedders = self.embedders(index_uid.clone(), embedders)?;
+                let embedders = self.embedders(&index_uid, embedders)?;
                 for operation in operations {
                     match operation {
                         DocumentOperation::Replace(_content_uuid) => {
@@ -284,7 +284,7 @@ impl IndexScheduler {
                         .embedding_configs()
                         .embedding_configs(index_wtxn)
                         .map_err(|err| Error::from_milli(err.into(), Some(index_uid.clone())))?;
-                    let embedders = self.embedders(index_uid.clone(), embedders)?;
+                    let embedders = self.embedders(&index_uid, embedders)?;
 
                     progress.update_progress(DocumentEditionProgress::Indexing);
                     congestion = Some(
@@ -434,7 +434,7 @@ impl IndexScheduler {
                         .embedding_configs()
                         .embedding_configs(index_wtxn)
                         .map_err(|err| Error::from_milli(err.into(), Some(index_uid.clone())))?;
-                    let embedders = self.embedders(index_uid.clone(), embedders)?;
+                    let embedders = self.embedders(&index_uid, embedders)?;
 
                     progress.update_progress(DocumentDeletionProgress::Indexing);
                     congestion = Some(

crates/index-scheduler/src/scheduler/test.rs

@@ -722,7 +722,7 @@ fn basic_get_stats() {
     let kind = index_creation_task("whalo", "fish");
     let _task = index_scheduler.register(kind, None, false).unwrap();
 
-    snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r#"
+    snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r###"
     {
       "indexes": {
         "catto": 1,
@@ -746,6 +746,7 @@ fn basic_get_stats() {
         "indexDeletion": 0,
         "indexSwap": 0,
         "indexUpdate": 0,
+        "networkTopologyChange": 0,
         "settingsUpdate": 0,
         "snapshotCreation": 0,
         "taskCancelation": 0,
@@ -753,7 +754,7 @@ fn basic_get_stats() {
         "upgradeDatabase": 0
       }
     }
-    "#);
+    "###);
 
     handle.advance_till([Start, BatchCreated]);
     snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r#"

crates/index-scheduler/src/scheduler/test_embedders.rs

@@ -121,7 +121,7 @@ fn import_vectors() {
         insta::assert_json_snapshot!(simple_hf_config.embedder_options);
         let simple_hf_name = name.clone();
 
-        let configs = index_scheduler.embedders("doggos".to_string(), configs).unwrap();
+        let configs = index_scheduler.embedders("doggos", configs).unwrap();
         let hf_runtime = configs.get(&simple_hf_name).unwrap();
         let hf_embedder = &hf_runtime.embedder;
         let beagle_embed = hf_embedder

crates/index-scheduler/src/test_utils.rs

@@ -126,7 +126,7 @@ impl IndexScheduler {
         std::fs::create_dir_all(&options.auth_path).unwrap();
         let auth_env = open_auth_store_env(&options.auth_path).unwrap();
         let index_scheduler =
-            Self::new(options, auth_env, version, sender, planned_failures).unwrap();
+            Self::new(options, auth_env, version, None, sender, planned_failures).unwrap();
 
         // To be 100% consistent between all test we're going to start the scheduler right now
         // and ensure it's in the expected starting state.

crates/index-scheduler/src/utils.rs

@@ -285,6 +285,7 @@ pub fn swap_index_uid_in_task(task: &mut Task, swap: (&str, &str)) {
         | K::DumpCreation { .. }
         | K::Export { .. }
         | K::UpgradeDatabase { .. }
+        | K::NetworkTopologyChange { .. }
         | K::SnapshotCreation => (),
     };
     if let Some(Details::IndexSwap { swaps }) = &mut task.details {
@@ -618,6 +619,9 @@ impl crate::IndexScheduler {
                     Details::UpgradeDatabase { from: _, to: _ } => {
                         assert_eq!(kind.as_kind(), Kind::UpgradeDatabase);
                     }
+                    Details::NetworkTopologyChange { .. } => {
+                        assert_eq!(kind.as_kind(), Kind::NetworkTopologyChange);
+                    }
                 }
             }
 

crates/meili-snap/src/lib.rs

@@ -271,9 +271,10 @@ macro_rules! json_string {
 
 #[cfg(test)]
 mod tests {
+    use uuid::Uuid;
+
     use crate as meili_snap;
     use crate::UUID_IN_MESSAGE_RE;
-    use uuid::Uuid;
 
     #[test]
     fn snap() {

crates/meilisearch-types/src/enterprise_edition/network.rs

@@ -5,31 +5,85 @@
 
 use std::collections::BTreeMap;
 
-use milli::update::new::indexer::enterprise_edition::sharding::Shards;
+use deserr::Deserr;
+use milli::update::new::indexer::enterprise_edition::sharding::{Shard, Shards};
+use milli::update::Setting;
 use serde::{Deserialize, Serialize};
+use utoipa::ToSchema;
+
+use crate::deserr::DeserrJsonError;
+use crate::error::deserr_codes::{
+    InvalidNetworkRemotes, InvalidNetworkSearchApiKey, InvalidNetworkSelf, InvalidNetworkSharding,
+    InvalidNetworkUrl, InvalidNetworkWriteApiKey,
+};
+
+#[derive(Clone, Debug, Deserr, ToSchema, Serialize, Deserialize, PartialEq, Eq)]
+#[deserr(error = DeserrJsonError, rename_all = camelCase, deny_unknown_fields)]
+#[serde(rename_all = "camelCase")]
+#[schema(rename_all = "camelCase")]
+pub struct Network {
+    #[schema(value_type = Option<BTreeMap<String, Remote>>, example = json!("http://localhost:7700"))]
+    #[deserr(default, error = DeserrJsonError<InvalidNetworkRemotes>)]
+    #[serde(default)]
+    pub remotes: Setting<BTreeMap<String, Option<Remote>>>,
+    #[schema(value_type = Option<String>, example = json!("ms-00"), rename = "self")]
+    #[serde(default, rename = "self")]
+    #[deserr(default, rename = "self", error = DeserrJsonError<InvalidNetworkSelf>)]
+    pub local: Setting<String>,
+    #[schema(value_type = Option<bool>, example = json!(true))]
+    #[serde(default)]
+    #[deserr(default, error = DeserrJsonError<InvalidNetworkSharding>)]
+    pub sharding: Setting<bool>,
+}
+
+#[derive(Clone, Debug, Deserr, ToSchema, Serialize, Deserialize, PartialEq, Eq)]
+#[deserr(error = DeserrJsonError<InvalidNetworkRemotes>, rename_all = camelCase, deny_unknown_fields)]
+#[serde(rename_all = "camelCase")]
+#[schema(rename_all = "camelCase")]
+pub struct Remote {
+    #[schema(value_type = Option<String>, example = json!({
+        "ms-0": Remote { url: Setting::Set("http://localhost:7700".into()), search_api_key: Setting::Reset, write_api_key: Setting::Reset },
+        "ms-1": Remote { url: Setting::Set("http://localhost:7701".into()), search_api_key: Setting::Set("foo".into()), write_api_key: Setting::Set("bar".into()) },
+        "ms-2": Remote { url: Setting::Set("http://localhost:7702".into()), search_api_key: Setting::Set("bar".into()), write_api_key: Setting::Set("foo".into()) },
+    }))]
+    #[deserr(default, error = DeserrJsonError<InvalidNetworkUrl>)]
+    #[serde(default)]
+    pub url: Setting<String>,
+    #[schema(value_type = Option<String>, example = json!("XWnBI8QHUc-4IlqbKPLUDuhftNq19mQtjc6JvmivzJU"))]
+    #[deserr(default, error = DeserrJsonError<InvalidNetworkSearchApiKey>)]
+    #[serde(default)]
+    pub search_api_key: Setting<String>,
+    #[schema(value_type = Option<String>, example = json!("XWnBI8QHUc-4IlqbKPLUDuhftNq19mQtjc6JvmivzJU"))]
+    #[deserr(default, error = DeserrJsonError<InvalidNetworkWriteApiKey>)]
+    #[serde(default)]
+    pub write_api_key: Setting<String>,
+}
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, Default)]
 #[serde(rename_all = "camelCase")]
-pub struct Network {
+pub struct DbNetwork {
     #[serde(default, rename = "self")]
     pub local: Option<String>,
     #[serde(default)]
-    pub remotes: BTreeMap<String, Remote>,
+    pub remotes: BTreeMap<String, DbRemote>,
     #[serde(default)]
     pub sharding: bool,
 }
 
-impl Network {
+impl DbNetwork {
     pub fn shards(&self) -> Option<Shards> {
         if self.sharding {
-            let this = self.local.as_deref().expect("Inconsistent `sharding` and `self`");
-            let others = self
-                .remotes
-                .keys()
-                .filter(|name| name.as_str() != this)
-                .map(|name| name.to_owned())
-                .collect();
-            Some(Shards { own: vec![this.to_owned()], others })
+            let this = self.local.as_deref();
+
+            Some(Shards(
+                self.remotes
+                    .keys()
+                    .map(|name| Shard {
+                        is_own: Some(name.as_str()) == this,
+                        name: name.to_owned(),
+                    })
+                    .collect(),
+            ))
         } else {
             None
         }
@@ -38,7 +92,7 @@ impl Network {
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
 #[serde(rename_all = "camelCase")]
-pub struct Remote {
+pub struct DbRemote {
     pub url: String,
     #[serde(default)]
     pub search_api_key: Option<String>,

crates/meilisearch-types/src/error.rs

@@ -5,6 +5,7 @@ use actix_web::{self as aweb, HttpResponseBuilder};
 use aweb::http::header;
 use aweb::rt::task::JoinError;
 use convert_case::Casing;
+use milli::cellulite;
 use milli::heed::{Error as HeedError, MdbError};
 use serde::{Deserialize, Serialize};
 use utoipa::ToSchema;
@@ -239,6 +240,7 @@ InconsistentDocumentChangeHeaders              , InvalidRequest       , BAD_REQU
 InvalidDocumentFilter                          , InvalidRequest       , BAD_REQUEST ;
 InvalidDocumentSort                            , InvalidRequest       , BAD_REQUEST ;
 InvalidDocumentGeoField                        , InvalidRequest       , BAD_REQUEST ;
+InvalidDocumentGeojsonField                    , InvalidRequest       , BAD_REQUEST ;
 InvalidHeaderValue                             , InvalidRequest       , BAD_REQUEST ;
 InvalidVectorDimensions                        , InvalidRequest       , BAD_REQUEST ;
 InvalidVectorsType                             , InvalidRequest       , BAD_REQUEST ;
@@ -501,7 +503,9 @@ impl ErrorCode for milli::Error {
                         Code::InvalidFacetSearchFacetName
                     }
                     UserError::CriterionError(_) => Code::InvalidSettingsRankingRules,
-                    UserError::InvalidGeoField { .. } => Code::InvalidDocumentGeoField,
+                    UserError::InvalidGeoField { .. } | UserError::GeoJsonError(_) => {
+                        Code::InvalidDocumentGeoField
+                    }
                     UserError::InvalidVectorDimensions { .. }
                     | UserError::InvalidIndexingVectorDimensions { .. } => {
                         Code::InvalidVectorDimensions
@@ -525,6 +529,19 @@ impl ErrorCode for milli::Error {
                     | UserError::DocumentEditionCompilationError(_) => {
                         Code::EditDocumentsByFunctionError
                     }
+                    UserError::NetworkShardingWithoutSelf => Code::InvalidNetworkSharding,
+                    UserError::NetworkMissingUrl(_) => Code::MissingNetworkUrl,
+                    UserError::CelluliteError(err) => match err {
+                        cellulite::Error::BuildCanceled
+                        | cellulite::Error::VersionMismatchOnBuild(_)
+                        | cellulite::Error::DatabaseDoesntExists
+                        | cellulite::Error::Heed(_)
+                        | cellulite::Error::InvalidGeometry(_)
+                        | cellulite::Error::InternalDocIdMissing(_, _)
+                        | cellulite::Error::CannotConvertLineToCell(_, _, _) => Code::Internal,
+                        cellulite::Error::InvalidGeoJson(_) => Code::InvalidDocumentGeojsonField,
+                    },
+                    UserError::MalformedGeojson(_) => Code::InvalidDocumentGeojsonField,
                 }
             }
         }

crates/meilisearch-types/src/lib.rs

@@ -1,3 +1,5 @@
+#![allow(clippy::result_large_err)]
+
 pub mod batch_view;
 pub mod batches;
 pub mod compression;

crates/meilisearch-types/src/task_view.rs

@@ -7,6 +7,7 @@ use time::{Duration, OffsetDateTime};
 use utoipa::ToSchema;
 
 use crate::batches::BatchId;
+use crate::enterprise_edition::network::Network;
 use crate::error::ResponseError;
 use crate::settings::{Settings, Unchecked};
 use crate::tasks::{
@@ -142,6 +143,9 @@ pub struct DetailsView {
     pub old_index_uid: Option<String>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub new_index_uid: Option<String>,
+    // network
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub network: Option<Network>,
 }
 
 impl DetailsView {
@@ -314,6 +318,10 @@ impl DetailsView {
                 // We should never be able to batch multiple renames at the same time.
                 (Some(left), Some(_right)) => Some(left),
             },
+            network: match (&self.network, &other.network) {
+                (None, None) => None,
+                (_, Some(network)) | (Some(network), None) => Some(network.clone()),
+            },
         }
     }
 }
@@ -415,6 +423,9 @@ impl From<Details> for DetailsView {
                 upgrade_to: Some(format!("v{}.{}.{}", to.0, to.1, to.2)),
                 ..Default::default()
             },
+            Details::NetworkTopologyChange { network: new_network } => {
+                DetailsView { network: new_network, ..Default::default() }
+            }
         }
     }
 }

crates/meilisearch-types/src/tasks.rs

@@ -15,6 +15,7 @@ use utoipa::{schema, ToSchema};
 use uuid::Uuid;
 
 use crate::batches::BatchId;
+use crate::enterprise_edition::network::Network;
 use crate::error::ResponseError;
 use crate::index_uid_pattern::IndexUidPattern;
 use crate::keys::Key;
@@ -58,6 +59,7 @@ impl Task {
             | TaskDeletion { .. }
             | Export { .. }
             | UpgradeDatabase { .. }
+            | NetworkTopologyChange { .. }
             | IndexSwap { .. } => None,
             DocumentAdditionOrUpdate { index_uid, .. }
             | DocumentEdition { index_uid, .. }
@@ -94,7 +96,8 @@ impl Task {
             | KindWithContent::DumpCreation { .. }
             | KindWithContent::SnapshotCreation
             | KindWithContent::Export { .. }
-            | KindWithContent::UpgradeDatabase { .. } => None,
+            | KindWithContent::UpgradeDatabase { .. }
+            | KindWithContent::NetworkTopologyChange { .. } => None,
         }
     }
 }
@@ -170,6 +173,10 @@ pub enum KindWithContent {
     UpgradeDatabase {
         from: (u32, u32, u32),
     },
+    NetworkTopologyChange {
+        network: Option<Network>,
+        origin: Option<Origin>,
+    },
 }
 
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, ToSchema)]
@@ -206,6 +213,7 @@ impl KindWithContent {
             KindWithContent::SnapshotCreation => Kind::SnapshotCreation,
             KindWithContent::Export { .. } => Kind::Export,
             KindWithContent::UpgradeDatabase { .. } => Kind::UpgradeDatabase,
+            KindWithContent::NetworkTopologyChange { .. } => Kind::NetworkTopologyChange,
         }
     }
 
@@ -218,6 +226,7 @@ impl KindWithContent {
             | TaskCancelation { .. }
             | TaskDeletion { .. }
             | Export { .. }
+            | NetworkTopologyChange { .. }
             | UpgradeDatabase { .. } => vec![],
             DocumentAdditionOrUpdate { index_uid, .. }
             | DocumentEdition { index_uid, .. }
@@ -325,6 +334,9 @@ impl KindWithContent {
                     versioning::VERSION_PATCH,
                 ),
             }),
+            KindWithContent::NetworkTopologyChange { network: new_network, origin: _ } => {
+                Some(Details::NetworkTopologyChange { network: new_network.clone() })
+            }
         }
     }
 
@@ -407,6 +419,9 @@ impl KindWithContent {
                     versioning::VERSION_PATCH,
                 ),
             }),
+            KindWithContent::NetworkTopologyChange { network: new_network, origin: _s } => {
+                Some(Details::NetworkTopologyChange { network: new_network.clone() })
+            }
         }
     }
 }
@@ -469,6 +484,9 @@ impl From<&KindWithContent> for Option<Details> {
                     versioning::VERSION_PATCH,
                 ),
             }),
+            KindWithContent::NetworkTopologyChange { network: new_network, origin: _ } => {
+                Some(Details::NetworkTopologyChange { network: new_network.clone() })
+            }
         }
     }
 }
@@ -579,6 +597,7 @@ pub enum Kind {
     SnapshotCreation,
     Export,
     UpgradeDatabase,
+    NetworkTopologyChange,
 }
 
 impl Kind {
@@ -597,7 +616,8 @@ impl Kind {
             | Kind::DumpCreation
             | Kind::Export
             | Kind::UpgradeDatabase
-            | Kind::SnapshotCreation => false,
+            | Kind::SnapshotCreation
+            | Kind::NetworkTopologyChange => false,
         }
     }
 }
@@ -618,6 +638,7 @@ impl Display for Kind {
             Kind::SnapshotCreation => write!(f, "snapshotCreation"),
             Kind::Export => write!(f, "export"),
             Kind::UpgradeDatabase => write!(f, "upgradeDatabase"),
+            Kind::NetworkTopologyChange => write!(f, "networkTopologyChange"),
         }
     }
 }
@@ -653,6 +674,8 @@ impl FromStr for Kind {
             Ok(Kind::Export)
         } else if kind.eq_ignore_ascii_case("upgradeDatabase") {
             Ok(Kind::UpgradeDatabase)
+        } else if kind.eq_ignore_ascii_case("networkTopologyChange") {
+            Ok(Kind::NetworkTopologyChange)
         } else {
             Err(ParseTaskKindError(kind.to_owned()))
         }
@@ -738,6 +761,9 @@ pub enum Details {
         from: (u32, u32, u32),
         to: (u32, u32, u32),
     },
+    NetworkTopologyChange {
+        network: Option<Network>,
+    },
 }
 
 #[derive(Debug, PartialEq, Clone, Serialize, Deserialize, ToSchema)]
@@ -805,6 +831,7 @@ impl Details {
             | Self::Dump { .. }
             | Self::Export { .. }
             | Self::UpgradeDatabase { .. }
+            | Self::NetworkTopologyChange { .. }
             | Self::IndexSwap { .. } => (),
         }
 

crates/meilisearch/src/lib.rs

@@ -216,7 +216,10 @@ enum OnFailure {
     KeepDb,
 }
 
-pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<AuthController>)> {
+pub fn setup_meilisearch(
+    opt: &Opt,
+    handle: tokio::runtime::Handle,
+) -> anyhow::Result<(Arc<IndexScheduler>, Arc<AuthController>)> {
     let index_scheduler_opt = IndexSchedulerOptions {
         version_file_path: opt.db_path.join(VERSION_FILE_NAME),
         auth_path: opt.db_path.join("auth"),
@@ -256,6 +259,7 @@ pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<
                     index_scheduler_opt,
                     OnFailure::RemoveDb,
                     binary_version, // the db is empty
+                    handle,
                 )?,
                 Err(e) => {
                     std::fs::remove_dir_all(&opt.db_path)?;
@@ -273,7 +277,7 @@ pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<
             bail!("snapshot doesn't exist at {}", snapshot_path.display())
         // the snapshot and the db exist, and we can ignore the snapshot because of the ignore_snapshot_if_db_exists flag
         } else {
-            open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version)?
+            open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version, handle)?
         }
     } else if let Some(ref path) = opt.import_dump {
         let src_path_exists = path.exists();
@@ -284,6 +288,7 @@ pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<
                 index_scheduler_opt,
                 OnFailure::RemoveDb,
                 binary_version, // the db is empty
+                handle,
             )?;
             match import_dump(&opt.db_path, path, &mut index_scheduler, &mut auth_controller) {
                 Ok(()) => (index_scheduler, auth_controller),
@@ -304,10 +309,10 @@ pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<
         // the dump and the db exist and we can ignore the dump because of the ignore_dump_if_db_exists flag
         // or, the dump is missing but we can ignore that because of the ignore_missing_dump flag
         } else {
-            open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version)?
+            open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version, handle)?
         }
     } else {
-        open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version)?
+        open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version, handle)?
     };
 
     // We create a loop in a thread that registers snapshotCreation tasks
@@ -338,6 +343,7 @@ fn open_or_create_database_unchecked(
     index_scheduler_opt: IndexSchedulerOptions,
     on_failure: OnFailure,
     version: (u32, u32, u32),
+    handle: tokio::runtime::Handle,
 ) -> anyhow::Result<(IndexScheduler, AuthController)> {
     // we don't want to create anything in the data.ms yet, thus we
     // wrap our two builders in a closure that'll be executed later.
@@ -345,7 +351,7 @@ fn open_or_create_database_unchecked(
     let auth_env = open_auth_store_env(&index_scheduler_opt.auth_path).unwrap();
     let auth_controller = AuthController::new(auth_env.clone(), &opt.master_key);
     let index_scheduler_builder = || -> anyhow::Result<_> {
-        Ok(IndexScheduler::new(index_scheduler_opt, auth_env, version)?)
+        Ok(IndexScheduler::new(index_scheduler_opt, auth_env, version, Some(handle))?)
     };
 
     match (
@@ -452,6 +458,7 @@ fn open_or_create_database(
     index_scheduler_opt: IndexSchedulerOptions,
     empty_db: bool,
     binary_version: (u32, u32, u32),
+    handle: tokio::runtime::Handle,
 ) -> anyhow::Result<(IndexScheduler, AuthController)> {
     let version = if !empty_db {
         check_version(opt, &index_scheduler_opt, binary_version)?
@@ -459,7 +466,7 @@ fn open_or_create_database(
         binary_version
     };
 
-    open_or_create_database_unchecked(opt, index_scheduler_opt, OnFailure::KeepDb, version)
+    open_or_create_database_unchecked(opt, index_scheduler_opt, OnFailure::KeepDb, version, handle)
 }
 
 fn import_dump(
@@ -584,7 +591,7 @@ fn import_dump(
             let reader = DocumentsBatchReader::from_reader(reader)?;
 
             let embedder_configs = index.embedding_configs().embedding_configs(&wtxn)?;
-            let embedders = index_scheduler.embedders(uid.to_string(), embedder_configs)?;
+            let embedders = index_scheduler.embedders(&uid, embedder_configs)?;
 
             let builder = milli::update::IndexDocuments::new(
                 &mut wtxn,
@@ -612,7 +619,7 @@ fn import_dump(
 
             let mut indexer = indexer::DocumentOperation::new();
             let embedders = index.embedding_configs().embedding_configs(&rtxn)?;
-            let embedders = index_scheduler.embedders(uid.clone(), embedders)?;
+            let embedders = index_scheduler.embedders(&uid, embedders)?;
 
             let mmap = unsafe { memmap2::Mmap::map(index_reader.documents_file())? };
 

crates/meilisearch/src/main.rs

@@ -76,7 +76,10 @@ fn on_panic(info: &std::panic::PanicHookInfo) {
 
 #[actix_web::main]
 async fn main() -> anyhow::Result<()> {
-    try_main().await.inspect_err(|error| {
+    // won't panic inside of tokio::main
+    let runtime = tokio::runtime::Handle::current();
+
+    try_main(runtime).await.inspect_err(|error| {
         tracing::error!(%error);
         let mut current = error.source();
         let mut depth = 0;
@@ -88,7 +91,7 @@ async fn main() -> anyhow::Result<()> {
     })
 }
 
-async fn try_main() -> anyhow::Result<()> {
+async fn try_main(runtime: tokio::runtime::Handle) -> anyhow::Result<()> {
     let (opt, config_read_from) = Opt::try_build()?;
 
     std::panic::set_hook(Box::new(on_panic));
@@ -122,7 +125,7 @@ async fn try_main() -> anyhow::Result<()> {
         _ => (),
     }
 
-    let (index_scheduler, auth_controller) = setup_meilisearch(&opt)?;
+    let (index_scheduler, auth_controller) = setup_meilisearch(&opt, runtime)?;
 
     let analytics =
         analytics::Analytics::new(&opt, index_scheduler.clone(), auth_controller.clone()).await;

crates/meilisearch/src/routes/chats/chat_completions.rs

@@ -282,8 +282,7 @@ async fn process_search_request(
     if let Some(search_rules) = auth_filter.get_index_search_rules(&index_uid) {
         add_search_rules(&mut query.filter, search_rules);
     }
-    let search_kind =
-        search_kind(&query, index_scheduler.get_ref(), index_uid.to_string(), &index)?;
+    let search_kind = search_kind(&query, index_scheduler.get_ref(), &index_uid, &index)?;
 
     let permit = search_queue.try_get_search_permit().await?;
     let features = index_scheduler.features();
@@ -300,7 +299,7 @@ async fn process_search_request(
         let (search, _is_finite_pagination, _max_total_hits, _offset) =
             prepare_search(&index_cloned, &rtxn, &query, &search_kind, time_budget, features)?;
 
-        match search_from_kind(index_uid, search_kind, search) {
+        match search_from_kind(&index_uid, search_kind, search) {
             Ok((search_results, _)) => Ok((rtxn, Ok(search_results))),
             Err(MeilisearchHttpError::Milli {
                 error: meilisearch_types::milli::Error::UserError(user_error),

crates/meilisearch/src/routes/indexes/documents.rs

@@ -45,7 +45,7 @@ use crate::extractors::authentication::policies::*;
 use crate::extractors::authentication::GuardedData;
 use crate::extractors::payload::Payload;
 use crate::extractors::sequential_extractor::SeqHandler;
-use crate::routes::indexes::enterprise_edition::proxy::{proxy, Body};
+use crate::routes::indexes::enterprise_edition::proxy::{check_leader, proxy, Body};
 use crate::routes::indexes::search::fix_sort_query_parameters;
 use crate::routes::{
     get_task_id, is_dry_run, PaginationView, SummarizedTaskView, PAGINATION_DEFAULT_LIMIT,
@@ -340,6 +340,7 @@ pub async fn delete_document(
     let DocumentParam { index_uid, document_id } = path.into_inner();
     let index_uid = IndexUid::try_from(index_uid)?;
     let network = index_scheduler.network();
+    let origin = check_leader(&req, &network)?;
 
     analytics.publish(
         DocumentsDeletionAggregator {
@@ -363,7 +364,7 @@ pub async fn delete_document(
     };
 
     if network.sharding && !dry_run {
-        proxy(&index_scheduler, &index_uid, &req, network, Body::none(), &task).await?;
+        proxy(&index_scheduler, &index_uid, &req, origin, network, Body::none(), &task).await?;
     }
 
     let task: SummarizedTaskView = task.into();
@@ -946,6 +947,7 @@ async fn document_addition(
 ) -> Result<SummarizedTaskView, MeilisearchHttpError> {
     let mime_type = extract_mime_type(req)?;
     let network = index_scheduler.network();
+    let origin = check_leader(&req, &network)?;
 
     let format = match (
         mime_type.as_ref().map(|m| (m.type_().as_str(), m.subtype().as_str())),
@@ -1081,6 +1083,7 @@ async fn document_addition(
                 &index_scheduler,
                 &index_uid,
                 req,
+                origin,
                 network,
                 Body::with_ndjson_payload(file),
                 &task,
@@ -1168,6 +1171,7 @@ pub async fn delete_documents_batch(
     debug!(parameters = ?body, "Delete documents by batch");
     let index_uid = IndexUid::try_from(index_uid.into_inner())?;
     let network = index_scheduler.network();
+    let origin = check_leader(&req, &network)?;
 
     analytics.publish(
         DocumentsDeletionAggregator {
@@ -1194,7 +1198,8 @@ pub async fn delete_documents_batch(
     };
 
     if network.sharding && !dry_run {
-        proxy(&index_scheduler, &index_uid, &req, network, Body::Inline(body), &task).await?;
+        proxy(&index_scheduler, &index_uid, &req, origin, network, Body::Inline(body), &task)
+            .await?;
     }
 
     let task: SummarizedTaskView = task.into();
@@ -1254,6 +1259,7 @@ pub async fn delete_documents_by_filter(
     let index_uid = index_uid.into_inner();
     let filter = body.into_inner();
     let network = index_scheduler.network();
+    let origin = check_leader(&req, &network)?;
 
     analytics.publish(
         DocumentsDeletionAggregator {
@@ -1286,7 +1292,8 @@ pub async fn delete_documents_by_filter(
     };
 
     if network.sharding && !dry_run {
-        proxy(&index_scheduler, &index_uid, &req, network, Body::Inline(filter), &task).await?;
+        proxy(&index_scheduler, &index_uid, &req, origin, network, Body::Inline(filter), &task)
+            .await?;
     }
 
     let task: SummarizedTaskView = task.into();
@@ -1384,6 +1391,7 @@ pub async fn edit_documents_by_function(
         .check_edit_documents_by_function("Using the documents edit route")?;
 
     let network = index_scheduler.network();
+    let origin = check_leader(&req, &network)?;
 
     let index_uid = IndexUid::try_from(index_uid.into_inner())?;
     let index_uid = index_uid.into_inner();
@@ -1436,7 +1444,8 @@ pub async fn edit_documents_by_function(
     };
 
     if network.sharding && !dry_run {
-        proxy(&index_scheduler, &index_uid, &req, network, Body::Inline(params), &task).await?;
+        proxy(&index_scheduler, &index_uid, &req, origin, network, Body::Inline(params), &task)
+            .await?;
     }
 
     let task: SummarizedTaskView = task.into();
@@ -1483,6 +1492,7 @@ pub async fn clear_all_documents(
 ) -> Result<HttpResponse, ResponseError> {
     let index_uid = IndexUid::try_from(index_uid.into_inner())?;
     let network = index_scheduler.network();
+    let origin = check_leader(&req, &network)?;
 
     analytics.publish(
         DocumentsDeletionAggregator {
@@ -1505,7 +1515,7 @@ pub async fn clear_all_documents(
     };
 
     if network.sharding && !dry_run {
-        proxy(&index_scheduler, &index_uid, &req, network, Body::none(), &task).await?;
+        proxy(&index_scheduler, &index_uid, &req, origin, network, Body::none(), &task).await?;
     }
 
     let task: SummarizedTaskView = task.into();

crates/meilisearch/src/routes/indexes/enterprise_edition/proxy.rs

@@ -38,6 +38,27 @@ impl Body<()> {
     }
 }
 
+pub fn check_leader(
+    req: &HttpRequest,
+    network: &meilisearch_types::enterprise_edition::network::DbNetwork,
+) -> Result<Option<Origin>, MeilisearchHttpError> {
+    match origin_from_req(req)? {
+        Some(origin) => Ok(Some(origin)),
+        None => {
+            let this = network
+                .local
+                .as_deref()
+                .expect("inconsistent `network.sharding` and `network.self`");
+
+            let is_leader = this == todo!();
+            if !is_leader {
+                return Err(MeilisearchHttpError::NotLeader { leader: todo!() });
+            }
+            Ok(None)
+        }
+    }
+}
+
 /// If necessary, proxies the passed request to the network and update the task description.
 ///
 /// This function reads the custom headers from the request to determine if must proxy the request or if the request
@@ -52,11 +73,12 @@ pub async fn proxy<T: serde::Serialize>(
     index_scheduler: &IndexScheduler,
     index_uid: &str,
     req: &HttpRequest,
-    network: meilisearch_types::enterprise_edition::network::Network,
+    origin: Option<Origin>,
+    network: meilisearch_types::enterprise_edition::network::DbNetwork,
     body: Body<T>,
     task: &meilisearch_types::tasks::Task,
 ) -> Result<(), MeilisearchHttpError> {
-    match origin_from_req(req)? {
+    match origin {
         Some(origin) => {
             index_scheduler.set_task_network(task.uid, TaskNetwork::Origin { origin })?
         }

crates/meilisearch/src/routes/indexes/facet_search.rs

@@ -260,7 +260,7 @@ pub async fn search(
     }
 
     let index = index_scheduler.index(&index_uid)?;
-    let search_kind = search_kind(&search_query, &index_scheduler, index_uid.to_string(), &index)?;
+    let search_kind = search_kind(&search_query, &index_scheduler, &index_uid, &index)?;
     let permit = search_queue.try_get_search_permit().await?;
     let search_result = tokio::task::spawn_blocking(move || {
         perform_facet_search(

crates/meilisearch/src/routes/indexes/search.rs

@@ -339,13 +339,12 @@ pub async fn search_with_url_query(
 
     let index = index_scheduler.index(&index_uid)?;
 
-    let search_kind =
-        search_kind(&query, index_scheduler.get_ref(), index_uid.to_string(), &index)?;
+    let search_kind = search_kind(&query, index_scheduler.get_ref(), &index_uid, &index)?;
     let retrieve_vector = RetrieveVectors::new(query.retrieve_vectors);
     let permit = search_queue.try_get_search_permit().await?;
     let search_result = tokio::task::spawn_blocking(move || {
         perform_search(
-            index_uid.to_string(),
+            &index_uid,
             &index,
             query,
             search_kind,
@@ -445,14 +444,13 @@ pub async fn search_with_post(
 
     let index = index_scheduler.index(&index_uid)?;
 
-    let search_kind =
-        search_kind(&query, index_scheduler.get_ref(), index_uid.to_string(), &index)?;
+    let search_kind = search_kind(&query, index_scheduler.get_ref(), &index_uid, &index)?;
     let retrieve_vectors = RetrieveVectors::new(query.retrieve_vectors);
 
     let permit = search_queue.try_get_search_permit().await?;
     let search_result = tokio::task::spawn_blocking(move || {
         perform_search(
-            index_uid.to_string(),
+            &index_uid,
             &index,
             query,
             search_kind,
@@ -480,7 +478,7 @@ pub async fn search_with_post(
 pub fn search_kind(
     query: &SearchQuery,
     index_scheduler: &IndexScheduler,
-    index_uid: String,
+    index_uid: &str,
     index: &milli::Index,
 ) -> Result<SearchKind, ResponseError> {
     let is_placeholder_query =

crates/meilisearch/src/routes/indexes/similar.rs

@@ -227,7 +227,7 @@ async fn similar(
 
     let (embedder_name, embedder, quantized) = SearchKind::embedder(
         &index_scheduler,
-        index_uid.to_string(),
+        index_uid.as_str(),
         &index,
         &query.embedder,
         None,

crates/meilisearch/src/routes/mod.rs

@@ -39,7 +39,6 @@ use crate::routes::features::RuntimeTogglableFeatures;
 use crate::routes::indexes::documents::{DocumentDeletionByFilter, DocumentEditionByFunction};
 use crate::routes::indexes::IndexView;
 use crate::routes::multi_search::SearchResults;
-use crate::routes::network::{Network, Remote};
 use crate::routes::swap_indexes::SwapIndexesPayload;
 use crate::routes::webhooks::{WebhookResults, WebhookSettings, WebhookWithMetadata};
 use crate::search::{
@@ -102,7 +101,7 @@ mod webhooks;
         url = "/",
         description = "Local server",
     )),
-    components(schemas(PaginationView<KeyView>, PaginationView<IndexView>, IndexView, DocumentDeletionByFilter, AllBatches, BatchStats, ProgressStepView, ProgressView, BatchView, RuntimeTogglableFeatures, SwapIndexesPayload, DocumentEditionByFunction, MergeFacets, FederationOptions, SearchQueryWithIndex, Federation, FederatedSearch, FederatedSearchResult, SearchResults, SearchResultWithIndex, SimilarQuery, SimilarResult, PaginationView<serde_json::Value>, BrowseQuery, UpdateIndexRequest, IndexUid, IndexCreateRequest, KeyView, Action, CreateApiKey, UpdateStderrLogs, LogMode, GetLogs, IndexStats, Stats, HealthStatus, HealthResponse, VersionResponse, Code, ErrorType, AllTasks, TaskView, Status, DetailsView, ResponseError, Settings<Unchecked>, Settings<Checked>, TypoSettings, MinWordSizeTyposSetting, FacetingSettings, PaginationSettings, SummarizedTaskView, Kind, Network, Remote, FilterableAttributesRule, FilterableAttributesPatterns, AttributePatterns, FilterableAttributesFeatures, FilterFeatures, Export, WebhookSettings, WebhookResults, WebhookWithMetadata, meilisearch_types::milli::vector::VectorStoreBackend))
+    components(schemas(PaginationView<KeyView>, PaginationView<IndexView>, IndexView, DocumentDeletionByFilter, AllBatches, BatchStats, ProgressStepView, ProgressView, BatchView, RuntimeTogglableFeatures, SwapIndexesPayload, DocumentEditionByFunction, MergeFacets, FederationOptions, SearchQueryWithIndex, Federation, FederatedSearch, FederatedSearchResult, SearchResults, SearchResultWithIndex, SimilarQuery, SimilarResult, PaginationView<serde_json::Value>, BrowseQuery, UpdateIndexRequest, IndexUid, IndexCreateRequest, KeyView, Action, CreateApiKey, UpdateStderrLogs, LogMode, GetLogs, IndexStats, Stats, HealthStatus, HealthResponse, VersionResponse, Code, ErrorType, AllTasks, TaskView, Status, DetailsView, ResponseError, Settings<Unchecked>, Settings<Checked>, TypoSettings, MinWordSizeTyposSetting, FacetingSettings, PaginationSettings, SummarizedTaskView, Kind, meilisearch_types::enterprise_edition::network::Network, meilisearch_types::enterprise_edition::network::Remote, FilterableAttributesRule, FilterableAttributesPatterns, AttributePatterns, FilterableAttributesFeatures, FilterFeatures, Export, WebhookSettings, WebhookResults, WebhookWithMetadata, meilisearch_types::milli::vector::VectorStoreBackend))
 )]
 pub struct MeilisearchApi;
 

crates/meilisearch/src/routes/multi_search.rs

@@ -239,26 +239,23 @@ pub async fn multi_search_with_post(
                         })
                         .with_index(query_index)?;
 
-                    let index_uid_str = index_uid.to_string();
-
-                    let search_kind = search_kind(
-                        &query,
-                        index_scheduler.get_ref(),
-                        index_uid_str.clone(),
-                        &index,
-                    )
-                    .with_index(query_index)?;
+                    let search_kind =
+                        search_kind(&query, index_scheduler.get_ref(), &index_uid, &index)
+                            .with_index(query_index)?;
                     let retrieve_vector = RetrieveVectors::new(query.retrieve_vectors);
 
-                    let search_result = tokio::task::spawn_blocking(move || {
-                        perform_search(
-                            index_uid_str.clone(),
-                            &index,
-                            query,
-                            search_kind,
-                            retrieve_vector,
-                            features,
-                        )
+                    let search_result = tokio::task::spawn_blocking({
+                        let index_uid = index_uid.clone();
+                        move || {
+                            perform_search(
+                                &index_uid,
+                                &index,
+                                query,
+                                search_kind,
+                                retrieve_vector,
+                                features,
+                            )
+                        }
                     })
                     .await
                     .with_index(query_index)?;

crates/meilisearch/src/routes/network.rs

@@ -1,28 +1,21 @@
-use std::collections::BTreeMap;
-
 use actix_web::web::{self, Data};
 use actix_web::{HttpRequest, HttpResponse};
 use deserr::actix_web::AwebJson;
-use deserr::Deserr;
 use index_scheduler::IndexScheduler;
-use itertools::{EitherOrBoth, Itertools};
 use meilisearch_types::deserr::DeserrJsonError;
-use meilisearch_types::enterprise_edition::network::{Network as DbNetwork, Remote as DbRemote};
-use meilisearch_types::error::deserr_codes::{
-    InvalidNetworkRemotes, InvalidNetworkSearchApiKey, InvalidNetworkSelf, InvalidNetworkSharding,
-    InvalidNetworkUrl, InvalidNetworkWriteApiKey,
-};
+use meilisearch_types::enterprise_edition::network::{Network, Remote};
 use meilisearch_types::error::ResponseError;
 use meilisearch_types::keys::actions;
 use meilisearch_types::milli::update::Setting;
 use serde::Serialize;
 use tracing::debug;
-use utoipa::{OpenApi, ToSchema};
+use utoipa::OpenApi;
 
 use crate::analytics::{Aggregate, Analytics};
 use crate::extractors::authentication::policies::ActionPolicy;
 use crate::extractors::authentication::GuardedData;
 use crate::extractors::sequential_extractor::SeqHandler;
+use crate::routes::SummarizedTaskView;
 
 #[derive(OpenApi)]
 #[openapi(
@@ -31,7 +24,7 @@ use crate::extractors::sequential_extractor::SeqHandler;
         name = "Network",
         description = "The `/network` route allows you to describe the topology of a network of Meilisearch instances.
 
-This route is **synchronous**. This means that no task object will be returned, and any change to the network will be made available immediately.",
+This route is **asynchronous**. A task uid will be returned, and any change to the network will be effective after the corresponding task has been processed.",
         external_docs(url = "https://www.meilisearch.com/docs/reference/api/network"),
     )),
 )]
@@ -83,73 +76,6 @@ async fn get_network(
     Ok(HttpResponse::Ok().json(network))
 }
 
-#[derive(Debug, Deserr, ToSchema, Serialize)]
-#[deserr(error = DeserrJsonError<InvalidNetworkRemotes>, rename_all = camelCase, deny_unknown_fields)]
-#[serde(rename_all = "camelCase")]
-#[schema(rename_all = "camelCase")]
-pub struct Remote {
-    #[schema(value_type = Option<String>, example = json!({
-        "ms-0": Remote { url: Setting::Set("http://localhost:7700".into()), search_api_key: Setting::Reset, write_api_key: Setting::Reset },
-        "ms-1": Remote { url: Setting::Set("http://localhost:7701".into()), search_api_key: Setting::Set("foo".into()), write_api_key: Setting::Set("bar".into()) },
-        "ms-2": Remote { url: Setting::Set("http://localhost:7702".into()), search_api_key: Setting::Set("bar".into()), write_api_key: Setting::Set("foo".into()) },
-    }))]
-    #[deserr(default, error = DeserrJsonError<InvalidNetworkUrl>)]
-    #[serde(default)]
-    pub url: Setting<String>,
-    #[schema(value_type = Option<String>, example = json!("XWnBI8QHUc-4IlqbKPLUDuhftNq19mQtjc6JvmivzJU"))]
-    #[deserr(default, error = DeserrJsonError<InvalidNetworkSearchApiKey>)]
-    #[serde(default)]
-    pub search_api_key: Setting<String>,
-    #[schema(value_type = Option<String>, example = json!("XWnBI8QHUc-4IlqbKPLUDuhftNq19mQtjc6JvmivzJU"))]
-    #[deserr(default, error = DeserrJsonError<InvalidNetworkWriteApiKey>)]
-    #[serde(default)]
-    pub write_api_key: Setting<String>,
-}
-
-#[derive(Debug, Deserr, ToSchema, Serialize)]
-#[deserr(error = DeserrJsonError, rename_all = camelCase, deny_unknown_fields)]
-#[serde(rename_all = "camelCase")]
-#[schema(rename_all = "camelCase")]
-pub struct Network {
-    #[schema(value_type = Option<BTreeMap<String, Remote>>, example = json!("http://localhost:7700"))]
-    #[deserr(default, error = DeserrJsonError<InvalidNetworkRemotes>)]
-    #[serde(default)]
-    pub remotes: Setting<BTreeMap<String, Option<Remote>>>,
-    #[schema(value_type = Option<String>, example = json!("ms-00"), rename = "self")]
-    #[serde(default, rename = "self")]
-    #[deserr(default, rename = "self", error = DeserrJsonError<InvalidNetworkSelf>)]
-    pub local: Setting<String>,
-    #[schema(value_type = Option<bool>, example = json!(true))]
-    #[serde(default)]
-    #[deserr(default, error = DeserrJsonError<InvalidNetworkSharding>)]
-    pub sharding: Setting<bool>,
-}
-
-impl Remote {
-    pub fn try_into_db_node(self, name: &str) -> Result<DbRemote, ResponseError> {
-        Ok(DbRemote {
-            url: self
-                .url
-                .set()
-                .ok_or(ResponseError::from_msg(
-                    format!("Missing field `.remotes.{name}.url`"),
-                    meilisearch_types::error::Code::MissingNetworkUrl,
-                ))
-                .and_then(|url| {
-                    if let Err(error) = url::Url::parse(&url) {
-                        return Err(ResponseError::from_msg(
-                            format!("Invalid `.remotes.{name}.url` (`{url}`): {error}"),
-                            meilisearch_types::error::Code::InvalidNetworkUrl,
-                        ));
-                    }
-                    Ok(url)
-                })?,
-            search_api_key: self.search_api_key.set(),
-            write_api_key: self.write_api_key.set(),
-        })
-    }
-}
-
 #[derive(Serialize)]
 pub struct PatchNetworkAnalytics {
     network_size: usize,
@@ -208,111 +134,58 @@ async fn patch_network(
     index_scheduler.features().check_network("Using the /network route")?;
 
     let new_network = new_network.0;
-    let old_network = index_scheduler.network();
     debug!(parameters = ?new_network, "Patch network");
 
-    let merged_self = match new_network.local {
-        Setting::Set(new_self) => Some(new_self),
-        Setting::Reset => None,
-        Setting::NotSet => old_network.local,
-    };
-
-    let merged_sharding = match new_network.sharding {
-        Setting::Set(new_sharding) => new_sharding,
-        Setting::Reset => false,
-        Setting::NotSet => old_network.sharding,
-    };
-
-    if merged_sharding && merged_self.is_none() {
-        return Err(ResponseError::from_msg(
-            "`.sharding`: enabling the sharding requires `.self` to be set\n  - Hint: Disable `sharding` or set `self` to a value.".into(),
-            meilisearch_types::error::Code::InvalidNetworkSharding,
-        ));
-    }
-
-    let merged_remotes = match new_network.remotes {
-        Setting::Set(new_remotes) => {
-            let mut merged_remotes = BTreeMap::new();
-            for either_or_both in old_network
-                .remotes
-                .into_iter()
-                .merge_join_by(new_remotes.into_iter(), |left, right| left.0.cmp(&right.0))
-            {
-                match either_or_both {
-                    EitherOrBoth::Both((key, old), (_, Some(new))) => {
-                        let DbRemote {
-                            url: old_url,
-                            search_api_key: old_search_api_key,
-                            write_api_key: old_write_api_key,
-                        } = old;
-
-                        let Remote {
-                            url: new_url,
-                            search_api_key: new_search_api_key,
-                            write_api_key: new_write_api_key,
-                        } = new;
-
-                        let merged = DbRemote {
-                            url: match new_url {
-                                Setting::Set(new_url) => {
-                                    if let Err(error) = url::Url::parse(&new_url) {
-                                        return Err(ResponseError::from_msg(
-                                            format!("Invalid `.remotes.{key}.url` (`{new_url}`): {error}"),
-                                            meilisearch_types::error::Code::InvalidNetworkUrl,
-                                        ));
-                                    }
-                                    new_url
-                                }
-                                Setting::Reset => {
-                                    return Err(ResponseError::from_msg(
-                                        format!(
-                                            "Field `.remotes.{key}.url` cannot be set to `null`"
-                                        ),
-                                        meilisearch_types::error::Code::InvalidNetworkUrl,
-                                    ))
-                                }
-                                Setting::NotSet => old_url,
-                            },
-                            search_api_key: match new_search_api_key {
-                                Setting::Set(new_search_api_key) => Some(new_search_api_key),
-                                Setting::Reset => None,
-                                Setting::NotSet => old_search_api_key,
-                            },
-                            write_api_key: match new_write_api_key {
-                                Setting::Set(new_write_api_key) => Some(new_write_api_key),
-                                Setting::Reset => None,
-                                Setting::NotSet => old_write_api_key,
-                            },
-                        };
-                        merged_remotes.insert(key, merged);
-                    }
-                    EitherOrBoth::Both((_, _), (_, None)) | EitherOrBoth::Right((_, None)) => {}
-                    EitherOrBoth::Left((key, node)) => {
-                        merged_remotes.insert(key, node);
-                    }
-                    EitherOrBoth::Right((key, Some(node))) => {
-                        let node = node.try_into_db_node(&key)?;
-                        merged_remotes.insert(key, node);
+    // check the URLs of all remotes
+    if let Setting::Set(remotes) = &new_network.remotes {
+        for (remote_name, remote) in remotes.iter() {
+            let Some(remote) = remote else {
+                continue;
+            };
+            match &remote.url {
+                Setting::Set(new_url) => {
+                    if let Err(error) = url::Url::parse(&new_url) {
+                        return Err(ResponseError::from_msg(
+                            format!("Invalid `.remotes.{remote_name}.url` (`{new_url}`): {error}"),
+                            meilisearch_types::error::Code::InvalidNetworkUrl,
+                        ));
                     }
                 }
+                Setting::Reset => {
+                    return Err(ResponseError::from_msg(
+                        format!("Field `.remotes.{remote_name}.url` cannot be set to `null`"),
+                        meilisearch_types::error::Code::InvalidNetworkUrl,
+                    ))
+                }
+                Setting::NotSet => (),
             }
-            merged_remotes
         }
-        Setting::Reset => BTreeMap::new(),
-        Setting::NotSet => old_network.remotes,
-    };
+    }
 
     analytics.publish(
         PatchNetworkAnalytics {
-            network_size: merged_remotes.len(),
-            network_has_self: merged_self.is_some(),
+            network_size: new_network
+                .remotes
+                .as_ref()
+                .set()
+                .map(|remotes| remotes.len())
+                .unwrap_or_default(),
+            network_has_self: new_network.local.as_ref().set().is_some(),
         },
         &req,
     );
 
-    let merged_network =
-        DbNetwork { local: merged_self, remotes: merged_remotes, sharding: merged_sharding };
-    index_scheduler.put_network(merged_network.clone())?;
-    debug!(returns = ?merged_network, "Patch network");
-    Ok(HttpResponse::Ok().json(merged_network))
+    let task = index_scheduler.register(
+        meilisearch_types::tasks::KindWithContent::NetworkTopologyChange {
+            network: Some(new_network),
+            origin: None,
+        },
+        None,
+        false,
+    )?;
+    debug!(returns = ?task, "Patch network");
+
+    let task: SummarizedTaskView = task.into();
+
+    return Ok(HttpResponse::Accepted().json(task));
 }

crates/meilisearch/src/routes/tasks_test.rs

@@ -226,14 +226,14 @@ mod tests {
         {
             let params = "types=createIndex";
             let err = deserr_query_params::<TaskDeletionOrCancelationQuery>(params).unwrap_err();
-            snapshot!(meili_snap::json_string!(err), @r#"
+            snapshot!(meili_snap::json_string!(err), @r###"
             {
-              "message": "Invalid value in parameter `types`: `createIndex` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`.",
+              "message": "Invalid value in parameter `types`: `createIndex` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`, `networkTopologyChange`.",
               "code": "invalid_task_types",
               "type": "invalid_request",
               "link": "https://docs.meilisearch.com/errors#invalid_task_types"
             }
-            "#);
+            "###);
         }
     }
     #[test]

crates/meilisearch/src/routes/webhooks.rs

@@ -1,3 +1,4 @@
+use core::convert::Infallible;
 use std::collections::BTreeMap;
 use std::str::FromStr;
 
@@ -7,7 +8,6 @@ use actix_http::header::{
 };
 use actix_web::web::{self, Data, Path};
 use actix_web::{HttpRequest, HttpResponse};
-use core::convert::Infallible;
 use deserr::actix_web::AwebJson;
 use deserr::{DeserializeError, Deserr, ValuePointerRef};
 use index_scheduler::IndexScheduler;
@@ -24,12 +24,12 @@ use tracing::debug;
 use url::Url;
 use utoipa::{OpenApi, ToSchema};
 use uuid::Uuid;
+use WebhooksError::*;
 
 use crate::analytics::{Aggregate, Analytics};
 use crate::extractors::authentication::policies::ActionPolicy;
 use crate::extractors::authentication::GuardedData;
 use crate::extractors::sequential_extractor::SeqHandler;
-use WebhooksError::*;
 
 #[derive(OpenApi)]
 #[openapi(

crates/meilisearch/src/search/federated/perform.rs

@@ -9,7 +9,7 @@ use std::vec::{IntoIter, Vec};
 use actix_http::StatusCode;
 use index_scheduler::{IndexScheduler, RoFeatures};
 use itertools::Itertools;
-use meilisearch_types::enterprise_edition::network::{Network, Remote};
+use meilisearch_types::enterprise_edition::network::{DbNetwork, DbRemote};
 use meilisearch_types::error::ResponseError;
 use meilisearch_types::milli::order_by_map::OrderByMap;
 use meilisearch_types::milli::score_details::{ScoreDetails, WeightedScoreValue};
@@ -456,7 +456,7 @@ fn merge_metadata(
 }
 
 type LocalQueriesByIndex = BTreeMap<String, Vec<QueryByIndex>>;
-type RemoteQueriesByHost = BTreeMap<String, (Remote, Vec<SearchQueryWithIndex>)>;
+type RemoteQueriesByHost = BTreeMap<String, (DbRemote, Vec<SearchQueryWithIndex>)>;
 
 struct PartitionedQueries {
     local_queries_by_index: LocalQueriesByIndex,
@@ -477,7 +477,7 @@ impl PartitionedQueries {
         &mut self,
         federated_query: SearchQueryWithIndex,
         query_index: usize,
-        network: &Network,
+        network: &DbNetwork,
         features: RoFeatures,
     ) -> Result<(), ResponseError> {
         if let Some(pagination_field) = federated_query.has_pagination() {
@@ -672,7 +672,7 @@ struct SearchByIndexParams<'a> {
     features: RoFeatures,
     is_proxy: bool,
     has_remote: bool,
-    network: &'a Network,
+    network: &'a DbNetwork,
 }
 
 struct SearchByIndex {
@@ -755,8 +755,7 @@ impl SearchByIndex {
             // use an immediately invoked lambda to capture the result without returning from the function
 
             let res: Result<(), ResponseError> = (|| {
-                let search_kind =
-                    search_kind(&query, params.index_scheduler, index_uid.to_string(), &index)?;
+                let search_kind = search_kind(&query, params.index_scheduler, &index_uid, &index)?;
 
                 let canonicalization_kind = match (&search_kind, &query.q) {
                     (SearchKind::SemanticOnly { .. }, _) => {
@@ -806,11 +805,11 @@ impl SearchByIndex {
                     {
                         Some((previous_ranking_rules, previous_query_index, previous_index_uid))
                     } else {
-                        Some((ranking_rules, query_index, index_uid.clone()))
+                        Some((ranking_rules, query_index, index_uid.to_string()))
                     };
                 } else {
                     self.previous_query_data =
-                        Some((ranking_rules, query_index, index_uid.clone()));
+                        Some((ranking_rules, query_index, index_uid.to_string()));
                 }
 
                 match search_kind {
@@ -839,7 +838,7 @@ impl SearchByIndex {
                 search.limit(params.required_hit_count);
 
                 let (result, _semantic_hit_count) =
-                    super::super::search_from_kind(index_uid.to_string(), search_kind, search)?;
+                    super::super::search_from_kind(&index_uid, search_kind, search)?;
                 let format = AttributesFormat {
                     attributes_to_retrieve: query.attributes_to_retrieve,
                     retrieve_vectors,

crates/meilisearch/src/search/federated/proxy.rs

@@ -1,6 +1,6 @@
 pub use error::ProxySearchError;
 use error::ReqwestErrorWithoutUrl;
-use meilisearch_types::enterprise_edition::network::Remote;
+use meilisearch_types::enterprise_edition::network::DbRemote;
 use rand::Rng as _;
 use reqwest::{Client, Response, StatusCode};
 use serde::de::DeserializeOwned;
@@ -94,7 +94,7 @@ pub struct ProxySearchParams {
 
 /// Performs a federated search on a remote host and returns the results
 pub async fn proxy_search(
-    node: &Remote,
+    node: &DbRemote,
     queries: Vec<SearchQueryWithIndex>,
     federation: Federation,
     params: &ProxySearchParams,

crates/meilisearch/src/search/mod.rs

@@ -362,7 +362,7 @@ pub enum SearchKind {
 impl SearchKind {
     pub(crate) fn semantic(
         index_scheduler: &index_scheduler::IndexScheduler,
-        index_uid: String,
+        index_uid: &str,
         index: &Index,
         embedder_name: &str,
         vector_len: Option<usize>,
@@ -380,7 +380,7 @@ impl SearchKind {
 
     pub(crate) fn hybrid(
         index_scheduler: &index_scheduler::IndexScheduler,
-        index_uid: String,
+        index_uid: &str,
         index: &Index,
         embedder_name: &str,
         semantic_ratio: f32,
@@ -399,7 +399,7 @@ impl SearchKind {
 
     pub(crate) fn embedder(
         index_scheduler: &index_scheduler::IndexScheduler,
-        index_uid: String,
+        index_uid: &str,
         index: &Index,
         embedder_name: &str,
         vector_len: Option<usize>,
@@ -1114,7 +1114,7 @@ pub fn prepare_search<'t>(
 }
 
 pub fn perform_search(
-    index_uid: String,
+    index_uid: &str,
     index: &Index,
     query: SearchQuery,
     search_kind: SearchKind,
@@ -1299,27 +1299,24 @@ fn compute_facet_distribution_stats<S: AsRef<str>>(
 }
 
 pub fn search_from_kind(
-    index_uid: String,
+    index_uid: &str,
     search_kind: SearchKind,
     search: milli::Search<'_>,
 ) -> Result<(milli::SearchResult, Option<u32>), MeilisearchHttpError> {
+    let err = |e| MeilisearchHttpError::from_milli(e, Some(index_uid.to_string()));
     let (milli_result, semantic_hit_count) = match &search_kind {
         SearchKind::KeywordOnly => {
-            let results = search
-                .execute()
-                .map_err(|e| MeilisearchHttpError::from_milli(e, Some(index_uid.to_string())))?;
+            let results = search.execute().map_err(err)?;
             (results, None)
         }
         SearchKind::SemanticOnly { .. } => {
-            let results = search
-                .execute()
-                .map_err(|e| MeilisearchHttpError::from_milli(e, Some(index_uid.to_string())))?;
+            let results = search.execute().map_err(err)?;
             let semantic_hit_count = results.document_scores.len() as u32;
             (results, Some(semantic_hit_count))
         }
-        SearchKind::Hybrid { semantic_ratio, .. } => search
-            .execute_hybrid(*semantic_ratio)
-            .map_err(|e| MeilisearchHttpError::from_milli(e, Some(index_uid)))?,
+        SearchKind::Hybrid { semantic_ratio, .. } => {
+            search.execute_hybrid(*semantic_ratio).map_err(err)?
+        }
     };
     Ok((milli_result, semantic_hit_count))
 }

crates/meilisearch/tests/batches/errors.rs

@@ -40,14 +40,14 @@ async fn batch_bad_types() {
 
     let (response, code) = server.batches_filter("types=doggo").await;
     snapshot!(code, @"400 Bad Request");
-    snapshot!(json_string!(response), @r#"
+    snapshot!(json_string!(response), @r###"
     {
-      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`.",
+      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`, `networkTopologyChange`.",
       "code": "invalid_task_types",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_task_types"
     }
-    "#);
+    "###);
 }
 
 #[actix_rt::test]

crates/meilisearch/tests/common/mod.rs

@@ -522,6 +522,26 @@ pub async fn shared_index_with_geo_documents() -> &'static Index<'static, Shared
         .await
 }
 
+pub async fn shared_index_geojson_documents() -> &'static Index<'static, Shared> {
+    static INDEX: OnceCell<Index<'static, Shared>> = OnceCell::const_new();
+    INDEX
+        .get_or_init(|| async {
+            // Retrieved from https://gitlab-forge.din.developpement-durable.gouv.fr/pub/geomatique/descartes/d-map/-/blob/main/demo/examples/commons/countries.geojson?ref_type=heads
+            let server = Server::new_shared();
+            let index = server._index("SHARED_GEOJSON_DOCUMENTS").to_shared();
+            let countries = include_str!("../documents/geojson/assets/countries.json");
+            let lille = serde_json::from_str::<serde_json::Value>(countries).unwrap();
+            let (response, _code) = index._add_documents(Value(lille), Some("name")).await;
+            server.wait_task(response.uid()).await.succeeded();
+
+            let (response, _code) =
+                index._update_settings(json!({"filterableAttributes": ["_geojson"]})).await;
+            server.wait_task(response.uid()).await.succeeded();
+            index
+        })
+        .await
+}
+
 pub async fn shared_index_for_fragments() -> Index<'static, Shared> {
     static INDEX: OnceCell<(Server<Shared>, String)> = OnceCell::const_new();
     let (server, uid) = INDEX

crates/meilisearch/tests/common/server.rs

@@ -49,8 +49,8 @@ impl Server<Owned> {
         }
 
         let options = default_settings(dir.path());
-
-        let (index_scheduler, auth) = setup_meilisearch(&options).unwrap();
+        let handle = tokio::runtime::Handle::current();
+        let (index_scheduler, auth) = setup_meilisearch(&options, handle).unwrap();
         let service = Service { index_scheduler, auth, options, api_key: None };
 
         Server { service, _dir: Some(dir), _marker: PhantomData }
@@ -65,7 +65,9 @@ impl Server<Owned> {
 
         options.master_key = Some("MASTER_KEY".to_string());
 
-        let (index_scheduler, auth) = setup_meilisearch(&options).unwrap();
+        let handle = tokio::runtime::Handle::current();
+
+        let (index_scheduler, auth) = setup_meilisearch(&options, handle).unwrap();
         let service = Service { index_scheduler, auth, options, api_key: None };
 
         Server { service, _dir: Some(dir), _marker: PhantomData }
@@ -78,7 +80,9 @@ impl Server<Owned> {
     }
 
     pub async fn new_with_options(options: Opt) -> Result<Self, anyhow::Error> {
-        let (index_scheduler, auth) = setup_meilisearch(&options)?;
+        let handle = tokio::runtime::Handle::current();
+
+        let (index_scheduler, auth) = setup_meilisearch(&options, handle)?;
         let service = Service { index_scheduler, auth, options, api_key: None };
 
         Ok(Server { service, _dir: None, _marker: PhantomData })
@@ -217,8 +221,9 @@ impl Server<Shared> {
         }
 
         let options = default_settings(dir.path());
+        let handle = tokio::runtime::Handle::current();
 
-        let (index_scheduler, auth) = setup_meilisearch(&options).unwrap();
+        let (index_scheduler, auth) = setup_meilisearch(&options, handle).unwrap();
         let service = Service { index_scheduler, auth, api_key: None, options };
 
         Server { service, _dir: Some(dir), _marker: PhantomData }

crates/meilisearch/tests/documents/add_documents.rs

@@ -1,6 +1,3 @@
-use crate::common::encoder::Encoder;
-use crate::common::{default_settings, GetAllDocumentsOptions, Server, Value};
-use crate::json;
 use actix_web::test;
 use meili_snap::{json_string, snapshot};
 use meilisearch::Opt;
@@ -8,6 +5,10 @@ use time::format_description::well_known::Rfc3339;
 use time::OffsetDateTime;
 use uuid::Uuid;
 
+use crate::common::encoder::Encoder;
+use crate::common::{default_settings, GetAllDocumentsOptions, Server, Value};
+use crate::json;
+
 /// This is the basic usage of our API and every other tests uses the content-type application/json
 #[actix_rt::test]
 async fn add_documents_test_json_content_types() {

crates/meilisearch/tests/documents/errors.rs

@@ -134,14 +134,14 @@ async fn get_all_documents_bad_filter() {
 
     let (response, code) = index.get_all_documents_raw("?filter=doggo").await;
     snapshot!(code, @"400 Bad Request");
-    snapshot!(json_string!(response), @r###"
+    snapshot!(json_string!(response), @r#"
     {
-      "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `doggo`.\n1:6 doggo",
+      "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `doggo`.\n1:6 doggo",
       "code": "invalid_document_filter",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_document_filter"
     }
-    "###);
+    "#);
 
     let (response, code) = index.get_all_documents_raw("?filter=doggo=bernese").await;
     snapshot!(code, @"400 Bad Request");
@@ -523,14 +523,14 @@ async fn delete_document_by_filter() {
     // send bad filter
     let (response, code) = index.delete_document_by_filter(json!({ "filter": "hello"})).await;
     snapshot!(code, @"400 Bad Request");
-    snapshot!(response, @r###"
+    snapshot!(response, @r#"
     {
-      "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `hello`.\n1:6 hello",
+      "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `hello`.\n1:6 hello",
       "code": "invalid_document_filter",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_document_filter"
     }
-    "###);
+    "#);
 
     // send empty filter
     let (response, code) = index.delete_document_by_filter(json!({ "filter": ""})).await;
@@ -724,14 +724,14 @@ async fn fetch_document_by_filter() {
 
     let (response, code) = index.fetch_documents(json!({ "filter": "cool doggo" })).await;
     snapshot!(code, @"400 Bad Request");
-    snapshot!(response, @r###"
+    snapshot!(response, @r#"
     {
-      "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `cool doggo`.\n1:11 cool doggo",
+      "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `cool doggo`.\n1:11 cool doggo",
       "code": "invalid_document_filter",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_document_filter"
     }
-    "###);
+    "#);
 
     let (response, code) = index.fetch_documents(json!({ "filter": "doggo = bernese" })).await;
     snapshot!(code, @"400 Bad Request");

crates/meilisearch/tests/documents/geojson/assets/lille.geojson

@@ -0,0 +1,547 @@
+{
+    "type": "Polygon",
+    "coordinates": [
+      [
+        [
+          3.11681,
+          50.63646
+        ],
+        [
+          3.11945,
+          50.63488
+        ],
+        [
+          3.12134,
+          50.63504
+        ],
+        [
+          3.12064,
+          50.63127
+        ],
+        [
+          3.12203,
+          50.62785
+        ],
+        [
+          3.12389,
+          50.6262
+        ],
+        [
+          3.12161,
+          50.62358
+        ],
+        [
+          3.12547,
+          50.62114
+        ],
+        [
+          3.12447,
+          50.61874
+        ],
+        [
+          3.12288,
+          50.61988
+        ],
+        [
+          3.12054,
+          50.61846
+        ],
+        [
+          3.11846,
+          50.61754
+        ],
+        [
+          3.11482,
+          50.6207
+        ],
+        [
+          3.11232,
+          50.6188
+        ],
+        [
+          3.10936,
+          50.61727
+        ],
+        [
+          3.10822,
+          50.61765
+        ],
+        [
+          3.10603,
+          50.61536
+        ],
+        [
+          3.1041,
+          50.61596
+        ],
+        [
+          3.10017,
+          50.6186
+        ],
+        [
+          3.09688,
+          50.61714
+        ],
+        [
+          3.09575,
+          50.61795
+        ],
+        [
+          3.0891,
+          50.61532
+        ],
+        [
+          3.08625,
+          50.61792
+        ],
+        [
+          3.07948,
+          50.61428
+        ],
+        [
+          3.07146,
+          50.6066
+        ],
+        [
+          3.06819,
+          50.60918
+        ],
+        [
+          3.06502,
+          50.61046
+        ],
+        [
+          3.06223,
+          50.61223
+        ],
+        [
+          3.05925,
+          50.60659
+        ],
+        [
+          3.05463,
+          50.60077
+        ],
+        [
+          3.04906,
+          50.6008
+        ],
+        [
+          3.04726,
+          50.6035
+        ],
+        [
+          3.04328,
+          50.60667
+        ],
+        [
+          3.04155,
+          50.60417
+        ],
+        [
+          3.03767,
+          50.60456
+        ],
+        [
+          3.03528,
+          50.60538
+        ],
+        [
+          3.03239,
+          50.60725
+        ],
+        [
+          3.0254,
+          50.6111
+        ],
+        [
+          3.02387,
+          50.6125
+        ],
+        [
+          3.0248,
+          50.61344
+        ],
+        [
+          3.02779,
+          50.61418
+        ],
+        [
+          3.02414,
+          50.6169
+        ],
+        [
+          3.02312,
+          50.61975
+        ],
+        [
+          3.02172,
+          50.62082
+        ],
+        [
+          3.01953,
+          50.62484
+        ],
+        [
+          3.01811,
+          50.62529
+        ],
+        [
+          3.01313,
+          50.62558
+        ],
+        [
+          3.01385,
+          50.62695
+        ],
+        [
+          3.00844,
+          50.62717
+        ],
+        [
+          3.0056,
+          50.6267
+        ],
+        [
+          3.00229,
+          50.62557
+        ],
+        [
+          3.00119,
+          50.62723
+        ],
+        [
+          2.99769,
+          50.62901
+        ],
+        [
+          2.99391,
+          50.62732
+        ],
+        [
+          2.98971,
+          50.63036
+        ],
+        [
+          2.9862,
+          50.63328
+        ],
+        [
+          2.98178,
+          50.63404
+        ],
+        [
+          2.97917,
+          50.63499
+        ],
+        [
+          2.97284,
+          50.63429
+        ],
+        [
+          2.97174,
+          50.63365
+        ],
+        [
+          2.97002,
+          50.63366
+        ],
+        [
+          2.96956,
+          50.63506
+        ],
+        [
+          2.97046,
+          50.6365
+        ],
+        [
+          2.96878,
+          50.63833
+        ],
+        [
+          2.97039,
+          50.6395
+        ],
+        [
+          2.97275,
+          50.64183
+        ],
+        [
+          2.97225,
+          50.64381
+        ],
+        [
+          2.9745,
+          50.64442
+        ],
+        [
+          2.97474,
+          50.64648
+        ],
+        [
+          2.97091,
+          50.65108
+        ],
+        [
+          2.96975,
+          50.65361
+        ],
+        [
+          2.97061,
+          50.65513
+        ],
+        [
+          2.96929,
+          50.65739
+        ],
+        [
+          2.97072,
+          50.6581
+        ],
+        [
+          2.97973,
+          50.66048
+        ],
+        [
+          2.98369,
+          50.66123
+        ],
+        [
+          2.9865,
+          50.65959
+        ],
+        [
+          2.9896,
+          50.65845
+        ],
+        [
+          2.9963,
+          50.65666
+        ],
+        [
+          2.99903,
+          50.65552
+        ],
+        [
+          3.00274,
+          50.65235
+        ],
+        [
+          3.00714,
+          50.64887
+        ],
+        [
+          3.01088,
+          50.64845
+        ],
+        [
+          3.01318,
+          50.64541
+        ],
+        [
+          3.01974,
+          50.63972
+        ],
+        [
+          3.02317,
+          50.63813
+        ],
+        [
+          3.02639,
+          50.63613
+        ],
+        [
+          3.029,
+          50.63521
+        ],
+        [
+          3.03414,
+          50.6382
+        ],
+        [
+          3.03676,
+          50.63888
+        ],
+        [
+          3.03686,
+          50.64147
+        ],
+        [
+          3.03791,
+          50.64379
+        ],
+        [
+          3.0409,
+          50.64577
+        ],
+        [
+          3.04582,
+          50.64807
+        ],
+        [
+          3.05132,
+          50.64866
+        ],
+        [
+          3.05055,
+          50.64949
+        ],
+        [
+          3.05244,
+          50.65055
+        ],
+        [
+          3.05784,
+          50.64927
+        ],
+        [
+          3.0596,
+          50.65105
+        ],
+        [
+          3.06414,
+          50.65041
+        ],
+        [
+          3.06705,
+          50.64936
+        ],
+        [
+          3.07023,
+          50.64706
+        ],
+        [
+          3.07203,
+          50.64355
+        ],
+        [
+          3.07526,
+          50.64188
+        ],
+        [
+          3.0758,
+          50.64453
+        ],
+        [
+          3.07753,
+          50.64381
+        ],
+        [
+          3.07861,
+          50.64542
+        ],
+        [
+          3.08299,
+          50.64725
+        ],
+        [
+          3.08046,
+          50.64912
+        ],
+        [
+          3.08349,
+          50.65082
+        ],
+        [
+          3.08354,
+          50.65155
+        ],
+        [
+          3.08477,
+          50.65312
+        ],
+        [
+          3.08542,
+          50.65654
+        ],
+        [
+          3.08753,
+          50.65687
+        ],
+        [
+          3.09032,
+          50.65602
+        ],
+        [
+          3.09018,
+          50.65142
+        ],
+        [
+          3.09278,
+          50.65086
+        ],
+        [
+          3.09402,
+          50.64982
+        ],
+        [
+          3.09908,
+          50.65146
+        ],
+        [
+          3.10316,
+          50.65227
+        ],
+        [
+          3.09726,
+          50.64723
+        ],
+        [
+          3.09387,
+          50.64358
+        ],
+        [
+          3.09357,
+          50.64095
+        ],
+        [
+          3.09561,
+          50.64133
+        ],
+        [
+          3.09675,
+          50.64018
+        ],
+        [
+          3.09454,
+          50.63891
+        ],
+        [
+          3.09627,
+          50.63693
+        ],
+        [
+          3.09795,
+          50.63713
+        ],
+        [
+          3.09919,
+          50.63576
+        ],
+        [
+          3.10324,
+          50.6351
+        ],
+        [
+          3.10613,
+          50.63532
+        ],
+        [
+          3.10649,
+          50.63434
+        ],
+        [
+          3.1109,
+          50.63525
+        ],
+        [
+          3.11502,
+          50.63504
+        ],
+        [
+          3.11681,
+          50.63646
+        ]
+      ]
+    ]
+  }

crates/meilisearch/tests/documents/geojson/mod.rs

@@ -0,0 +1,420 @@
+use meili_snap::{json_string, snapshot};
+
+use crate::common::{shared_index_geojson_documents, Server};
+use crate::json;
+
+const LILLE: &str = include_str!("assets/lille.geojson");
+
+#[actix_rt::test]
+async fn basic_add_settings_and_geojson_documents() {
+    let server = Server::new_shared();
+    let index = server.unique_index();
+    let (task, _status_code) =
+        index.update_settings(json!({"filterableAttributes": ["_geojson"]})).await;
+    server.wait_task(task.uid()).await.succeeded();
+
+    let (response, _) = index.search_get("?filter=_geoPolygon([0,0],[0,2],[2,2],[2,0])").await;
+    snapshot!(response,
+    @r#"
+    {
+      "hits": [],
+      "query": "",
+      "processingTimeMs": "[duration]",
+      "limit": 20,
+      "offset": 0,
+      "estimatedTotalHits": 0
+    }
+    "#);
+
+    let lille: serde_json::Value = serde_json::from_str(LILLE).unwrap();
+    let documents = json!([
+        {
+            "id": "missing",
+        },
+        {
+            "id": "point",
+            "_geojson": { "type": "Point", "coordinates": [1, 1] },
+        },
+        {
+            "id": "lille",
+            "_geojson": lille,
+        },
+    ]);
+
+    let (task, _status_code) = index.add_documents(documents, None).await;
+    let response = server.wait_task(task.uid()).await.succeeded();
+    snapshot!(json_string!(response, { ".uid" => "[uid]", ".batchUid" => "[batch_uid]", ".duration" => "[duration]", ".enqueuedAt" => "[date]", ".startedAt" => "[date]", ".finishedAt" => "[date]" }),
+        @r#"
+    {
+      "uid": "[uid]",
+      "batchUid": "[batch_uid]",
+      "indexUid": "[uuid]",
+      "status": "succeeded",
+      "type": "documentAdditionOrUpdate",
+      "canceledBy": null,
+      "details": {
+        "receivedDocuments": 3,
+        "indexedDocuments": 3
+      },
+      "error": null,
+      "duration": "[duration]",
+      "enqueuedAt": "[date]",
+      "startedAt": "[date]",
+      "finishedAt": "[date]"
+    }
+    "#);
+
+    let (response, code) = index.get_all_documents_raw("?ids=missing,point").await;
+
+    snapshot!(code, @"200 OK");
+    snapshot!(response,
+    @r#"
+    {
+      "results": [
+        {
+          "id": "missing"
+        },
+        {
+          "id": "point",
+          "_geojson": {
+            "type": "Point",
+            "coordinates": [
+              1,
+              1
+            ]
+          }
+        }
+      ],
+      "offset": 0,
+      "limit": 20,
+      "total": 2
+    }
+    "#);
+
+    let (response, _code) = index.search_get("?filter=_geoPolygon([0,0],[0,2],[2,2],[2,0])").await;
+    snapshot!(response,
+    @r#"
+    {
+      "hits": [
+        {
+          "id": "point",
+          "_geojson": {
+            "type": "Point",
+            "coordinates": [
+              1,
+              1
+            ]
+          }
+        }
+      ],
+      "query": "",
+      "processingTimeMs": "[duration]",
+      "limit": 20,
+      "offset": 0,
+      "estimatedTotalHits": 1
+    }
+    "#);
+}
+
+#[actix_rt::test]
+async fn basic_add_geojson_documents_and_settings() {
+    let server = Server::new_shared();
+    let index = server.unique_index();
+
+    let lille: serde_json::Value = serde_json::from_str(LILLE).unwrap();
+    let documents = json!([
+        {
+            "id": "missing",
+        },
+        {
+            "id": "point",
+            "_geojson": { "type": "Point", "coordinates": [1, 1] },
+        },
+        {
+            "id": "lille",
+            "_geojson": lille,
+        },
+    ]);
+
+    let (task, _status_code) = index.add_documents(documents, None).await;
+    let response = server.wait_task(task.uid()).await.succeeded();
+    snapshot!(response,
+        @r#"
+    {
+      "uid": "[uid]",
+      "batchUid": "[batch_uid]",
+      "indexUid": "[uuid]",
+      "status": "succeeded",
+      "type": "documentAdditionOrUpdate",
+      "canceledBy": null,
+      "details": {
+        "receivedDocuments": 3,
+        "indexedDocuments": 3
+      },
+      "error": null,
+      "duration": "[duration]",
+      "enqueuedAt": "[date]",
+      "startedAt": "[date]",
+      "finishedAt": "[date]"
+    }
+    "#);
+
+    let (response, _code) = index.search_get("?filter=_geoPolygon([0,0],[0,2],[2,2],[2,0])").await;
+    snapshot!(response,
+    @r#"
+    {
+      "message": "Index `[uuid]`: Attribute `_geojson` is not filterable. This index does not have configured filterable attributes.\n14:15 _geoPolygon([0,0],[0,2],[2,2],[2,0])",
+      "code": "invalid_search_filter",
+      "type": "invalid_request",
+      "link": "https://docs.meilisearch.com/errors#invalid_search_filter"
+    }
+    "#);
+
+    let (task, _status_code) =
+        index.update_settings(json!({"filterableAttributes": ["_geojson"]})).await;
+    server.wait_task(task.uid()).await.succeeded();
+    let (response, _code) = index.search_get("?filter=_geoPolygon([0,0],[0,2],[2,2],[2,0])").await;
+    snapshot!(response,
+    @r#"
+    {
+      "hits": [
+        {
+          "id": "point",
+          "_geojson": {
+            "type": "Point",
+            "coordinates": [
+              1,
+              1
+            ]
+          }
+        }
+      ],
+      "query": "",
+      "processingTimeMs": "[duration]",
+      "limit": 20,
+      "offset": 0,
+      "estimatedTotalHits": 1
+    }
+    "#);
+}
+
+#[actix_rt::test]
+async fn add_and_remove_geojson() {
+    let server = Server::new_shared();
+    let index = server.unique_index();
+    index.update_settings(json!({"filterableAttributes": ["_geojson"]})).await;
+
+    let documents = json!([
+        {
+            "id": "missing",
+        },
+        {
+            "id": 0,
+            "_geojson": { "type": "Point", "coordinates": [1, 1] },
+        }
+    ]);
+    let (task, _status_code) = index.add_documents(documents, None).await;
+    server.wait_task(task.uid()).await.succeeded();
+    let (response, _code) =
+        index.search_get("?filter=_geoPolygon([0,0],[0,0.9],[0.9,0.9],[0.9,0])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 0);
+    let (response, _code) = index.search_get("?filter=_geoPolygon([0,0],[0,2],[2,2],[2,0])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 1);
+
+    let (task, _) = index.delete_document(0).await;
+    server.wait_task(task.uid()).await.succeeded();
+    let (response, _code) =
+        index.search_get("?filter=_geoPolygon([0,0],[0,0.9],[0.9,0.9],[0.9,0])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 0);
+    let (response, _code) = index.search_get("?filter=_geoPolygon([0,0],[0,2],[2,2],[2,0])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 0);
+
+    // add it back
+    let documents = json!([
+        {
+            "id": 0,
+            "_geojson": { "type": "Point", "coordinates": [1, 1] },
+        }
+    ]);
+    let (task, _status_code) = index.add_documents(documents, None).await;
+    server.wait_task(task.uid()).await.succeeded();
+    let (response, _code) =
+        index.search_get("?filter=_geoPolygon([0,0],[0,0.9],[0.9,0.9],[0.9,0])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 0);
+    let (response, _code) = index.search_get("?filter=_geoPolygon([0,0],[0,2],[2,2],[2,0])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 1);
+}
+
+#[actix_rt::test]
+async fn partial_update_geojson() {
+    let server = Server::new_shared();
+    let index = server.unique_index();
+    let (task, _) = index.update_settings(json!({"filterableAttributes": ["_geojson"]})).await;
+    server.wait_task(task.uid()).await.succeeded();
+
+    let documents = json!([
+        {
+            "id": 0,
+            "_geojson": { "type": "Point", "coordinates": [1, 1] },
+        }
+    ]);
+    let (task, _status_code) = index.add_documents(documents, None).await;
+    server.wait_task(task.uid()).await.succeeded();
+    let (response, _code) =
+        index.search_get("?filter=_geoPolygon([0,0],[0,0.9],[0.9,0.9],[0.9,0])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 0);
+    let (response, _code) = index.search_get("?filter=_geoPolygon([0,0],[0,2],[2,2],[2,0])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 1);
+
+    let documents = json!([
+        {
+            "id": 0,
+            "_geojson": { "type": "Point", "coordinates": [0.5, 0.5] },
+        }
+    ]);
+    let (task, _status_code) = index.update_documents(documents, None).await;
+    server.wait_task(task.uid()).await.succeeded();
+    let (response, _code) =
+        index.search_get("?filter=_geoPolygon([0,0],[0,0.9],[0.9,0.9],[0.9,0])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 1);
+    let (response, _code) = index.search_get("?filter=_geoPolygon([0,0],[0,2],[2,2],[2,0])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 1);
+    let (response, _code) =
+        index.search_get("?filter=_geoPolygon([0.9,0.9],[0.9,2],[2,2],[2,0.9])").await;
+    assert_eq!(response.get("hits").unwrap().as_array().unwrap().len(), 0);
+}
+
+#[actix_rt::test]
+async fn geo_bounding_box() {
+    let index = shared_index_geojson_documents().await;
+
+    // The bounding box is a polygon over middle Europe
+    let (response, code) =
+        index.search_get("?filter=_geoBoundingBox([50.53987503447863,21.43443989912143],[43.76393151539099,0.54979129195425])&attributesToRetrieve=name").await;
+    snapshot!(code, @"200 OK");
+    snapshot!(response, @r#"
+    {
+      "hits": [
+        {
+          "name": "Austria"
+        },
+        {
+          "name": "Belgium"
+        },
+        {
+          "name": "Bosnia_and_Herzegovina"
+        },
+        {
+          "name": "Switzerland"
+        },
+        {
+          "name": "Czech_Republic"
+        },
+        {
+          "name": "Germany"
+        },
+        {
+          "name": "France"
+        },
+        {
+          "name": "Croatia"
+        },
+        {
+          "name": "Hungary"
+        },
+        {
+          "name": "Italy"
+        },
+        {
+          "name": "Luxembourg"
+        },
+        {
+          "name": "Netherlands"
+        },
+        {
+          "name": "Poland"
+        },
+        {
+          "name": "Romania"
+        },
+        {
+          "name": "Republic_of_Serbia"
+        },
+        {
+          "name": "Slovakia"
+        },
+        {
+          "name": "Slovenia"
+        }
+      ],
+      "query": "",
+      "processingTimeMs": "[duration]",
+      "limit": 20,
+      "offset": 0,
+      "estimatedTotalHits": 17
+    }
+    "#);
+
+    // Between Russia and Alaska
+    let (response, code) = index
+        .search_get("?filter=_geoBoundingBox([70,-148],[63,152])&attributesToRetrieve=name")
+        .await;
+    snapshot!(code, @"200 OK");
+    snapshot!(response, @r#"
+    {
+      "hits": [
+        {
+          "name": "Canada"
+        },
+        {
+          "name": "Russia"
+        },
+        {
+          "name": "United_States_of_America"
+        }
+      ],
+      "query": "",
+      "processingTimeMs": "[duration]",
+      "limit": 20,
+      "offset": 0,
+      "estimatedTotalHits": 3
+    }
+    "#);
+}
+
+#[actix_rt::test]
+async fn geo_radius() {
+    let index = shared_index_geojson_documents().await;
+
+    // 200km around Luxembourg
+    let (response, code) = index
+        .search_get("?filter=_geoRadius(49.4369862,6.5576591,200000)&attributesToRetrieve=name")
+        .await;
+    snapshot!(code, @"200 OK");
+    snapshot!(response, @r#"
+    {
+      "hits": [
+        {
+          "name": "Belgium"
+        },
+        {
+          "name": "Germany"
+        },
+        {
+          "name": "France"
+        },
+        {
+          "name": "Luxembourg"
+        },
+        {
+          "name": "Netherlands"
+        }
+      ],
+      "query": "",
+      "processingTimeMs": "[duration]",
+      "limit": 20,
+      "offset": 0,
+      "estimatedTotalHits": 5
+    }
+    "#);
+}

crates/meilisearch/tests/documents/mod.rs

@@ -1,5 +1,6 @@
 mod add_documents;
 mod delete_documents;
 mod errors;
+mod geojson;
 mod get_documents;
 mod update_documents;

crates/meilisearch/tests/index/update_index.rs

@@ -1,5 +1,4 @@
 use meili_snap::snapshot;
-
 use time::format_description::well_known::Rfc3339;
 use time::OffsetDateTime;
 

crates/meilisearch/tests/search/errors.rs

@@ -642,14 +642,14 @@ async fn filter_invalid_syntax_object() {
         &json!({"filterableAttributes": ["title"]}),
         &json!({"filter": "title & Glass"}),
         |response, code| {
-            snapshot!(response, @r###"
+            snapshot!(response, @r#"
             {
-              "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `title & Glass`.\n1:14 title & Glass",
+              "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `title & Glass`.\n1:14 title & Glass",
               "code": "invalid_search_filter",
               "type": "invalid_request",
               "link": "https://docs.meilisearch.com/errors#invalid_search_filter"
             }
-            "###);
+            "#);
             snapshot!(code, @"400 Bad Request");
         },
     )
@@ -663,14 +663,14 @@ async fn filter_invalid_syntax_array() {
         &json!({"filterableAttributes": ["title"]}),
         &json!({"filter": ["title & Glass"]}),
         |response, code| {
-            snapshot!(response, @r###"
+            snapshot!(response, @r#"
             {
-              "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `title & Glass`.\n1:14 title & Glass",
+              "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `title & Glass`.\n1:14 title & Glass",
               "code": "invalid_search_filter",
               "type": "invalid_request",
               "link": "https://docs.meilisearch.com/errors#invalid_search_filter"
             }
-            "###);
+            "#);
             snapshot!(code, @"400 Bad Request");
         },
     )

crates/meilisearch/tests/search/multi/proxy.rs

@@ -2,8 +2,7 @@ use std::sync::Arc;
 
 use actix_http::StatusCode;
 use meili_snap::{json_string, snapshot};
-use wiremock::matchers::method;
-use wiremock::matchers::{path, AnyMatcher};
+use wiremock::matchers::{method, path, AnyMatcher};
 use wiremock::{Mock, MockServer, Request, ResponseTemplate};
 
 use crate::common::{Server, Value, SCORE_DOCUMENTS};
@@ -128,7 +127,7 @@ async fn remote_sharding() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -137,7 +136,7 @@ async fn remote_sharding() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -146,7 +145,7 @@ async fn remote_sharding() {
     }
     "###);
     let (response, code) = ms2.set_network(json!({"self": "ms2"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms2",
@@ -193,11 +192,11 @@ async fn remote_sharding() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
     let (_response, status_code) = ms1.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
     let (_response, status_code) = ms2.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -443,7 +442,7 @@ async fn remote_sharding_retrieve_vectors() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -452,7 +451,7 @@ async fn remote_sharding_retrieve_vectors() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -461,7 +460,7 @@ async fn remote_sharding_retrieve_vectors() {
     }
     "###);
     let (response, code) = ms2.set_network(json!({"self": "ms2"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms2",
@@ -543,11 +542,11 @@ async fn remote_sharding_retrieve_vectors() {
     }});
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
     let (_response, status_code) = ms1.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
     let (_response, status_code) = ms2.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // multi vector search: one query per remote
 
@@ -936,7 +935,7 @@ async fn error_unregistered_remote() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -945,7 +944,7 @@ async fn error_unregistered_remote() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -984,9 +983,9 @@ async fn error_unregistered_remote() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
     let (_response, status_code) = ms1.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -1056,7 +1055,7 @@ async fn error_no_weighted_score() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -1065,7 +1064,7 @@ async fn error_no_weighted_score() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -1108,7 +1107,7 @@ async fn error_no_weighted_score() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -1191,7 +1190,7 @@ async fn error_bad_response() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -1200,7 +1199,7 @@ async fn error_bad_response() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -1246,7 +1245,7 @@ async fn error_bad_response() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -1330,7 +1329,7 @@ async fn error_bad_request() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -1339,7 +1338,7 @@ async fn error_bad_request() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -1378,7 +1377,7 @@ async fn error_bad_request() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -1462,7 +1461,7 @@ async fn error_bad_request_facets_by_index() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -1471,7 +1470,7 @@ async fn error_bad_request_facets_by_index() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -1511,7 +1510,7 @@ async fn error_bad_request_facets_by_index() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -1605,7 +1604,7 @@ async fn error_bad_request_facets_by_index_facet() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -1614,7 +1613,7 @@ async fn error_bad_request_facets_by_index_facet() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -1657,7 +1656,7 @@ async fn error_bad_request_facets_by_index_facet() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -1757,7 +1756,7 @@ async fn error_remote_does_not_answer() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -1766,7 +1765,7 @@ async fn error_remote_does_not_answer() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -1808,9 +1807,9 @@ async fn error_remote_does_not_answer() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
     let (_response, status_code) = ms1.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -1960,7 +1959,7 @@ async fn error_remote_404() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -1969,7 +1968,7 @@ async fn error_remote_404() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -2008,9 +2007,9 @@ async fn error_remote_404() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
     let (_response, status_code) = ms1.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -2157,7 +2156,7 @@ async fn error_remote_sharding_auth() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -2166,7 +2165,7 @@ async fn error_remote_sharding_auth() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -2212,7 +2211,7 @@ async fn error_remote_sharding_auth() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -2319,7 +2318,7 @@ async fn remote_sharding_auth() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -2328,7 +2327,7 @@ async fn remote_sharding_auth() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -2373,7 +2372,7 @@ async fn remote_sharding_auth() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -2476,7 +2475,7 @@ async fn error_remote_500() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -2485,7 +2484,7 @@ async fn error_remote_500() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -2528,9 +2527,9 @@ async fn error_remote_500() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
     let (_response, status_code) = ms1.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -2657,7 +2656,7 @@ async fn error_remote_500_once() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -2666,7 +2665,7 @@ async fn error_remote_500_once() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -2709,9 +2708,9 @@ async fn error_remote_500_once() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
     let (_response, status_code) = ms1.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";
@@ -2842,7 +2841,7 @@ async fn error_remote_timeout() {
     // set self
 
     let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms0",
@@ -2850,7 +2849,7 @@ async fn error_remote_timeout() {
     }
     "###);
     let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
-    snapshot!(code, @"200 OK");
+    snapshot!(code, @"202 Accepted");
     snapshot!(json_string!(response), @r###"
     {
       "self": "ms1",
@@ -2892,9 +2891,9 @@ async fn error_remote_timeout() {
     println!("{}", serde_json::to_string_pretty(&network).unwrap());
 
     let (_response, status_code) = ms0.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
     let (_response, status_code) = ms1.set_network(network.clone()).await;
-    snapshot!(status_code, @"200 OK");
+    snapshot!(status_code, @"202 Accepted");
 
     // perform multi-search
     let query = "badman returns";

crates/meilisearch/tests/search/pagination.rs

@@ -1,7 +1,8 @@
+use meili_snap::{json_string, snapshot};
+
 use super::shared_index_with_documents;
 use crate::common::Server;
 use crate::json;
-use meili_snap::{json_string, snapshot};
 
 #[actix_rt::test]
 async fn default_search_should_return_estimated_total_hit() {

crates/meilisearch/tests/settings/chat.rs

@@ -1,6 +1,7 @@
+use meili_snap::{json_string, snapshot};
+
 use crate::common::Server;
 use crate::json;
-use meili_snap::{json_string, snapshot};
 
 #[actix_rt::test]
 async fn set_reset_chat_issue_5772() {

crates/meilisearch/tests/similar/errors.rs

@@ -339,14 +339,14 @@ async fn filter_invalid_syntax_object() {
 
     index
         .similar(json!({"id": 287947, "filter": "title & Glass", "embedder": "manual"}), |response, code| {
-            snapshot!(response, @r###"
+            snapshot!(response, @r#"
             {
-              "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `title & Glass`.\n1:14 title & Glass",
+              "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `title & Glass`.\n1:14 title & Glass",
               "code": "invalid_similar_filter",
               "type": "invalid_request",
               "link": "https://docs.meilisearch.com/errors#invalid_similar_filter"
             }
-            "###);
+            "#);
             snapshot!(code, @"400 Bad Request");
         })
         .await;
@@ -377,14 +377,14 @@ async fn filter_invalid_syntax_array() {
 
     index
         .similar(json!({"id": 287947, "filter": ["title & Glass"], "embedder": "manual"}), |response, code| {
-            snapshot!(response, @r###"
+            snapshot!(response, @r#"
             {
-              "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, or `_geoBoundingBox` at `title & Glass`.\n1:14 title & Glass",
+              "message": "Was expecting an operation `=`, `!=`, `>=`, `>`, `<=`, `<`, `IN`, `NOT IN`, `TO`, `EXISTS`, `NOT EXISTS`, `IS NULL`, `IS NOT NULL`, `IS EMPTY`, `IS NOT EMPTY`, `CONTAINS`, `NOT CONTAINS`, `STARTS WITH`, `NOT STARTS WITH`, `_geoRadius`, `_geoBoundingBox` or `_geoPolygon` at `title & Glass`.\n1:14 title & Glass",
               "code": "invalid_similar_filter",
               "type": "invalid_request",
               "link": "https://docs.meilisearch.com/errors#invalid_similar_filter"
             }
-            "###);
+            "#);
             snapshot!(code, @"400 Bad Request");
         })
         .await;

crates/meilisearch/tests/tasks/errors.rs

@@ -95,14 +95,14 @@ async fn task_bad_types() {
 
     let (response, code) = server.tasks_filter("types=doggo").await;
     snapshot!(code, @"400 Bad Request");
-    snapshot!(json_string!(response), @r#"
+    snapshot!(json_string!(response), @r###"
     {
-      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`.",
+      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`, `networkTopologyChange`.",
       "code": "invalid_task_types",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_task_types"
     }
-    "#);
+    "###);
 
     let (response, code) = server.cancel_tasks("types=doggo").await;
     snapshot!(code, @"400 Bad Request");

crates/meilitool/src/main.rs

@@ -1,3 +1,5 @@
+#![allow(clippy::result_large_err)]
+
 use std::fs::{read_dir, read_to_string, remove_file, File};
 use std::io::{BufWriter, Write as _};
 use std::path::PathBuf;

crates/milli/Cargo.toml

@@ -19,6 +19,7 @@ bstr = "1.12.0"
 bytemuck = { version = "1.23.1", features = ["extern_crate_alloc"] }
 byteorder = "1.5.0"
 charabia = { version = "0.9.7", default-features = false }
+cellulite = "0.3.0"
 concat-arrays = "0.1.2"
 convert_case = "0.8.0"
 crossbeam-channel = "0.5.15"
@@ -27,6 +28,7 @@ either = { version = "1.15.0", features = ["serde"] }
 flatten-serde-json = { path = "../flatten-serde-json" }
 fst = "0.4.7"
 fxhash = "0.2.1"
+geojson = "0.24.2"
 geoutils = "0.5.1"
 grenad = { version = "0.5.0", default-features = false, features = [
     "rayon",
@@ -96,7 +98,7 @@ url = "2.5.4"
 hashbrown = "0.15.4"
 bumpalo = "3.18.1"
 bumparaw-collections = "0.1.4"
-steppe = { version = "0.4.0", default-features = false }
+steppe = { version = "0.4", default-features = false }
 thread_local = "1.1.9"
 allocator-api2 = "0.3.0"
 rustc-hash = "2.1.1"
@@ -116,6 +118,8 @@ twox-hash = { version = "2.1.1", default-features = false, features = [
     "xxhash3_64",
     "xxhash64",
 ] }
+geo-types = "0.7.16"
+zerometry = "0.1.0"
 
 [dev-dependencies]
 mimalloc = { version = "0.1.47", default-features = false }

crates/milli/src/constants.rs

@@ -11,3 +11,4 @@ const fn parse_u32(s: &str) -> u32 {
 
 pub const RESERVED_VECTORS_FIELD_NAME: &str = "_vectors";
 pub const RESERVED_GEO_FIELD_NAME: &str = "_geo";
+pub const RESERVED_GEOJSON_FIELD_NAME: &str = "_geojson";

crates/milli/src/documents/primary_key.rs

@@ -48,6 +48,7 @@ pub enum PrimaryKey<'a> {
     Nested { name: &'a str },
 }
 
+#[allow(clippy::large_enum_variant)]
 pub enum DocumentIdExtractionError {
     InvalidDocumentId(UserError),
     MissingDocumentId,

crates/milli/src/error.rs

@@ -10,17 +10,26 @@ use rhai::EvalAltResult;
 use serde_json::Value;
 use thiserror::Error;
 
-use crate::constants::RESERVED_GEO_FIELD_NAME;
+use crate::constants::{RESERVED_GEOJSON_FIELD_NAME, RESERVED_GEO_FIELD_NAME};
 use crate::documents::{self, DocumentsBatchCursorError};
 use crate::thread_pool_no_abort::PanicCatched;
 use crate::vector::settings::EmbeddingSettings;
 use crate::{CriterionError, DocumentId, FieldId, Object, SortError};
 
 pub fn is_reserved_keyword(keyword: &str) -> bool {
-    [RESERVED_GEO_FIELD_NAME, "_geoDistance", "_geoPoint", "_geoRadius", "_geoBoundingBox"]
-        .contains(&keyword)
+    [
+        RESERVED_GEO_FIELD_NAME,
+        RESERVED_GEOJSON_FIELD_NAME,
+        "_geoDistance",
+        "_geoPoint",
+        "_geoRadius",
+        "_geoBoundingBox",
+        "_geoPolygon",
+    ]
+    .contains(&keyword)
 }
 
+#[allow(clippy::large_enum_variant)]
 #[derive(Error, Debug)]
 pub enum Error {
     #[error("internal: {0}.")]
@@ -80,6 +89,8 @@ pub enum InternalError {
     #[error(transparent)]
     HannoyError(#[from] hannoy::Error),
     #[error(transparent)]
+    CelluliteError(#[from] cellulite::Error),
+    #[error(transparent)]
     VectorEmbeddingError(#[from] crate::vector::Error),
 }
 
@@ -99,6 +110,12 @@ pub enum SerializationError {
     InvalidNumberSerialization,
 }
 
+impl From<cellulite::Error> for Error {
+    fn from(error: cellulite::Error) -> Self {
+        Self::UserError(UserError::CelluliteError(error))
+    }
+}
+
 #[derive(Error, Debug)]
 pub enum FieldIdMapMissingEntry {
     #[error("unknown field id {field_id} coming from the {process} process")]
@@ -107,8 +124,13 @@ pub enum FieldIdMapMissingEntry {
     FieldName { field_name: String, process: &'static str },
 }
 
+#[allow(clippy::large_enum_variant)]
 #[derive(Error, Debug)]
 pub enum UserError {
+    #[error(transparent)]
+    CelluliteError(#[from] cellulite::Error),
+    #[error("Malformed geojson: {0}")]
+    MalformedGeojson(serde_json::Error),
     #[error("A document cannot contain more than 65,535 fields.")]
     AttributeLimitReached,
     #[error(transparent)]
@@ -153,6 +175,8 @@ and can not be more than 511 bytes.", .document_id.to_string()
     },
     #[error(transparent)]
     InvalidGeoField(#[from] Box<GeoError>),
+    #[error(transparent)]
+    GeoJsonError(#[from] geojson::Error),
     #[error("Invalid vector dimensions: expected: `{}`, found: `{}`.", .expected, .found)]
     InvalidVectorDimensions { expected: usize, found: usize },
     #[error("Invalid vector dimensions in document with id `{document_id}` in `._vectors.{embedder_name}`.\n  - note: embedding #{embedding_index} has dimensions {found}\n  - note: embedder `{embedder_name}` requires {expected}")]
@@ -408,6 +432,10 @@ and can not be more than 511 bytes.", .document_id.to_string()
     InvalidChatSettingsDocumentTemplateMaxBytes,
     #[error("{0}")]
     DocumentEmbeddingError(String),
+    #[error("enabling the sharding requires `.self` to be set\n  - Hint: Disable `sharding` or set `self` to a value.")]
+    NetworkShardingWithoutSelf,
+    #[error("Field `.remotes.{0}.url` cannot be set to `null`")]
+    NetworkMissingUrl(String),
 }
 
 impl From<crate::vector::Error> for Error {

crates/milli/src/fields_ids_map/metadata.rs

@@ -6,7 +6,9 @@ use heed::RoTxn;
 
 use super::FieldsIdsMap;
 use crate::attribute_patterns::{match_field_legacy, PatternMatch};
-use crate::constants::{RESERVED_GEO_FIELD_NAME, RESERVED_VECTORS_FIELD_NAME};
+use crate::constants::{
+    RESERVED_GEOJSON_FIELD_NAME, RESERVED_GEO_FIELD_NAME, RESERVED_VECTORS_FIELD_NAME,
+};
 use crate::{
     is_faceted_by, FieldId, FilterableAttributesFeatures, FilterableAttributesRule, Index,
     LocalizedAttributesRule, Result, Weight,
@@ -24,6 +26,8 @@ pub struct Metadata {
     pub asc_desc: bool,
     /// The field is a geo field (`_geo`, `_geo.lat`, `_geo.lng`).
     pub geo: bool,
+    /// The field is a geo json field (`_geojson`).
+    pub geo_json: bool,
     /// The id of the localized attributes rule if the field is localized.
     pub localized_attributes_rule_id: Option<NonZeroU16>,
     /// The id of the filterable attributes rule if the field is filterable.
@@ -269,6 +273,7 @@ impl MetadataBuilder {
                 distinct: false,
                 asc_desc: false,
                 geo: false,
+                geo_json: false,
                 localized_attributes_rule_id: None,
                 filterable_attributes_rule_id: None,
             };
@@ -295,6 +300,20 @@ impl MetadataBuilder {
                 distinct: false,
                 asc_desc: false,
                 geo: true,
+                geo_json: false,
+                localized_attributes_rule_id: None,
+                filterable_attributes_rule_id,
+            };
+        }
+        if match_field_legacy(RESERVED_GEOJSON_FIELD_NAME, field) == PatternMatch::Match {
+            debug_assert!(!sortable, "geojson fields should not be sortable");
+            return Metadata {
+                searchable: None,
+                sortable,
+                distinct: false,
+                asc_desc: false,
+                geo: false,
+                geo_json: true,
                 localized_attributes_rule_id: None,
                 filterable_attributes_rule_id,
             };
@@ -328,6 +347,7 @@ impl MetadataBuilder {
             distinct,
             asc_desc,
             geo: false,
+            geo_json: false,
             localized_attributes_rule_id,
             filterable_attributes_rule_id,
         }

crates/milli/src/filterable_attributes_rules.rs

@@ -5,7 +5,7 @@ use serde::{Deserialize, Serialize};
 use utoipa::ToSchema;
 
 use crate::attribute_patterns::{match_distinct_field, match_field_legacy, PatternMatch};
-use crate::constants::RESERVED_GEO_FIELD_NAME;
+use crate::constants::{RESERVED_GEOJSON_FIELD_NAME, RESERVED_GEO_FIELD_NAME};
 use crate::AttributePatterns;
 
 #[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Debug, ToSchema)]
@@ -34,6 +34,10 @@ impl FilterableAttributesRule {
         matches!(self, FilterableAttributesRule::Field(field_name) if field_name == RESERVED_GEO_FIELD_NAME)
     }
 
+    pub fn has_geojson(&self) -> bool {
+        matches!(self, FilterableAttributesRule::Field(field_name) if field_name == RESERVED_GEOJSON_FIELD_NAME)
+    }
+
     /// Get the features of the rule.
     pub fn features(&self) -> FilterableAttributesFeatures {
         match self {

crates/milli/src/index.rs

@@ -5,6 +5,7 @@ use std::fmt;
 use std::fs::File;
 use std::path::Path;
 
+use cellulite::Cellulite;
 use deserr::Deserr;
 use heed::types::*;
 use heed::{CompactionOption, Database, DatabaseStat, RoTxn, RwTxn, Unspecified, WithoutTls};
@@ -115,9 +116,10 @@ pub mod db_name {
     pub const FIELD_ID_DOCID_FACET_STRINGS: &str = "field-id-docid-facet-strings";
     pub const VECTOR_EMBEDDER_CATEGORY_ID: &str = "vector-embedder-category-id";
     pub const VECTOR_STORE: &str = "vector-arroy";
+    pub const CELLULITE: &str = "cellulite";
     pub const DOCUMENTS: &str = "documents";
 }
-const NUMBER_OF_DBS: u32 = 25;
+const NUMBER_OF_DBS: u32 = 25 + Cellulite::nb_dbs();
 
 #[derive(Clone)]
 pub struct Index {
@@ -183,6 +185,9 @@ pub struct Index {
     /// Vector store based on hannoy™.
     pub vector_store: hannoy::Database<Unspecified>,
 
+    /// Geo store based on cellulite™.
+    pub cellulite: Cellulite,
+
     /// Maps the document id to the document as an obkv store.
     pub(crate) documents: Database<BEU32, ObkvCodec>,
 }
@@ -239,6 +244,7 @@ impl Index {
         let embedder_category_id =
             env.create_database(&mut wtxn, Some(VECTOR_EMBEDDER_CATEGORY_ID))?;
         let vector_store = env.create_database(&mut wtxn, Some(VECTOR_STORE))?;
+        let cellulite = cellulite::Cellulite::create_from_env(&env, &mut wtxn, CELLULITE)?;
 
         let documents = env.create_database(&mut wtxn, Some(DOCUMENTS))?;
 
@@ -267,6 +273,7 @@ impl Index {
             field_id_docid_facet_strings,
             vector_store,
             embedder_category_id,
+            cellulite,
             documents,
         };
         if this.get_version(&wtxn)?.is_none() && creation {
@@ -1052,6 +1059,13 @@ impl Index {
         Ok(geo_filter)
     }
 
+    /// Returns true if the geo sorting feature is enabled.
+    pub fn is_geojson_filtering_enabled(&self, rtxn: &RoTxn<'_>) -> Result<bool> {
+        let geojson_filter =
+            self.filterable_attributes_rules(rtxn)?.iter().any(|field| field.has_geojson());
+        Ok(geojson_filter)
+    }
+
     pub fn asc_desc_fields(&self, rtxn: &RoTxn<'_>) -> Result<HashSet<String>> {
         let asc_desc_fields = self
             .criteria(rtxn)?
@@ -1882,6 +1896,7 @@ impl Index {
             field_id_docid_facet_strings,
             vector_store,
             embedder_category_id,
+            cellulite,
             documents,
         } = self;
 
@@ -1955,6 +1970,17 @@ impl Index {
         sizes.insert("embedder_category_id", embedder_category_id.stat(rtxn).map(compute_size)?);
         sizes.insert("documents", documents.stat(rtxn).map(compute_size)?);
 
+        // Cellulite
+        const _CELLULITE_DB_CHECK: () = {
+            if Cellulite::nb_dbs() != 4 {
+                panic!("Cellulite database count has changed, please update the code accordingly.")
+            }
+        };
+        sizes.insert("cellulite_item", cellulite.item_db_stats(rtxn).map(compute_size)?);
+        sizes.insert("cellulite_cell", cellulite.cell_db_stats(rtxn).map(compute_size)?);
+        sizes.insert("cellulite_update", cellulite.update_db_stats(rtxn).map(compute_size)?);
+        sizes.insert("cellulite_metadata", cellulite.metadata_db_stats(rtxn).map(compute_size)?);
+
         Ok(sizes)
     }
 }

crates/milli/src/lib.rs

@@ -54,7 +54,7 @@ pub use search::new::{
 };
 use serde_json::Value;
 pub use thread_pool_no_abort::{PanicCatched, ThreadPoolNoAbort, ThreadPoolNoAbortBuilder};
-pub use {arroy, charabia as tokenizer, hannoy, heed, rhai};
+pub use {arroy, cellulite, charabia as tokenizer, hannoy, heed, rhai};
 
 pub use self::asc_desc::{AscDesc, AscDescError, Member, SortError};
 pub use self::attribute_patterns::{AttributePatterns, PatternMatch};
@@ -87,7 +87,7 @@ pub use self::search::{
 };
 pub use self::update::ChannelCongestion;
 
-pub type Result<T> = std::result::Result<T, error::Error>;
+pub type Result<T, E = error::Error> = std::result::Result<T, E>;
 
 pub type Attribute = u32;
 pub type BEU16 = heed::types::U16<heed::byteorder::BE>;

crates/milli/src/progress.rs

@@ -278,30 +278,6 @@ impl<U: Send + Sync + 'static> Step for VariableNameStep<U> {
     }
 }
 
-// Integration with steppe
-
-impl steppe::Progress for Progress {
-    fn update(&self, sub_progress: impl steppe::Step) {
-        self.update_progress(Compat(sub_progress));
-    }
-}
-
-struct Compat<T: steppe::Step>(T);
-
-impl<T: steppe::Step> Step for Compat<T> {
-    fn name(&self) -> Cow<'static, str> {
-        self.0.name()
-    }
-
-    fn current(&self) -> u32 {
-        self.0.current().try_into().unwrap_or(u32::MAX)
-    }
-
-    fn total(&self) -> u32 {
-        self.0.total().try_into().unwrap_or(u32::MAX)
-    }
-}
-
 impl Step for arroy::MainStep {
     fn name(&self) -> Cow<'static, str> {
         match self {
@@ -343,3 +319,27 @@ impl Step for arroy::SubStep {
         self.max
     }
 }
+
+// Integration with steppe
+
+impl steppe::Progress for Progress {
+    fn update(&self, sub_progress: impl steppe::Step) {
+        self.update_progress(Compat(sub_progress));
+    }
+}
+
+struct Compat<T: steppe::Step>(T);
+
+impl<T: steppe::Step> Step for Compat<T> {
+    fn name(&self) -> Cow<'static, str> {
+        self.0.name()
+    }
+
+    fn current(&self) -> u32 {
+        self.0.current().try_into().unwrap_or(u32::MAX)
+    }
+
+    fn total(&self) -> u32 {
+        self.0.total().try_into().unwrap_or(u32::MAX)
+    }
+}

crates/milli/src/search/facet/filter.rs

@@ -12,7 +12,9 @@ use roaring::{MultiOps, RoaringBitmap};
 use serde_json::Value;
 
 use super::facet_range_search;
-use crate::constants::{RESERVED_GEO_FIELD_NAME, RESERVED_VECTORS_FIELD_NAME};
+use crate::constants::{
+    RESERVED_GEOJSON_FIELD_NAME, RESERVED_GEO_FIELD_NAME, RESERVED_VECTORS_FIELD_NAME,
+};
 use crate::error::{Error, UserError};
 use crate::filterable_attributes_rules::{filtered_matching_patterns, matching_features};
 use crate::heed_codec::facet::{FacetGroupKey, FacetGroupKeyCodec, FacetGroupValueCodec};
@@ -36,6 +38,7 @@ pub struct Filter<'a> {
 pub enum BadGeoError {
     Lat(f64),
     Lng(f64),
+    InvalidResolution(usize),
     BoundingBoxTopIsBelowBottom(f64, f64),
 }
 
@@ -47,16 +50,23 @@ impl Display for BadGeoError {
             Self::BoundingBoxTopIsBelowBottom(top, bottom) => {
                 write!(f, "The top latitude `{top}` is below the bottom latitude `{bottom}`.")
             }
+            Self::InvalidResolution(resolution) => write!(
+                f,
+                "Invalid resolution `{resolution}`. Resolution must be between 3 and 1000."
+            ),
             Self::Lat(lat) => write!(
                 f,
-                "Bad latitude `{}`. Latitude must be contained between -90 and 90 degrees. ",
+                "Bad latitude `{}`. Latitude must be contained between -90 and 90 degrees.",
                 lat
             ),
-            Self::Lng(lng) => write!(
-                f,
-                "Bad longitude `{}`. Longitude must be contained between -180 and 180 degrees. ",
-                lng
-            ),
+            Self::Lng(lng) => {
+                let normalized = (lng + 180.0).rem_euclid(360.0) - 180.0;
+                write!(
+                    f,
+                    "Bad longitude `{}`. Longitude must be contained between -180 and 180 degrees. Hint: try using `{normalized}` instead.",
+                    lng
+                )
+            }
         }
     }
 }
@@ -612,50 +622,61 @@ impl<'a> Filter<'a> {
                 .union(),
             FilterCondition::And(subfilters) => {
                 let mut subfilters_iter = subfilters.iter();
-                if let Some(first_subfilter) = subfilters_iter.next() {
-                    let mut bitmap = Self::inner_evaluate(
-                        &(first_subfilter.clone()).into(),
+                let Some(first_subfilter) = subfilters_iter.next() else {
+                    return Ok(RoaringBitmap::new());
+                };
+
+                let mut bitmap = Self::inner_evaluate(
+                    &(first_subfilter.clone()).into(),
+                    rtxn,
+                    index,
+                    field_ids_map,
+                    filterable_attribute_rules,
+                    universe,
+                )?;
+                for f in subfilters_iter {
+                    if bitmap.is_empty() {
+                        return Ok(bitmap);
+                    }
+                    // TODO We are doing the intersections two times,
+                    //      it could be more efficient
+                    //      Can't I just replace this `&=` by an `=`?
+                    bitmap &= Self::inner_evaluate(
+                        &(f.clone()).into(),
                         rtxn,
                         index,
                         field_ids_map,
                         filterable_attribute_rules,
-                        universe,
+                        Some(&bitmap),
                     )?;
-                    for f in subfilters_iter {
-                        if bitmap.is_empty() {
-                            return Ok(bitmap);
-                        }
-                        // TODO We are doing the intersections two times,
-                        //      it could be more efficient
-                        //      Can't I just replace this `&=` by an `=`?
-                        bitmap &= Self::inner_evaluate(
-                            &(f.clone()).into(),
-                            rtxn,
-                            index,
-                            field_ids_map,
-                            filterable_attribute_rules,
-                            Some(&bitmap),
-                        )?;
-                    }
-                    Ok(bitmap)
-                } else {
-                    Ok(RoaringBitmap::new())
                 }
+                Ok(bitmap)
             }
             FilterCondition::VectorExists { fid: _, embedder, filter } => {
                 super::filter_vector::evaluate(rtxn, index, universe, embedder.clone(), filter)
             }
-            FilterCondition::GeoLowerThan { point, radius } => {
+            FilterCondition::GeoLowerThan { point, radius, resolution: res_token } => {
+                let base_point: [f64; 2] =
+                    [point[0].parse_finite_float()?, point[1].parse_finite_float()?];
+                if !(-90.0..=90.0).contains(&base_point[0]) {
+                    return Err(point[0].as_external_error(BadGeoError::Lat(base_point[0])))?;
+                }
+                if !(-180.0..=180.0).contains(&base_point[1]) {
+                    return Err(point[1].as_external_error(BadGeoError::Lng(base_point[1])))?;
+                }
+                let radius = radius.parse_finite_float()?;
+                let mut resolution = 125;
+                if let Some(res_token) = res_token {
+                    resolution = res_token.parse_finite_float()? as usize;
+                    if !(3..=1000).contains(&resolution) {
+                        return Err(
+                            res_token.as_external_error(BadGeoError::InvalidResolution(resolution))
+                        )?;
+                    }
+                }
+
+                let mut r1 = None;
                 if index.is_geo_filtering_enabled(rtxn)? {
-                    let base_point: [f64; 2] =
-                        [point[0].parse_finite_float()?, point[1].parse_finite_float()?];
-                    if !(-90.0..=90.0).contains(&base_point[0]) {
-                        return Err(point[0].as_external_error(BadGeoError::Lat(base_point[0])))?;
-                    }
-                    if !(-180.0..=180.0).contains(&base_point[1]) {
-                        return Err(point[1].as_external_error(BadGeoError::Lng(base_point[1])))?;
-                    }
-                    let radius = radius.parse_finite_float()?;
                     let rtree = match index.geo_rtree(rtxn)? {
                         Some(rtree) => rtree,
                         None => return Ok(RoaringBitmap::new()),
@@ -671,52 +692,72 @@ impl<'a> Filter<'a> {
                         })
                         .map(|point| point.data.0)
                         .collect();
+                    r1 = Some(result);
+                }
 
-                    Ok(result)
-                } else {
-                    Err(point[0].as_external_error(FilterError::AttributeNotFilterable {
-                        attribute: RESERVED_GEO_FIELD_NAME,
-                        filterable_patterns: filtered_matching_patterns(
-                            filterable_attribute_rules,
-                            &|features| features.is_filterable(),
-                        ),
-                    }))?
+                let mut r2 = None;
+                if index.is_geojson_filtering_enabled(rtxn)? {
+                    let point = geo_types::Point::new(base_point[1], base_point[0]);
+
+                    let result = index.cellulite.in_circle(rtxn, point, radius, resolution)?;
+
+                    r2 = Some(RoaringBitmap::from_iter(result)); // TODO: Remove once we update roaring in meilisearch
+                }
+
+                match (r1, r2) {
+                    (Some(r1), Some(r2)) => Ok(r1 | r2),
+                    (Some(r1), None) => Ok(r1),
+                    (None, Some(r2)) => Ok(r2),
+                    (None, None) => {
+                        Err(point[0].as_external_error(FilterError::AttributeNotFilterable {
+                            attribute: &format!(
+                                "{RESERVED_GEO_FIELD_NAME}/{RESERVED_GEOJSON_FIELD_NAME}"
+                            ),
+                            filterable_patterns: filtered_matching_patterns(
+                                filterable_attribute_rules,
+                                &|features| features.is_filterable(),
+                            ),
+                        }))?
+                    }
                 }
             }
             FilterCondition::GeoBoundingBox { top_right_point, bottom_left_point } => {
-                if index.is_geo_filtering_enabled(rtxn)? {
-                    let top_right: [f64; 2] = [
-                        top_right_point[0].parse_finite_float()?,
-                        top_right_point[1].parse_finite_float()?,
-                    ];
-                    let bottom_left: [f64; 2] = [
-                        bottom_left_point[0].parse_finite_float()?,
-                        bottom_left_point[1].parse_finite_float()?,
-                    ];
-                    if !(-90.0..=90.0).contains(&top_right[0]) {
-                        return Err(
-                            top_right_point[0].as_external_error(BadGeoError::Lat(top_right[0]))
-                        )?;
-                    }
-                    if !(-180.0..=180.0).contains(&top_right[1]) {
-                        return Err(
-                            top_right_point[1].as_external_error(BadGeoError::Lng(top_right[1]))
-                        )?;
-                    }
-                    if !(-90.0..=90.0).contains(&bottom_left[0]) {
-                        return Err(bottom_left_point[0]
-                            .as_external_error(BadGeoError::Lat(bottom_left[0])))?;
-                    }
-                    if !(-180.0..=180.0).contains(&bottom_left[1]) {
-                        return Err(bottom_left_point[1]
-                            .as_external_error(BadGeoError::Lng(bottom_left[1])))?;
-                    }
-                    if top_right[0] < bottom_left[0] {
-                        return Err(bottom_left_point[1].as_external_error(
-                            BadGeoError::BoundingBoxTopIsBelowBottom(top_right[0], bottom_left[0]),
-                        ))?;
-                    }
+                let top_right: [f64; 2] = [
+                    top_right_point[0].parse_finite_float()?,
+                    top_right_point[1].parse_finite_float()?,
+                ];
+                let bottom_left: [f64; 2] = [
+                    bottom_left_point[0].parse_finite_float()?,
+                    bottom_left_point[1].parse_finite_float()?,
+                ];
+                if !(-90.0..=90.0).contains(&top_right[0]) {
+                    return Err(
+                        top_right_point[0].as_external_error(BadGeoError::Lat(top_right[0]))
+                    )?;
+                }
+                if !(-180.0..=180.0).contains(&top_right[1]) {
+                    return Err(
+                        top_right_point[1].as_external_error(BadGeoError::Lng(top_right[1]))
+                    )?;
+                }
+                if !(-90.0..=90.0).contains(&bottom_left[0]) {
+                    return Err(
+                        bottom_left_point[0].as_external_error(BadGeoError::Lat(bottom_left[0]))
+                    )?;
+                }
+                if !(-180.0..=180.0).contains(&bottom_left[1]) {
+                    return Err(
+                        bottom_left_point[1].as_external_error(BadGeoError::Lng(bottom_left[1]))
+                    )?;
+                }
+                if top_right[0] < bottom_left[0] {
+                    return Err(bottom_left_point[1].as_external_error(
+                        BadGeoError::BoundingBoxTopIsBelowBottom(top_right[0], bottom_left[0]),
+                    ))?;
+                }
 
+                let mut r1 = None;
+                if index.is_geo_filtering_enabled(rtxn)? {
                     // Instead of writing a custom `GeoBoundingBox` filter we're simply going to re-use the range
                     // filter to create the following filter;
                     // `_geo.lat {top_right[0]} TO {bottom_left[0]} AND _geo.lng {top_right[1]} TO {bottom_left[1]}`
@@ -811,19 +852,76 @@ impl<'a> Filter<'a> {
                         )?
                     };
 
-                    Ok(selected_lat & selected_lng)
-                } else {
-                    Err(top_right_point[0].as_external_error(
+                    r1 = Some(selected_lat & selected_lng);
+                }
+
+                let mut r2 = None;
+                if index.is_geojson_filtering_enabled(rtxn)? {
+                    let polygon = geo_types::Polygon::new(
+                        geo_types::LineString(vec![
+                            geo_types::Coord { x: top_right[1], y: top_right[0] },
+                            geo_types::Coord { x: bottom_left[1], y: top_right[0] },
+                            geo_types::Coord { x: bottom_left[1], y: bottom_left[0] },
+                            geo_types::Coord { x: top_right[1], y: bottom_left[0] },
+                        ]),
+                        Vec::new(),
+                    );
+
+                    let result = index.cellulite.in_shape(rtxn, &polygon)?;
+
+                    r2 = Some(RoaringBitmap::from_iter(result)); // TODO: Remove once we update roaring in meilisearch
+                }
+
+                match (r1, r2) {
+                    (Some(r1), Some(r2)) => Ok(r1 | r2),
+                    (Some(r1), None) => Ok(r1),
+                    (None, Some(r2)) => Ok(r2),
+                    (None, None) => Err(top_right_point[0].as_external_error(
                         FilterError::AttributeNotFilterable {
-                            attribute: RESERVED_GEO_FIELD_NAME,
+                            attribute: &format!(
+                                "{RESERVED_GEO_FIELD_NAME}/{RESERVED_GEOJSON_FIELD_NAME}"
+                            ),
                             filterable_patterns: filtered_matching_patterns(
                                 filterable_attribute_rules,
                                 &|features| features.is_filterable(),
                             ),
                         },
-                    ))?
+                    ))?,
                 }
             }
+            FilterCondition::GeoPolygon { points } => {
+                if !index.is_geojson_filtering_enabled(rtxn)? {
+                    return Err(points[0][0].as_external_error(
+                        FilterError::AttributeNotFilterable {
+                            attribute: RESERVED_GEOJSON_FIELD_NAME,
+                            filterable_patterns: filtered_matching_patterns(
+                                filterable_attribute_rules,
+                                &|features| features.is_filterable(),
+                            ),
+                        },
+                    ))?;
+                }
+
+                let mut coords = Vec::new();
+                for [lat_token, lng_token] in points {
+                    let lat = lat_token.parse_finite_float()?;
+                    let lng = lng_token.parse_finite_float()?;
+                    if !(-90.0..=90.0).contains(&lat) {
+                        return Err(lat_token.as_external_error(BadGeoError::Lat(lat)))?;
+                    }
+                    if !(-180.0..=180.0).contains(&lng) {
+                        return Err(lng_token.as_external_error(BadGeoError::Lng(lng)))?;
+                    }
+                    coords.push(geo_types::Coord { x: lng, y: lat });
+                }
+
+                let polygon = geo_types::Polygon::new(geo_types::LineString(coords), Vec::new());
+                let result = index.cellulite.in_shape(rtxn, &polygon)?;
+
+                let result = roaring::RoaringBitmap::from_iter(result); // TODO: Remove once we update roaring in meilisearch
+
+                Ok(result)
+            }
         }
     }
 }
@@ -962,17 +1060,17 @@ mod tests {
         let rtxn = index.read_txn().unwrap();
         let filter = Filter::from_str("_geoRadius(42, 150, 10)").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        snapshot!(error.to_string(), @r###"
-        Attribute `_geo` is not filterable. This index does not have configured filterable attributes.
+        snapshot!(error.to_string(), @r"
+        Attribute `_geo/_geojson` is not filterable. This index does not have configured filterable attributes.
         12:14 _geoRadius(42, 150, 10)
-        "###);
+        ");
 
         let filter = Filter::from_str("_geoBoundingBox([42, 150], [30, 10])").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        snapshot!(error.to_string(), @r###"
-        Attribute `_geo` is not filterable. This index does not have configured filterable attributes.
+        snapshot!(error.to_string(), @r"
+        Attribute `_geo/_geojson` is not filterable. This index does not have configured filterable attributes.
         18:20 _geoBoundingBox([42, 150], [30, 10])
-        "###);
+        ");
 
         let filter = Filter::from_str("dog = \"bernese mountain\"").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
@@ -993,19 +1091,19 @@ mod tests {
 
         let rtxn = index.read_txn().unwrap();
 
-        let filter = Filter::from_str("_geoRadius(-100, 150, 10)").unwrap().unwrap();
+        let filter = Filter::from_str("_geoRadius(-90, 150, 10)").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        snapshot!(error.to_string(), @r###"
-        Attribute `_geo` is not filterable. Available filterable attribute patterns are: `title`.
-        12:16 _geoRadius(-100, 150, 10)
-        "###);
+        snapshot!(error.to_string(), @r"
+        Attribute `_geo/_geojson` is not filterable. Available filterable attribute patterns are: `title`.
+        12:15 _geoRadius(-90, 150, 10)
+        ");
 
         let filter = Filter::from_str("_geoBoundingBox([42, 150], [30, 10])").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        snapshot!(error.to_string(), @r###"
-        Attribute `_geo` is not filterable. Available filterable attribute patterns are: `title`.
+        snapshot!(error.to_string(), @r"
+        Attribute `_geo/_geojson` is not filterable. Available filterable attribute patterns are: `title`.
         18:20 _geoBoundingBox([42, 150], [30, 10])
-        "###);
+        ");
 
         let filter = Filter::from_str("name = 12").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
@@ -1153,38 +1251,34 @@ mod tests {
         // georadius have a bad latitude
         let filter = Filter::from_str("_geoRadius(-100, 150, 10)").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(
-            error.to_string().starts_with(
-                "Bad latitude `-100`. Latitude must be contained between -90 and 90 degrees."
-            ),
-            "{}",
-            error.to_string()
-        );
+        snapshot!(error.to_string(), @r"
+        Bad latitude `-100`. Latitude must be contained between -90 and 90 degrees.
+        12:16 _geoRadius(-100, 150, 10)
+        ");
 
         // georadius have a bad latitude
         let filter = Filter::from_str("_geoRadius(-90.0000001, 150, 10)").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(error.to_string().contains(
-            "Bad latitude `-90.0000001`. Latitude must be contained between -90 and 90 degrees."
-        ));
+        snapshot!(error.to_string(), @r"
+        Bad latitude `-90.0000001`. Latitude must be contained between -90 and 90 degrees.
+        12:23 _geoRadius(-90.0000001, 150, 10)
+        ");
 
         // georadius have a bad longitude
         let filter = Filter::from_str("_geoRadius(-10, 250, 10)").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(
-            error.to_string().contains(
-                "Bad longitude `250`. Longitude must be contained between -180 and 180 degrees."
-            ),
-            "{}",
-            error.to_string(),
-        );
+        snapshot!(error.to_string(), @r"
+        Bad longitude `250`. Longitude must be contained between -180 and 180 degrees. Hint: try using `-110` instead.
+        17:20 _geoRadius(-10, 250, 10)
+        ");
 
         // georadius have a bad longitude
         let filter = Filter::from_str("_geoRadius(-10, 180.000001, 10)").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(error.to_string().contains(
-            "Bad longitude `180.000001`. Longitude must be contained between -180 and 180 degrees."
-        ));
+        snapshot!(error.to_string(), @r"
+        Bad longitude `180.000001`. Longitude must be contained between -180 and 180 degrees. Hint: try using `-179.999999` instead.
+        17:27 _geoRadius(-10, 180.000001, 10)
+        ");
     }
 
     #[test]
@@ -1207,73 +1301,73 @@ mod tests {
         let filter =
             Filter::from_str("_geoBoundingBox([-90.0000001, 150], [30, 10])").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(
-            error.to_string().starts_with(
-                "Bad latitude `-90.0000001`. Latitude must be contained between -90 and 90 degrees."
-            ),
-            "{}",
-            error.to_string()
-        );
+        snapshot!(error.to_string(), @r"
+        Bad latitude `-90.0000001`. Latitude must be contained between -90 and 90 degrees.
+        18:29 _geoBoundingBox([-90.0000001, 150], [30, 10])
+        ");
 
         // geoboundingbox top left coord have a bad latitude
         let filter =
             Filter::from_str("_geoBoundingBox([90.0000001, 150], [30, 10])").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(
-            error.to_string().starts_with(
-                "Bad latitude `90.0000001`. Latitude must be contained between -90 and 90 degrees."
-            ),
-            "{}",
-            error.to_string()
-        );
+        snapshot!(error.to_string(), @r"
+        Bad latitude `90.0000001`. Latitude must be contained between -90 and 90 degrees.
+        18:28 _geoBoundingBox([90.0000001, 150], [30, 10])
+        ");
 
         // geoboundingbox bottom right coord have a bad latitude
         let filter =
             Filter::from_str("_geoBoundingBox([30, 10], [-90.0000001, 150])").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(error.to_string().contains(
-            "Bad latitude `-90.0000001`. Latitude must be contained between -90 and 90 degrees."
-        ));
+        snapshot!(error.to_string(), @r"
+        Bad latitude `-90.0000001`. Latitude must be contained between -90 and 90 degrees.
+        28:39 _geoBoundingBox([30, 10], [-90.0000001, 150])
+        ");
 
         // geoboundingbox bottom right coord have a bad latitude
         let filter =
             Filter::from_str("_geoBoundingBox([30, 10], [90.0000001, 150])").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(error.to_string().contains(
-            "Bad latitude `90.0000001`. Latitude must be contained between -90 and 90 degrees."
-        ));
+        snapshot!(error.to_string(), @r"
+        Bad latitude `90.0000001`. Latitude must be contained between -90 and 90 degrees.
+        28:38 _geoBoundingBox([30, 10], [90.0000001, 150])
+        ");
 
         // geoboundingbox top left coord have a bad longitude
         let filter =
             Filter::from_str("_geoBoundingBox([-10, 180.000001], [30, 10])").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(error.to_string().contains(
-            "Bad longitude `180.000001`. Longitude must be contained between -180 and 180 degrees."
-        ));
+        snapshot!(error.to_string(), @r"
+        Bad longitude `180.000001`. Longitude must be contained between -180 and 180 degrees. Hint: try using `-179.999999` instead.
+        23:33 _geoBoundingBox([-10, 180.000001], [30, 10])
+        ");
 
         // geoboundingbox top left coord have a bad longitude
         let filter =
             Filter::from_str("_geoBoundingBox([-10, -180.000001], [30, 10])").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(error.to_string().contains(
-            "Bad longitude `-180.000001`. Longitude must be contained between -180 and 180 degrees."
-        ));
+        snapshot!(error.to_string(), @r"
+        Bad longitude `-180.000001`. Longitude must be contained between -180 and 180 degrees. Hint: try using `179.999999` instead.
+        23:34 _geoBoundingBox([-10, -180.000001], [30, 10])
+        ");
 
         // geoboundingbox bottom right coord have a bad longitude
         let filter =
             Filter::from_str("_geoBoundingBox([30, 10], [-10, -180.000001])").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(error.to_string().contains(
-            "Bad longitude `-180.000001`. Longitude must be contained between -180 and 180 degrees."
-        ));
+        snapshot!(error.to_string(), @r"
+        Bad longitude `-180.000001`. Longitude must be contained between -180 and 180 degrees. Hint: try using `179.999999` instead.
+        33:44 _geoBoundingBox([30, 10], [-10, -180.000001])
+        ");
 
         // geoboundingbox bottom right coord have a bad longitude
         let filter =
             Filter::from_str("_geoBoundingBox([30, 10], [-10, 180.000001])").unwrap().unwrap();
         let error = filter.evaluate(&rtxn, &index).unwrap_err();
-        assert!(error.to_string().contains(
-            "Bad longitude `180.000001`. Longitude must be contained between -180 and 180 degrees."
-        ));
+        snapshot!(error.to_string(), @r"
+        Bad longitude `180.000001`. Longitude must be contained between -180 and 180 degrees. Hint: try using `-179.999999` instead.
+        33:43 _geoBoundingBox([30, 10], [-10, 180.000001])
+        ");
     }
 
     #[test]

crates/milli/src/update/clear_documents.rs

@@ -48,6 +48,7 @@ impl<'t, 'i> ClearDocuments<'t, 'i> {
             field_id_docid_facet_strings,
             vector_store,
             embedder_category_id: _,
+            cellulite,
             documents,
         } = self.index;
 
@@ -90,6 +91,7 @@ impl<'t, 'i> ClearDocuments<'t, 'i> {
         field_id_docid_facet_strings.clear(self.wtxn)?;
         // vector
         vector_store.clear(self.wtxn)?;
+        cellulite.clear(self.wtxn)?;
 
         documents.clear(self.wtxn)?;
 

crates/milli/src/update/index_documents/extract/extract_geo_points.rs

@@ -2,6 +2,7 @@ use std::fs::File;
 use std::io::{self, BufReader};
 
 use concat_arrays::concat_arrays;
+use geojson::GeoJson;
 use serde_json::Value;
 
 use super::helpers::{create_writer, writer_into_reader, GrenadParameters};
@@ -9,7 +10,7 @@ use crate::error::GeoError;
 use crate::update::del_add::{DelAdd, KvReaderDelAdd, KvWriterDelAdd};
 use crate::update::index_documents::extract_finite_float_from_value;
 use crate::update::settings::{InnerIndexSettings, InnerIndexSettingsDiff};
-use crate::{FieldId, InternalError, Result};
+use crate::{DocumentId, FieldId, InternalError, Result, UserError};
 
 /// Extracts the geographical coordinates contained in each document under the `_geo` field.
 ///
@@ -107,3 +108,72 @@ fn extract_lat_lng(
         None => Ok(None),
     }
 }
+
+/// Extracts the geographical coordinates contained in each document under the `_geojson` field.
+///
+/// Returns the generated grenad reader containing the docid as key associated to its zerometry
+#[tracing::instrument(level = "trace", skip_all, target = "indexing::extract")]
+pub fn extract_geojson<R: io::Read + io::Seek>(
+    obkv_documents: grenad::Reader<R>,
+    indexer: GrenadParameters,
+    primary_key_id: FieldId,
+    settings_diff: &InnerIndexSettingsDiff,
+) -> Result<grenad::Reader<BufReader<File>>> {
+    let mut writer = create_writer(
+        indexer.chunk_compression_type,
+        indexer.chunk_compression_level,
+        tempfile::tempfile()?,
+    );
+
+    let mut cursor = obkv_documents.into_cursor()?;
+    while let Some((docid_bytes, value)) = cursor.move_on_next()? {
+        let obkv = obkv::KvReader::from_slice(value);
+        // since we only need the primary key when we throw an error
+        // we create this getter to lazily get it when needed
+        let document_id = || -> Value {
+            let reader = KvReaderDelAdd::from_slice(obkv.get(primary_key_id).unwrap());
+            let document_id =
+                reader.get(DelAdd::Deletion).or(reader.get(DelAdd::Addition)).unwrap();
+            serde_json::from_slice(document_id).unwrap()
+        };
+
+        // extract old version
+        let del_geojson =
+            extract_geojson_field(obkv, &settings_diff.old, DelAdd::Deletion, document_id)?;
+        // extract new version
+        let add_geojson =
+            extract_geojson_field(obkv, &settings_diff.new, DelAdd::Addition, document_id)?;
+
+        if del_geojson != add_geojson {
+            let mut obkv = KvWriterDelAdd::memory();
+            if del_geojson.is_some() {
+                // We don't need to store the geojson, we'll just delete it by id
+                obkv.insert(DelAdd::Deletion, [])?;
+            }
+            if let Some(geojson) = add_geojson {
+                obkv.insert(DelAdd::Addition, geojson.to_string().as_bytes())?;
+            }
+            let bytes = obkv.into_inner()?;
+            writer.insert(&docid_bytes[0..std::mem::size_of::<DocumentId>()], bytes)?;
+        }
+    }
+
+    writer_into_reader(writer)
+}
+
+fn extract_geojson_field(
+    obkv: &obkv::KvReader<FieldId>,
+    settings: &InnerIndexSettings,
+    deladd: DelAdd,
+    _document_id: impl Fn() -> Value,
+) -> Result<Option<GeoJson>> {
+    match settings.geojson_fid {
+        Some(fid) if settings.filterable_attributes_rules.iter().any(|rule| rule.has_geojson()) => {
+            let value = obkv.get(fid).map(KvReaderDelAdd::from_slice).and_then(|r| r.get(deladd));
+            Ok(value
+                .map(|v| GeoJson::from_reader(v).map_err(UserError::MalformedGeojson))
+                .transpose()?)
+        }
+        _ => Ok(None),
+    }
+}

crates/milli/src/update/index_documents/extract/mod.rs

@@ -31,6 +31,7 @@ use self::extract_word_position_docids::extract_word_position_docids;
 use super::helpers::{as_cloneable_grenad, CursorClonableMmap, GrenadParameters};
 use super::{helpers, TypedChunk};
 use crate::progress::EmbedderStats;
+use crate::update::index_documents::extract::extract_geo_points::extract_geojson;
 use crate::update::index_documents::extract::extract_vector_points::extract_embeddings_from_fragments;
 use crate::update::settings::InnerIndexSettingsDiff;
 use crate::vector::db::EmbedderInfo;
@@ -62,6 +63,7 @@ pub(crate) fn data_from_obkv_documents(
                         original_documents_chunk,
                         indexer,
                         lmdb_writer_sx.clone(),
+                        primary_key_id,
                         settings_diff.clone(),
                         embedder_info.clone(),
                         possible_embedding_mistakes.clone(),
@@ -228,10 +230,12 @@ pub fn request_threads() -> &'static ThreadPoolNoAbort {
 
 /// Extract chunked data and send it into lmdb_writer_sx sender:
 /// - documents
+#[allow(clippy::too_many_arguments)]
 fn send_original_documents_data(
     original_documents_chunk: Result<grenad::Reader<BufReader<File>>>,
     indexer: GrenadParameters,
     lmdb_writer_sx: Sender<Result<TypedChunk>>,
+    primary_key_id: FieldId,
     settings_diff: Arc<InnerIndexSettingsDiff>,
     embedder_info: Arc<Vec<(String, EmbedderInfo)>>,
     possible_embedding_mistakes: Arc<PossibleEmbeddingMistakes>,
@@ -240,6 +244,20 @@ fn send_original_documents_data(
     let original_documents_chunk =
         original_documents_chunk.and_then(|c| unsafe { as_cloneable_grenad(&c) })?;
 
+    if settings_diff.reindex_geojson() {
+        let documents_chunk_cloned = original_documents_chunk.clone();
+        let lmdb_writer_sx_cloned = lmdb_writer_sx.clone();
+        let settings_diff = settings_diff.clone();
+        rayon::spawn(move || {
+            let result =
+                extract_geojson(documents_chunk_cloned, indexer, primary_key_id, &settings_diff);
+            let _ = match result {
+                Ok(geojson) => lmdb_writer_sx_cloned.send(Ok(TypedChunk::GeoJson(geojson))),
+                Err(error) => lmdb_writer_sx_cloned.send(Err(error)),
+            };
+        });
+    }
+
     let index_vectors = (settings_diff.reindex_vectors() || !settings_diff.settings_update_only())
         // no point in indexing vectors without embedders
         && (!settings_diff.new.runtime_embedders.inner_as_ref().is_empty());

crates/milli/src/update/index_documents/mod.rs

@@ -523,7 +523,7 @@ where
                 .is_some_and(|conf| conf.is_quantized);
             let is_quantizing = embedder_config.is_some_and(|action| action.is_being_quantized);
 
-            pool.install(|| {
+            pool.install(|| -> Result<_> {
                 let mut writer =
                     VectorStore::new(backend, vector_store, embedder_index, was_quantized);
                 writer.build_and_quantize(
@@ -541,6 +541,8 @@ where
             .map_err(InternalError::from)??;
         }
 
+        self.index.cellulite.build(self.wtxn, &self.should_abort, &Progress::default())?;
+
         self.execute_prefix_databases(
             word_docids.map(MergerBuilder::build),
             exact_word_docids.map(MergerBuilder::build),

crates/milli/src/update/index_documents/transform.rs

@@ -820,19 +820,20 @@ impl<'a, 'i> Transform<'a, 'i> {
         let documents_count = documents_ids.len() as usize;
 
         // We initialize the sorter with the user indexing settings.
-        let mut original_sorter = if settings_diff.reindex_vectors() {
-            Some(create_sorter(
-                grenad::SortAlgorithm::Stable,
-                KeepFirst,
-                self.indexer_settings.chunk_compression_type,
-                self.indexer_settings.chunk_compression_level,
-                self.indexer_settings.max_nb_chunks,
-                self.indexer_settings.max_memory.map(|mem| mem / 2),
-                true,
-            ))
-        } else {
-            None
-        };
+        let mut original_sorter =
+            if settings_diff.reindex_vectors() || settings_diff.reindex_geojson() {
+                Some(create_sorter(
+                    grenad::SortAlgorithm::Stable,
+                    KeepFirst,
+                    self.indexer_settings.chunk_compression_type,
+                    self.indexer_settings.chunk_compression_level,
+                    self.indexer_settings.max_nb_chunks,
+                    self.indexer_settings.max_memory.map(|mem| mem / 2),
+                    true,
+                ))
+            } else {
+                None
+            };
 
         let backend = self.index.get_vector_store(wtxn)?.unwrap_or_default();
         let readers: BTreeMap<&str, (VectorStore, &RoaringBitmap)> = settings_diff

crates/milli/src/update/index_documents/typed_chunk.rs

@@ -30,7 +30,7 @@ use crate::vector::db::{EmbeddingStatusDelta, IndexEmbeddingConfig};
 use crate::vector::VectorStore;
 use crate::{
     lat_lng_to_xyz, CboRoaringBitmapCodec, DocumentId, FieldId, GeoPoint, Index, InternalError,
-    Result, SerializationError, U8StrStrCodec,
+    Result, SerializationError, U8StrStrCodec, UserError,
 };
 
 /// This struct accumulates and group the TypedChunks
@@ -85,6 +85,7 @@ pub(crate) enum TypedChunk {
     FieldIdFacetIsNullDocids(grenad::Reader<BufReader<File>>),
     FieldIdFacetIsEmptyDocids(grenad::Reader<BufReader<File>>),
     GeoPoints(grenad::Reader<BufReader<File>>),
+    GeoJson(grenad::Reader<BufReader<File>>),
     VectorPoints {
         remove_vectors: grenad::Reader<BufReader<File>>,
         // docid -> vector
@@ -614,6 +615,36 @@ pub(crate) fn write_typed_chunk_into_index(
             index.put_geo_rtree(wtxn, &rtree)?;
             index.put_geo_faceted_documents_ids(wtxn, &geo_faceted_docids)?;
         }
+        TypedChunk::GeoJson(_) => {
+            let span = tracing::trace_span!(target: "indexing::write_db", "geo_json");
+            let _entered = span.enter();
+
+            let mut builder = MergerBuilder::new(KeepFirst);
+            for typed_chunk in typed_chunks {
+                let TypedChunk::GeoJson(chunk) = typed_chunk else {
+                    unreachable!();
+                };
+
+                builder.push(chunk.into_cursor()?);
+            }
+            let merger = builder.build();
+
+            let mut iter = merger.into_stream_merger_iter()?;
+            while let Some((key, value)) = iter.next()? {
+                // convert the key back to a u32 (4 bytes)
+                let docid = key.try_into().map(DocumentId::from_be_bytes).unwrap();
+
+                let deladd_obkv = KvReaderDelAdd::from_slice(value);
+                if let Some(_value) = deladd_obkv.get(DelAdd::Deletion) {
+                    index.cellulite.delete(wtxn, docid)?;
+                }
+                if let Some(value) = deladd_obkv.get(DelAdd::Addition) {
+                    let geojson =
+                        geojson::GeoJson::from_reader(value).map_err(UserError::SerdeJson)?;
+                    index.cellulite.add(wtxn, docid, &geojson)?;
+                }
+            }
+        }
         TypedChunk::VectorPoints { .. } => {
             let span = tracing::trace_span!(target: "indexing::write_db", "vector_points");
             let _entered = span.enter();

crates/milli/src/update/new/channel.rs

@@ -139,6 +139,7 @@ pub enum ReceiverAction {
     LargeEntry(LargeEntry),
     LargeVectors(LargeVectors),
     LargeVector(LargeVector),
+    LargeGeoJson(LargeGeoJson),
 }
 
 /// An entry that cannot fit in the BBQueue buffers has been
@@ -193,6 +194,14 @@ impl LargeVector {
     }
 }
 
+#[derive(Debug)]
+pub struct LargeGeoJson {
+    /// The document id associated to the large geojson.
+    pub docid: DocumentId,
+    /// The large geojson that must be written.
+    pub geojson: Mmap,
+}
+
 impl<'a> WriterBbqueueReceiver<'a> {
     /// Tries to receive an action to do until the timeout occurs
     /// and if it does, consider it as a spurious wake up.
@@ -258,10 +267,12 @@ pub enum EntryHeader {
     DeleteVector(DeleteVector),
     SetVectors(SetVectors),
     SetVector(SetVector),
+    CelluliteItem(DocumentId),
+    CelluliteRemove(DocumentId),
 }
 
 impl EntryHeader {
-    const fn variant_size() -> usize {
+    pub const fn variant_size() -> usize {
         mem::size_of::<u8>()
     }
 
@@ -271,6 +282,8 @@ impl EntryHeader {
             EntryHeader::DeleteVector(_) => 1,
             EntryHeader::SetVectors(_) => 2,
             EntryHeader::SetVector(_) => 3,
+            EntryHeader::CelluliteItem(_) => 4,
+            EntryHeader::CelluliteRemove(_) => 5,
         }
     }
 
@@ -289,6 +302,14 @@ impl EntryHeader {
         Self::variant_size() + mem::size_of::<DeleteVector>()
     }
 
+    const fn total_cellulite_item_size(value_length: usize) -> usize {
+        Self::variant_size() + mem::size_of::<DocumentId>() + value_length
+    }
+
+    const fn total_cellulite_remove_size() -> usize {
+        Self::variant_size() + mem::size_of::<DocumentId>()
+    }
+
     /// The `dimensions` corresponds to the number of `f32` in the embedding.
     fn total_set_vectors_size(count: usize, dimensions: usize) -> usize {
         let embedding_size = dimensions * mem::size_of::<f32>();
@@ -306,6 +327,8 @@ impl EntryHeader {
             EntryHeader::DeleteVector(adv) => mem::size_of_val(adv),
             EntryHeader::SetVectors(asvs) => mem::size_of_val(asvs),
             EntryHeader::SetVector(asv) => mem::size_of_val(asv),
+            EntryHeader::CelluliteItem(docid) => mem::size_of_val(docid),
+            EntryHeader::CelluliteRemove(docid) => mem::size_of_val(docid),
         };
         Self::variant_size() + payload_size
     }
@@ -333,6 +356,16 @@ impl EntryHeader {
                 let header = checked::pod_read_unaligned(header_bytes);
                 EntryHeader::SetVector(header)
             }
+            4 => {
+                let header_bytes = &remaining[..mem::size_of::<DocumentId>()];
+                let header = checked::pod_read_unaligned(header_bytes);
+                EntryHeader::CelluliteItem(header)
+            }
+            5 => {
+                let header_bytes = &remaining[..mem::size_of::<DocumentId>()];
+                let header = checked::pod_read_unaligned(header_bytes);
+                EntryHeader::CelluliteRemove(header)
+            }
             id => panic!("invalid variant id: {id}"),
         }
     }
@@ -344,6 +377,8 @@ impl EntryHeader {
             EntryHeader::DeleteVector(adv) => bytemuck::bytes_of(adv),
             EntryHeader::SetVectors(asvs) => bytemuck::bytes_of(asvs),
             EntryHeader::SetVector(asv) => bytemuck::bytes_of(asv),
+            EntryHeader::CelluliteItem(docid) => bytemuck::bytes_of(docid),
+            EntryHeader::CelluliteRemove(docid) => bytemuck::bytes_of(docid),
         };
         *first = self.variant_id();
         remaining.copy_from_slice(payload_bytes);
@@ -548,6 +583,10 @@ impl<'b> ExtractorBbqueueSender<'b> {
         GeoSender(self)
     }
 
+    pub fn geojson<'a>(&'a self) -> GeoJsonSender<'a, 'b> {
+        GeoJsonSender(self)
+    }
+
     fn delete_vector(&self, docid: DocumentId) -> crate::Result<()> {
         let max_grant = self.max_grant;
         let refcell = self.producers.get().unwrap();
@@ -1139,3 +1178,72 @@ impl GeoSender<'_, '_> {
         )
     }
 }
+
+#[derive(Clone, Copy)]
+pub struct GeoJsonSender<'a, 'b>(&'a ExtractorBbqueueSender<'b>);
+
+impl GeoJsonSender<'_, '_> {
+    pub fn send_geojson(&self, docid: DocumentId, value: Vec<u8>) -> crate::Result<()> {
+        let max_grant = self.0.max_grant;
+        let refcell = self.0.producers.get().unwrap();
+        let mut producer = refcell.0.borrow_mut_or_yield();
+
+        let payload_header = EntryHeader::CelluliteItem(docid);
+        let value_length = value.len();
+        let total_length = EntryHeader::total_cellulite_item_size(value_length);
+        if total_length > max_grant {
+            let mut value_file = tempfile::tempfile().map(BufWriter::new)?;
+
+            let mut embedding_bytes = bytemuck::cast_slice(&value);
+            io::copy(&mut embedding_bytes, &mut value_file)?;
+
+            let value_file = value_file.into_inner().map_err(|ie| ie.into_error())?;
+            let geojson = unsafe { Mmap::map(&value_file)? }; // Safe because the file is never modified
+
+            let large_geojson = LargeGeoJson { docid, geojson };
+            self.0.sender.send(ReceiverAction::LargeGeoJson(large_geojson)).unwrap();
+
+            return Ok(());
+        }
+
+        // Spin loop to have a frame the size we requested.
+        reserve_and_write_grant(
+            &mut producer,
+            total_length,
+            &self.0.sender,
+            &self.0.sent_messages_attempts,
+            &self.0.blocking_sent_messages_attempts,
+            |grant| {
+                let header_size = payload_header.header_size();
+                let (header_bytes, remaining) = grant.split_at_mut(header_size);
+                payload_header.serialize_into(header_bytes);
+                remaining.copy_from_slice(&value);
+                Ok(())
+            },
+        )?;
+
+        Ok(())
+    }
+
+    pub fn delete_geojson(&self, docid: DocumentId) -> crate::Result<()> {
+        let refcell = self.0.producers.get().unwrap();
+        let mut producer = refcell.0.borrow_mut_or_yield();
+
+        let payload_header = EntryHeader::CelluliteRemove(docid);
+        let total_length = EntryHeader::total_cellulite_remove_size();
+
+        reserve_and_write_grant(
+            &mut producer,
+            total_length,
+            &self.0.sender,
+            &self.0.sent_messages_attempts,
+            &self.0.blocking_sent_messages_attempts,
+            |grant| {
+                payload_header.serialize_into(grant);
+                Ok(())
+            },
+        )?;
+
+        Ok(())
+    }
+}

crates/milli/src/update/new/document.rs

@@ -10,7 +10,9 @@ use serde_json::value::RawValue;
 
 use super::vector_document::VectorDocument;
 use super::{KvReaderFieldId, KvWriterFieldId};
-use crate::constants::{RESERVED_GEO_FIELD_NAME, RESERVED_VECTORS_FIELD_NAME};
+use crate::constants::{
+    RESERVED_GEOJSON_FIELD_NAME, RESERVED_GEO_FIELD_NAME, RESERVED_VECTORS_FIELD_NAME,
+};
 use crate::documents::FieldIdMapper;
 use crate::update::del_add::KvReaderDelAdd;
 use crate::update::new::thread_local::{FullySend, MostlySend, ThreadLocal};
@@ -29,10 +31,10 @@ pub trait Document<'doc> {
     /// Iterate over all **top-level** fields of the document, returning their name and raw JSON value.
     ///
     /// - The returned values *may* contain nested fields.
-    /// - The `_vectors` and `_geo` fields are **ignored** by this method, meaning  they are **not returned** by this method.
+    /// - The `_vectors`, `_geo` and `_geojson` fields are **ignored** by this method, meaning  they are **not returned** by this method.
     fn iter_top_level_fields(&self) -> impl Iterator<Item = Result<(&'doc str, &'doc RawValue)>>;
 
-    /// Number of top level fields, **excluding** `_vectors` and `_geo`
+    /// Number of top level fields, **excluding** `_vectors`, `_geo` and `_geojson`.
     fn top_level_fields_count(&self) -> usize;
 
     /// Get the **top-level** with the specified name, if exists.
@@ -50,11 +52,13 @@ pub trait Document<'doc> {
 
     /// Returns the unparsed value of the `_geo` field from the document data.
     ///
-    /// This field alone is insufficient to retrieve geo data, as they may be stored in a dedicated location in the database.
-    /// Use a [`super::geo_document::GeoDocument`] to access the vector.
-    ///
     /// This method is meant as a convenience for implementors of [`super::geo_document::GeoDocument`].
     fn geo_field(&self) -> Result<Option<&'doc RawValue>>;
+
+    /// Returns the unparsed value of the `_geojson` field from the document data.
+    ///
+    /// This method is meant as a convenience for implementors of [`super::geo_document::GeoDocument`].
+    fn geojson_field(&self) -> Result<Option<&'doc RawValue>>;
 }
 
 #[derive(Debug)]
@@ -90,7 +94,10 @@ impl<'t, Mapper: FieldIdMapper> Document<'t> for DocumentFromDb<'t, Mapper> {
                 Err(error) => return Some(Err(error.into())),
             };
 
-            if name == RESERVED_VECTORS_FIELD_NAME || name == RESERVED_GEO_FIELD_NAME {
+            if name == RESERVED_VECTORS_FIELD_NAME
+                || name == RESERVED_GEO_FIELD_NAME
+                || name == RESERVED_GEOJSON_FIELD_NAME
+            {
                 continue;
             }
 
@@ -113,19 +120,24 @@ impl<'t, Mapper: FieldIdMapper> Document<'t> for DocumentFromDb<'t, Mapper> {
         self.field(RESERVED_GEO_FIELD_NAME)
     }
 
+    fn geojson_field(&self) -> Result<Option<&'t RawValue>> {
+        self.field(RESERVED_GEOJSON_FIELD_NAME)
+    }
+
     fn top_level_fields_count(&self) -> usize {
         let has_vectors_field = self.vectors_field().unwrap_or(None).is_some();
         let has_geo_field = self.geo_field().unwrap_or(None).is_some();
+        let has_geojson_field = self.geojson_field().unwrap_or(None).is_some();
         let count = self.content.iter().count();
-        match (has_vectors_field, has_geo_field) {
-            (true, true) => count - 2,
-            (true, false) | (false, true) => count - 1,
-            (false, false) => count,
-        }
+
+        count - has_vectors_field as usize - has_geo_field as usize - has_geojson_field as usize
     }
 
     fn top_level_field(&self, k: &str) -> Result<Option<&'t RawValue>> {
-        if k == RESERVED_VECTORS_FIELD_NAME || k == RESERVED_GEO_FIELD_NAME {
+        if k == RESERVED_VECTORS_FIELD_NAME
+            || k == RESERVED_GEO_FIELD_NAME
+            || k == RESERVED_GEOJSON_FIELD_NAME
+        {
             return Ok(None);
         }
         self.field(k)
@@ -177,15 +189,16 @@ impl<'doc> Document<'doc> for DocumentFromVersions<'_, 'doc> {
         Ok(self.versions.geo_field())
     }
 
+    fn geojson_field(&self) -> Result<Option<&'doc RawValue>> {
+        Ok(self.versions.geojson_field())
+    }
+
     fn top_level_fields_count(&self) -> usize {
         let has_vectors_field = self.vectors_field().unwrap_or(None).is_some();
         let has_geo_field = self.geo_field().unwrap_or(None).is_some();
+        let has_geojson_field = self.geojson_field().unwrap_or(None).is_some();
         let count = self.versions.len();
-        match (has_vectors_field, has_geo_field) {
-            (true, true) => count - 2,
-            (true, false) | (false, true) => count - 1,
-            (false, false) => count,
-        }
+        count - has_vectors_field as usize - has_geo_field as usize - has_geojson_field as usize
     }
 
     fn top_level_field(&self, k: &str) -> Result<Option<&'doc RawValue>> {
@@ -265,6 +278,16 @@ impl<'d, 'doc: 'd, 't: 'd, Mapper: FieldIdMapper> Document<'d>
         db.geo_field()
     }
 
+    fn geojson_field(&self) -> Result<Option<&'d RawValue>> {
+        if let Some(geojson) = self.new_doc.geojson_field()? {
+            return Ok(Some(geojson));
+        }
+
+        let Some(db) = self.db else { return Ok(None) };
+
+        db.geojson_field()
+    }
+
     fn top_level_fields_count(&self) -> usize {
         self.iter_top_level_fields().count()
     }
@@ -296,6 +319,10 @@ where
         D::geo_field(self)
     }
 
+    fn geojson_field(&self) -> Result<Option<&'doc RawValue>> {
+        D::geojson_field(self)
+    }
+
     fn top_level_fields_count(&self) -> usize {
         D::top_level_fields_count(self)
     }
@@ -405,6 +432,13 @@ where
         unordered_field_buffer.push((fid, geo_value));
     }
 
+    if let Some(geojson_value) = document.geojson_field()? {
+        let fid = fields_ids_map
+            .id_or_insert(RESERVED_GEOJSON_FIELD_NAME)
+            .ok_or(UserError::AttributeLimitReached)?;
+        unordered_field_buffer.push((fid, geojson_value));
+    }
+
     unordered_field_buffer.sort_by_key(|(fid, _)| *fid);
     for (fid, value) in unordered_field_buffer.iter() {
         writer.insert(*fid, value.get().as_bytes()).unwrap();
@@ -441,9 +475,11 @@ impl<'doc> Versions<'doc> {
     }
 
     pub fn iter_top_level_fields(&self) -> impl Iterator<Item = (&'doc str, &'doc RawValue)> + '_ {
-        self.data
-            .iter()
-            .filter(|(k, _)| *k != RESERVED_VECTORS_FIELD_NAME && *k != RESERVED_GEO_FIELD_NAME)
+        self.data.iter().filter(|(k, _)| {
+            *k != RESERVED_VECTORS_FIELD_NAME
+                && *k != RESERVED_GEO_FIELD_NAME
+                && *k != RESERVED_GEOJSON_FIELD_NAME
+        })
     }
 
     pub fn vectors_field(&self) -> Option<&'doc RawValue> {
@@ -454,6 +490,10 @@ impl<'doc> Versions<'doc> {
         self.data.get(RESERVED_GEO_FIELD_NAME)
     }
 
+    pub fn geojson_field(&self) -> Option<&'doc RawValue> {
+        self.data.get(RESERVED_GEOJSON_FIELD_NAME)
+    }
+
     pub fn len(&self) -> usize {
         self.data.len()
     }
@@ -463,7 +503,10 @@ impl<'doc> Versions<'doc> {
     }
 
     pub fn top_level_field(&self, k: &str) -> Option<&'doc RawValue> {
-        if k == RESERVED_VECTORS_FIELD_NAME || k == RESERVED_GEO_FIELD_NAME {
+        if k == RESERVED_VECTORS_FIELD_NAME
+            || k == RESERVED_GEO_FIELD_NAME
+            || k == RESERVED_GEOJSON_FIELD_NAME
+        {
             return None;
         }
         self.data.get(k)
@@ -516,7 +559,10 @@ impl<'a, Mapper: FieldIdMapper> Document<'a> for KvDelAddDocument<'a, Mapper> {
                 Err(error) => return Some(Err(error.into())),
             };
 
-            if name == RESERVED_VECTORS_FIELD_NAME || name == RESERVED_GEO_FIELD_NAME {
+            if name == RESERVED_VECTORS_FIELD_NAME
+                || name == RESERVED_GEO_FIELD_NAME
+                || name == RESERVED_GEOJSON_FIELD_NAME
+            {
                 continue;
             }
 
@@ -549,7 +595,10 @@ impl<'a, Mapper: FieldIdMapper> Document<'a> for KvDelAddDocument<'a, Mapper> {
                 Err(_) => return Some(()),
             };
 
-            if name == RESERVED_VECTORS_FIELD_NAME || name == RESERVED_GEO_FIELD_NAME {
+            if name == RESERVED_VECTORS_FIELD_NAME
+                || name == RESERVED_GEO_FIELD_NAME
+                || name == RESERVED_GEOJSON_FIELD_NAME
+            {
                 continue;
             }
 
@@ -559,7 +608,10 @@ impl<'a, Mapper: FieldIdMapper> Document<'a> for KvDelAddDocument<'a, Mapper> {
     }
 
     fn top_level_field(&self, k: &str) -> Result<Option<&'a RawValue>> {
-        if k == RESERVED_VECTORS_FIELD_NAME || k == RESERVED_GEO_FIELD_NAME {
+        if k == RESERVED_VECTORS_FIELD_NAME
+            || k == RESERVED_GEO_FIELD_NAME
+            || k == RESERVED_GEOJSON_FIELD_NAME
+        {
             return Ok(None);
         }
         self.get(k)
@@ -572,6 +624,10 @@ impl<'a, Mapper: FieldIdMapper> Document<'a> for KvDelAddDocument<'a, Mapper> {
     fn geo_field(&self) -> Result<Option<&'a RawValue>> {
         self.get(RESERVED_GEO_FIELD_NAME)
     }
+
+    fn geojson_field(&self) -> Result<Option<&'a RawValue>> {
+        self.get(RESERVED_GEOJSON_FIELD_NAME)
+    }
 }
 
 pub struct DocumentIdentifiers<'doc> {

crates/milli/src/update/new/extract/documents.rs

@@ -5,7 +5,7 @@ use bumpalo::Bump;
 use hashbrown::HashMap;
 
 use super::DelAddRoaringBitmap;
-use crate::constants::RESERVED_GEO_FIELD_NAME;
+use crate::constants::{RESERVED_GEOJSON_FIELD_NAME, RESERVED_GEO_FIELD_NAME};
 use crate::update::new::channel::{DocumentsSender, ExtractorBbqueueSender};
 use crate::update::new::document::{write_to_obkv, Document, DocumentContext, DocumentIdentifiers};
 use crate::update::new::indexer::document_changes::{Extractor, IndexingContext};
@@ -75,7 +75,11 @@ impl<'extractor> Extractor<'extractor> for DocumentsExtractor<'_, '_> {
                         .geo_field()
                         .transpose()
                         .map(|res| res.map(|rv| (RESERVED_GEO_FIELD_NAME, rv)));
-                    for res in content.iter_top_level_fields().chain(geo_iter) {
+                    let geojson_iter = content
+                        .geojson_field()
+                        .transpose()
+                        .map(|res| res.map(|rv| (RESERVED_GEOJSON_FIELD_NAME, rv)));
+                    for res in content.iter_top_level_fields().chain(geo_iter).chain(geojson_iter) {
                         let (f, _) = res?;
                         let entry = document_extractor_data
                             .field_distribution_delta
@@ -94,7 +98,11 @@ impl<'extractor> Extractor<'extractor> for DocumentsExtractor<'_, '_> {
                         .geo_field()
                         .transpose()
                         .map(|res| res.map(|rv| (RESERVED_GEO_FIELD_NAME, rv)));
-                    for res in content.iter_top_level_fields().chain(geo_iter) {
+                    let geojson_iter = content
+                        .geojson_field()
+                        .transpose()
+                        .map(|res| res.map(|rv| (RESERVED_GEOJSON_FIELD_NAME, rv)));
+                    for res in content.iter_top_level_fields().chain(geo_iter).chain(geojson_iter) {
                         let (f, _) = res?;
                         let entry = document_extractor_data
                             .field_distribution_delta
@@ -108,7 +116,11 @@ impl<'extractor> Extractor<'extractor> for DocumentsExtractor<'_, '_> {
                         .geo_field()
                         .transpose()
                         .map(|res| res.map(|rv| (RESERVED_GEO_FIELD_NAME, rv)));
-                    for res in content.iter_top_level_fields().chain(geo_iter) {
+                    let geojson_iter = content
+                        .geojson_field()
+                        .transpose()
+                        .map(|res| res.map(|rv| (RESERVED_GEOJSON_FIELD_NAME, rv)));
+                    for res in content.iter_top_level_fields().chain(geo_iter).chain(geojson_iter) {
                         let (f, _) = res?;
                         let entry = document_extractor_data
                             .field_distribution_delta
@@ -143,7 +155,11 @@ impl<'extractor> Extractor<'extractor> for DocumentsExtractor<'_, '_> {
                         .geo_field()
                         .transpose()
                         .map(|res| res.map(|rv| (RESERVED_GEO_FIELD_NAME, rv)));
-                    for res in content.iter_top_level_fields().chain(geo_iter) {
+                    let geojson_iter = content
+                        .geojson_field()
+                        .transpose()
+                        .map(|res| res.map(|rv| (RESERVED_GEOJSON_FIELD_NAME, rv)));
+                    for res in content.iter_top_level_fields().chain(geo_iter).chain(geojson_iter) {
                         let (f, _) = res?;
                         let entry = document_extractor_data
                             .field_distribution_delta

crates/milli/src/update/new/extract/geo/cellulite.rs

@@ -0,0 +1,261 @@
+use std::cell::RefCell;
+use std::fs::File;
+use std::io::{BufReader, BufWriter, ErrorKind, Read, Seek as _, Write as _};
+use std::mem;
+use std::str::FromStr;
+
+use bumpalo::Bump;
+use byteorder::{BigEndian, ReadBytesExt, WriteBytesExt};
+use cellulite::zerometry::ZerometryCodec;
+use geo_types::Geometry;
+use geojson::GeoJson;
+use heed::{BytesEncode, RoTxn};
+use zerometry::Zerometry;
+
+use crate::update::new::channel::GeoJsonSender;
+use crate::update::new::document::{Document, DocumentContext};
+use crate::update::new::indexer::document_changes::Extractor;
+use crate::update::new::ref_cell_ext::RefCellExt as _;
+use crate::update::new::thread_local::MostlySend;
+use crate::update::new::DocumentChange;
+use crate::update::GrenadParameters;
+use crate::{DocumentId, Index, InternalError, Result, UserError};
+
+pub struct GeoJsonExtractor {
+    grenad_parameters: GrenadParameters,
+}
+
+impl GeoJsonExtractor {
+    pub fn new(
+        rtxn: &RoTxn,
+        index: &Index,
+        grenad_parameters: GrenadParameters,
+    ) -> Result<Option<Self>> {
+        if index.is_geojson_filtering_enabled(rtxn)? {
+            Ok(Some(GeoJsonExtractor { grenad_parameters }))
+        } else {
+            Ok(None)
+        }
+    }
+}
+
+pub struct GeoJsonExtractorData<'extractor> {
+    /// The set of documents ids that were removed. If a document sees its geo
+    /// point being updated, we first put it in the deleted and then in the inserted.
+    removed: bumpalo::collections::Vec<'extractor, DocumentId>,
+    inserted: bumpalo::collections::Vec<'extractor, (DocumentId, &'extractor [u8])>,
+    /// Contains a packed list of `ExtractedGeoPoint` of the inserted geo points
+    /// data structures if we have spilled to disk.
+    spilled_removed: Option<BufWriter<File>>,
+    /// Contains a packed list of `ExtractedGeoPoint` of the inserted geo points
+    /// data structures if we have spilled to disk.
+    spilled_inserted: Option<BufWriter<File>>,
+}
+
+impl<'extractor> GeoJsonExtractorData<'extractor> {
+    pub fn freeze(self) -> Result<FrozenGeoJsonExtractorData<'extractor>> {
+        let GeoJsonExtractorData { removed, inserted, spilled_removed, spilled_inserted } = self;
+
+        Ok(FrozenGeoJsonExtractorData {
+            removed: removed.into_bump_slice(),
+            inserted: inserted.into_bump_slice(),
+            spilled_removed: spilled_removed
+                .map(|bw| bw.into_inner().map(BufReader::new).map_err(|iie| iie.into_error()))
+                .transpose()?,
+            spilled_inserted: spilled_inserted
+                .map(|bw| bw.into_inner().map(BufReader::new).map_err(|iie| iie.into_error()))
+                .transpose()?,
+        })
+    }
+}
+
+unsafe impl MostlySend for GeoJsonExtractorData<'_> {}
+
+pub struct FrozenGeoJsonExtractorData<'extractor> {
+    pub removed: &'extractor [DocumentId],
+    pub inserted: &'extractor [(DocumentId, &'extractor [u8])],
+    pub spilled_removed: Option<BufReader<File>>,
+    pub spilled_inserted: Option<BufReader<File>>,
+}
+
+impl FrozenGeoJsonExtractorData<'_> {
+    pub fn iter_and_clear_removed(&mut self, channel: GeoJsonSender<'_, '_>) -> Result<()> {
+        for docid in mem::take(&mut self.removed) {
+            channel.delete_geojson(*docid).unwrap();
+        }
+
+        if let Some(mut spilled) = self.spilled_removed.take() {
+            spilled.rewind()?;
+
+            loop {
+                let docid = match spilled.read_u32::<BigEndian>() {
+                    Ok(docid) => docid,
+                    Err(e) if e.kind() == ErrorKind::UnexpectedEof => break,
+                    Err(e) => return Err(InternalError::SerdeJson(serde_json::Error::io(e)).into()),
+                };
+                channel.delete_geojson(docid).unwrap();
+            }
+        }
+
+        Ok(())
+    }
+
+    pub fn iter_and_clear_inserted(&mut self, channel: GeoJsonSender<'_, '_>) -> Result<()> {
+        for (docid, _buf) in mem::take(&mut self.inserted) {
+            channel.send_geojson(*docid, _buf.to_vec()).unwrap();
+        }
+
+        if let Some(mut spilled) = self.spilled_inserted.take() {
+            spilled.rewind()?;
+
+            loop {
+                let docid = match spilled.read_u32::<BigEndian>() {
+                    Ok(docid) => docid,
+                    Err(e) if e.kind() == ErrorKind::UnexpectedEof => break,
+                    Err(e) => return Err(InternalError::SerdeJson(serde_json::Error::io(e)).into()),
+                };
+                let size = match spilled.read_u32::<BigEndian>() {
+                    Ok(size) => size,
+                    Err(e) => return Err(InternalError::SerdeJson(serde_json::Error::io(e)).into()),
+                };
+                let mut buf = vec![0; size as usize];
+                spilled
+                    .read_exact(&mut buf)
+                    .map_err(|e| InternalError::SerdeJson(serde_json::Error::io(e)))?;
+                channel.send_geojson(docid, buf).unwrap();
+            }
+        }
+
+        Ok(())
+    }
+}
+
+impl<'extractor> Extractor<'extractor> for GeoJsonExtractor {
+    type Data = RefCell<GeoJsonExtractorData<'extractor>>;
+
+    fn init_data<'doc>(&'doc self, extractor_alloc: &'extractor Bump) -> Result<Self::Data> {
+        Ok(RefCell::new(GeoJsonExtractorData {
+            removed: bumpalo::collections::Vec::new_in(extractor_alloc),
+            inserted: bumpalo::collections::Vec::new_in(extractor_alloc),
+            spilled_inserted: None,
+            spilled_removed: None,
+        }))
+    }
+
+    fn process<'doc>(
+        &'doc self,
+        changes: impl Iterator<Item = Result<DocumentChange<'doc>>>,
+        context: &'doc DocumentContext<'doc, 'extractor, '_, '_, Self::Data>,
+    ) -> Result<()> {
+        let rtxn = &context.rtxn;
+        let index = context.index;
+        let max_memory = self.grenad_parameters.max_memory_by_thread();
+        let db_fields_ids_map = context.db_fields_ids_map;
+        let mut data_ref = context.data.borrow_mut_or_yield();
+
+        for change in changes {
+            if data_ref.spilled_removed.is_none()
+                && max_memory.is_some_and(|mm| context.extractor_alloc.allocated_bytes() >= mm)
+            {
+                // We must spill as we allocated too much memory
+                data_ref.spilled_removed = tempfile::tempfile().map(BufWriter::new).map(Some)?;
+                data_ref.spilled_inserted = tempfile::tempfile().map(BufWriter::new).map(Some)?;
+            }
+
+            match change? {
+                DocumentChange::Deletion(deletion) => {
+                    let docid = deletion.docid();
+                    let current = deletion.current(rtxn, index, db_fields_ids_map)?;
+
+                    if let Some(_geojson) = current.geojson_field()? {
+                        match &mut data_ref.spilled_removed {
+                            Some(file) => {
+                                file.write_u32::<BigEndian>(docid)?;
+                            }
+                            None => {
+                                data_ref.removed.push(docid);
+                            }
+                        }
+                    }
+                }
+                DocumentChange::Update(update) => {
+                    let current = update.current(rtxn, index, db_fields_ids_map)?;
+                    let docid = update.docid();
+
+                    let current_geo = current.geojson_field()?;
+
+                    let updated_geo =
+                        update.merged(rtxn, index, db_fields_ids_map)?.geojson_field()?;
+
+                    if current_geo.map(|c| c.get()) != updated_geo.map(|u| u.get()) {
+                        // If the current and new geo points are different it means that
+                        // we need to replace the current by the new point and therefore
+                        // delete the current point from cellulite.
+                        if let Some(_geojson) = current_geo {
+                            match &mut data_ref.spilled_removed {
+                                Some(file) => {
+                                    file.write_u32::<BigEndian>(docid)?;
+                                }
+                                None => {
+                                    data_ref.removed.push(docid);
+                                }
+                            }
+                        }
+
+                        if let Some(geojson) = updated_geo {
+                            let geojson =
+                                GeoJson::from_str(geojson.get()).map_err(UserError::from)?;
+                            let mut geometry =
+                                Geometry::try_from(geojson).map_err(UserError::from)?;
+                            cellulite::densify_geom(&mut geometry);
+
+                            let buf = ZerometryCodec::bytes_encode(&geometry).unwrap();
+
+                            match &mut data_ref.spilled_inserted {
+                                Some(file) => {
+                                    file.write_u32::<BigEndian>(docid)?;
+                                    file.write_u32::<BigEndian>(buf.len() as u32)?;
+                                    file.write_all(&buf)?;
+                                }
+                                None => {
+                                    let mut bvec =
+                                        bumpalo::collections::Vec::new_in(context.extractor_alloc);
+                                    bvec.extend_from_slice(&buf);
+                                    data_ref.inserted.push((docid, bvec.into_bump_slice()));
+                                }
+                            }
+                        }
+                    }
+                }
+                DocumentChange::Insertion(insertion) => {
+                    let docid = insertion.docid();
+                    let inserted_geo = insertion.inserted().geojson_field()?;
+
+                    if let Some(geojson) = inserted_geo {
+                        let geojson = GeoJson::from_str(geojson.get()).map_err(UserError::from)?;
+                        let mut geometry = Geometry::try_from(geojson).map_err(UserError::from)?;
+                        cellulite::densify_geom(&mut geometry);
+                        let mut bytes = Vec::new();
+                        Zerometry::write_from_geometry(&mut bytes, &geometry)?;
+
+                        match &mut data_ref.spilled_inserted {
+                            Some(file) => {
+                                file.write_u32::<BigEndian>(docid)?;
+                                file.write_u32::<BigEndian>(bytes.len() as u32)?;
+                                file.write_all(&bytes)?;
+                            }
+                            None => {
+                                let mut bvec =
+                                    bumpalo::collections::Vec::new_in(context.extractor_alloc);
+                                bvec.extend_from_slice(&bytes);
+                                data_ref.inserted.push((docid, bvec.into_bump_slice()));
+                            }
+                        }
+                    }
+                }
+            }
+        }
+
+        Ok(())
+    }
+}

crates/milli/src/update/new/extract/geo/mod.rs

@@ -18,6 +18,8 @@ use crate::update::new::DocumentChange;
 use crate::update::GrenadParameters;
 use crate::{lat_lng_to_xyz, DocumentId, GeoPoint, Index, InternalError, Result};
 
+pub mod cellulite;
+
 pub struct GeoExtractor {
     grenad_parameters: GrenadParameters,
 }

crates/milli/src/update/new/extract/vectors/mod.rs

@@ -653,7 +653,7 @@ impl<'a, 'b, 'extractor> Chunks<'a, 'b, 'extractor> {
 
                 settings_delta.try_for_each_fragment_diff(
                     session.embedder_name(),
-                    |fragment_diff| {
+                    |fragment_diff| -> Result<()> {
                         let extractor = RequestFragmentExtractor::new(fragment_diff.new, doc_alloc)
                             .ignore_errors();
                         let old = if full_reindex {

crates/milli/src/update/new/indexer/document_changes.rs

@@ -24,7 +24,7 @@ pub trait Extractor<'extractor>: Sync {
     fn process<'doc>(
         &'doc self,
         changes: impl Iterator<Item = Result<DocumentChange<'doc>>>,
-        context: &'doc DocumentContext<Self::Data>,
+        context: &'doc DocumentContext<'doc, 'extractor, '_, '_, Self::Data>,
     ) -> Result<()>;
 }
 

crates/milli/src/update/new/indexer/enterprise_edition/sharding.rs

@@ -5,18 +5,24 @@
 
 use std::hash::{BuildHasher as _, BuildHasherDefault};
 
-pub struct Shards {
-    pub own: Vec<String>,
-    pub others: Vec<String>,
+pub struct Shards(pub Vec<Shard>);
+
+pub struct Shard {
+    pub is_own: bool,
+    pub name: String,
 }
 
 impl Shards {
     pub fn must_process(&self, docid: &str) -> bool {
+        self.processing_shard(docid).map(|shard| shard.is_own).unwrap_or_default()
+    }
+
+    pub fn processing_shard<'a>(&'a self, docid: &str) -> Option<&'a Shard> {
         let hasher = BuildHasherDefault::<twox_hash::XxHash3_64>::new();
-        let to_hash = |shard: &String| hasher.hash_one((shard, docid));
+        let to_hash = |shard: &'a Shard| (shard, hasher.hash_one((&shard.name, docid)));
 
-        let max_hash = self.others.iter().map(to_hash).max().unwrap_or_default();
-
-        self.own.iter().map(to_hash).any(|hash| hash > max_hash)
+        let shard =
+            self.0.iter().map(to_hash).max_by_key(|(_, hash)| *hash).map(|(shard, _)| shard);
+        shard
     }
 }

crates/milli/src/update/new/indexer/extract.rs

@@ -16,9 +16,10 @@ use super::settings_changes::settings_change_extract;
 use crate::documents::{FieldIdMapper, PrimaryKey};
 use crate::progress::{EmbedderStats, MergingWordCache};
 use crate::proximity::ProximityPrecision;
+use crate::update::new::extract::cellulite::GeoJsonExtractor;
 use crate::update::new::extract::EmbeddingExtractor;
 use crate::update::new::indexer::settings_changes::DocumentsIndentifiers;
-use crate::update::new::merger::merge_and_send_rtree;
+use crate::update::new::merger::{merge_and_send_cellulite, merge_and_send_rtree};
 use crate::update::new::{merge_and_send_docids, merge_and_send_facet_docids, FacetDatabases};
 use crate::update::settings::SettingsDelta;
 use crate::vector::db::{EmbedderInfo, IndexEmbeddingConfig};
@@ -317,6 +318,37 @@ where
             &indexing_context.must_stop_processing,
         )?;
     }
+
+    'cellulite: {
+        let Some(extractor) =
+            GeoJsonExtractor::new(&rtxn, index, *indexing_context.grenad_parameters)?
+        else {
+            break 'cellulite;
+        };
+        let datastore = ThreadLocal::with_capacity(rayon::current_num_threads());
+
+        {
+            let span = tracing::trace_span!(target: "indexing::documents::extract", "cellulite");
+            let _entered = span.enter();
+
+            extract(
+                document_changes,
+                &extractor,
+                indexing_context,
+                extractor_allocs,
+                &datastore,
+                IndexingStep::WritingGeoJson,
+            )?;
+        }
+
+        merge_and_send_cellulite(
+            datastore,
+            &rtxn,
+            index,
+            extractor_sender.geojson(),
+            &indexing_context.must_stop_processing,
+        )?;
+    }
     indexing_context.progress.update_progress(IndexingStep::WaitingForDatabaseWrites);
     finished_extraction.store(true, std::sync::atomic::Ordering::Relaxed);
 

crates/milli/src/update/new/indexer/mod.rs

@@ -187,6 +187,13 @@ where
             facet_field_ids_delta,
         )?;
 
+        indexing_context.progress.update_progress(IndexingStep::BuildingGeoJson);
+        index.cellulite.build(
+            wtxn,
+            &indexing_context.must_stop_processing,
+            indexing_context.progress,
+        )?;
+
         indexing_context.progress.update_progress(IndexingStep::Finalizing);
 
         Ok(congestion) as Result<_>
@@ -317,6 +324,13 @@ where
         })
         .unwrap()?;
 
+        indexing_context.progress.update_progress(IndexingStep::BuildingGeoJson);
+        index.cellulite.build(
+            wtxn,
+            &indexing_context.must_stop_processing,
+            indexing_context.progress,
+        )?;
+
         indexing_context.progress.update_progress(IndexingStep::Finalizing);
 
         Ok(congestion) as Result<_>

crates/milli/src/update/new/indexer/write.rs

@@ -16,7 +16,7 @@ use crate::update::settings::InnerIndexSettings;
 use crate::vector::db::IndexEmbeddingConfig;
 use crate::vector::settings::EmbedderAction;
 use crate::vector::{Embedder, Embeddings, RuntimeEmbedders, VectorStore};
-use crate::{Error, Index, InternalError, Result, UserError};
+use crate::{DocumentId, Error, Index, InternalError, Result, UserError};
 
 pub fn write_to_db(
     mut writer_receiver: WriterBbqueueReceiver<'_>,
@@ -72,6 +72,14 @@ pub fn write_to_db(
                 let embedding = large_vector.read_embedding(*dimensions);
                 writer.add_item_in_store(wtxn, docid, extractor_id, embedding)?;
             }
+            ReceiverAction::LargeGeoJson(LargeGeoJson { docid, geojson }) => {
+                // It cannot be a deletion because it's large. Deletions are always small
+                let geojson: &[u8] = &geojson;
+                index
+                    .cellulite
+                    .add_raw_zerometry(wtxn, docid, geojson)
+                    .map_err(InternalError::CelluliteError)?;
+            }
         }
 
         // Every time the is a message in the channel we search
@@ -263,6 +271,19 @@ pub fn write_from_bbqueue(
                     writer.add_item_in_store(wtxn, docid, extractor_id, embedding)?;
                 }
             }
+            EntryHeader::CelluliteItem(docid) => {
+                let frame = frame_with_header.frame();
+                let skip = EntryHeader::variant_size() + std::mem::size_of::<DocumentId>();
+                let geojson = &frame[skip..];
+
+                index
+                    .cellulite
+                    .add_raw_zerometry(wtxn, docid, geojson)
+                    .map_err(InternalError::CelluliteError)?;
+            }
+            EntryHeader::CelluliteRemove(docid) => {
+                index.cellulite.delete(wtxn, docid).map_err(InternalError::CelluliteError)?;
+            }
         }
     }
 

crates/milli/src/update/new/merger.rs

@@ -13,6 +13,7 @@ use super::extract::{
     FacetKind, GeoExtractorData,
 };
 use crate::update::facet::new_incremental::FacetFieldIdChange;
+use crate::update::new::extract::cellulite::GeoJsonExtractorData;
 use crate::{CboRoaringBitmapCodec, FieldId, GeoPoint, Index, InternalError, Result};
 
 #[tracing::instrument(level = "trace", skip_all, target = "indexing::merge")]
@@ -62,6 +63,30 @@ where
     Ok(())
 }
 
+#[tracing::instrument(level = "trace", skip_all, target = "indexing::merge")]
+pub fn merge_and_send_cellulite<'extractor, MSP>(
+    datastore: impl IntoIterator<Item = RefCell<GeoJsonExtractorData<'extractor>>>,
+    _rtxn: &RoTxn,
+    _index: &Index,
+    geojson_sender: GeoJsonSender<'_, '_>,
+    must_stop_processing: &MSP,
+) -> Result<()>
+where
+    MSP: Fn() -> bool + Sync,
+{
+    for data in datastore {
+        if must_stop_processing() {
+            return Err(InternalError::AbortedIndexation.into());
+        }
+
+        let mut frozen = data.into_inner().freeze()?;
+        frozen.iter_and_clear_removed(geojson_sender)?;
+        frozen.iter_and_clear_inserted(geojson_sender)?;
+    }
+
+    Ok(())
+}
+
 #[tracing::instrument(level = "trace", skip_all, target = "indexing::merge")]
 pub fn merge_and_send_docids<MSP, D>(
     mut caches: Vec<BalancedCaches<'_>>,

crates/milli/src/update/new/reindex.rs

@@ -1,6 +1,7 @@
 use heed::RwTxn;
 
 use super::document::{Document, DocumentFromDb};
+use crate::constants::{RESERVED_GEOJSON_FIELD_NAME, RESERVED_GEO_FIELD_NAME};
 use crate::progress::{self, AtomicSubStep, Progress};
 use crate::{FieldDistribution, Index, Result};
 
@@ -22,8 +23,13 @@ pub fn field_distribution(index: &Index, wtxn: &mut RwTxn<'_>, progress: &Progre
         let Some(document) = DocumentFromDb::new(docid, wtxn, index, &field_id_map)? else {
             continue;
         };
-        let geo_iter = document.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
-        for res in document.iter_top_level_fields().chain(geo_iter) {
+        let geo_iter =
+            document.geo_field().transpose().map(|res| res.map(|rv| (RESERVED_GEO_FIELD_NAME, rv)));
+        let geojson_iter = document
+            .geojson_field()
+            .transpose()
+            .map(|res| res.map(|rv| (RESERVED_GEOJSON_FIELD_NAME, rv)));
+        for res in document.iter_top_level_fields().chain(geo_iter).chain(geojson_iter) {
             let (field_name, _) = res?;
             if let Some(count) = distribution.get_mut(field_name) {
                 *count += 1;

crates/milli/src/update/new/steps.rs

@@ -12,11 +12,13 @@ make_enum_progress! {
         MergingWordCaches,
         MergingWordProximity,
         WritingGeoPoints,
+        WritingGeoJson,
         WaitingForDatabaseWrites,
         WaitingForExtractors,
         WritingEmbeddingsToDatabase,
         PostProcessingFacets,
         PostProcessingWords,
+        BuildingGeoJson,
         Finalizing,
     }
 }

crates/milli/src/update/new/words_prefix_docids.rs

@@ -193,7 +193,7 @@ impl WordPrefixIntegerDocids {
         // We access this HashMap in parallel to compute the *union* of all
         // of them and *serialize* them into files. There is one file by CPU.
         let local_entries = ThreadLocal::with_capacity(rayon::current_num_threads());
-        prefixes.into_par_iter().map(AsRef::as_ref).try_for_each(|prefix| {
+        prefixes.into_par_iter().map(AsRef::as_ref).try_for_each(|prefix| -> Result<()> {
             let refcell = local_entries.get_or(|| {
                 let file = BufWriter::new(spooled_tempfile(
                     self.max_memory_by_thread.unwrap_or(usize::MAX),

crates/milli/src/update/settings.rs

@@ -15,7 +15,7 @@ use super::del_add::{DelAdd, DelAddOperation};
 use super::index_documents::{IndexDocumentsConfig, Transform};
 use super::{ChatSettings, IndexerConfig};
 use crate::attribute_patterns::PatternMatch;
-use crate::constants::RESERVED_GEO_FIELD_NAME;
+use crate::constants::{RESERVED_GEOJSON_FIELD_NAME, RESERVED_GEO_FIELD_NAME};
 use crate::criterion::Criterion;
 use crate::disabled_typos_terms::DisabledTyposTerms;
 use crate::error::UserError::{self, InvalidChatSettingsDocumentTemplateMaxBytes};
@@ -1862,7 +1862,10 @@ impl InnerIndexSettingsDiff {
     }
 
     pub fn any_reindexing_needed(&self) -> bool {
-        self.reindex_searchable() || self.reindex_facets() || self.reindex_vectors()
+        self.reindex_searchable()
+            || self.reindex_facets()
+            || self.reindex_vectors()
+            || self.reindex_geojson()
     }
 
     pub fn reindex_searchable(&self) -> bool {
@@ -1971,6 +1974,11 @@ impl InnerIndexSettingsDiff {
         !self.embedding_config_updates.is_empty()
     }
 
+    pub fn reindex_geojson(&self) -> bool {
+        self.old.filterable_attributes_rules.iter().any(|rule| rule.has_geojson())
+            != self.new.filterable_attributes_rules.iter().any(|rule| rule.has_geojson())
+    }
+
     pub fn settings_update_only(&self) -> bool {
         self.settings_update_only
     }
@@ -1979,6 +1987,11 @@ impl InnerIndexSettingsDiff {
         self.old.geo_fields_ids != self.new.geo_fields_ids
             || (!self.settings_update_only && self.new.geo_fields_ids.is_some())
     }
+
+    pub fn run_geojson_indexing(&self) -> bool {
+        self.old.geojson_fid != self.new.geojson_fid
+            || (!self.settings_update_only && self.new.geojson_fid.is_some())
+    }
 }
 
 #[derive(Clone)]
@@ -1999,6 +2012,7 @@ pub(crate) struct InnerIndexSettings {
     pub runtime_embedders: RuntimeEmbedders,
     pub embedder_category_id: HashMap<String, u8>,
     pub geo_fields_ids: Option<(FieldId, FieldId)>,
+    pub geojson_fid: Option<FieldId>,
     pub prefix_search: PrefixSearch,
     pub facet_search: bool,
 }
@@ -2038,6 +2052,7 @@ impl InnerIndexSettings {
             }
             _ => None,
         };
+        let geo_json_fid = fields_ids_map.id(RESERVED_GEOJSON_FIELD_NAME);
         let localized_attributes_rules =
             index.localized_attributes_rules(rtxn)?.unwrap_or_default();
         let filterable_attributes_rules = index.filterable_attributes_rules(rtxn)?;
@@ -2066,6 +2081,7 @@ impl InnerIndexSettings {
             runtime_embedders,
             embedder_category_id,
             geo_fields_ids,
+            geojson_fid: geo_json_fid,
             prefix_search,
             facet_search,
             disabled_typos_terms,