Compare commits

...

8 Commits

Author SHA1 Message Date
Louis Dureuil
a9ac68289f WIP 2025-11-27 17:21:40 +01:00
Louis Dureuil
cc3daa7cd3 rustfmt 2025-11-27 14:20:46 +01:00
Louis Dureuil
df737293f1 Cargo clippy 2025-11-27 14:20:17 +01:00
Louis Dureuil
584c343d59 WIP 2025-11-27 12:34:15 +01:00
Louis Dureuil
e3d5b1025e WIP 2025-11-27 11:31:04 +01:00
Louis Dureuil
72dcaee29a WIP 2025-11-26 17:52:29 +01:00
Louis Dureuil
60716f13e1 WIP 2025-11-26 16:19:40 +01:00
Louis Dureuil
c7986ca8c9 WIP 2025-11-26 10:30:05 +01:00
43 changed files with 3796 additions and 977 deletions

2
Cargo.lock generated
View File

@@ -3206,6 +3206,7 @@ dependencies = [
"enum-iterator",
"file-store",
"flate2",
"hashbrown 0.15.5",
"indexmap",
"insta",
"maplit",
@@ -4123,6 +4124,7 @@ dependencies = [
"flate2",
"fst",
"insta",
"itertools 0.14.0",
"meili-snap",
"memmap2",
"milli",

View File

@@ -9,8 +9,9 @@ use meilisearch_types::error::ResponseError;
use meilisearch_types::keys::Key;
use meilisearch_types::milli::update::IndexDocumentsMethod;
use meilisearch_types::settings::Unchecked;
use meilisearch_types::tasks::enterprise_edition::network::{DbTaskNetwork, NetworkTopologyChange};
use meilisearch_types::tasks::{
Details, ExportIndexSettings, IndexSwap, KindWithContent, Status, Task, TaskId, TaskNetwork,
Details, ExportIndexSettings, IndexSwap, KindWithContent, Status, Task, TaskId,
};
use meilisearch_types::InstanceUid;
use roaring::RoaringBitmap;
@@ -95,7 +96,7 @@ pub struct TaskDump {
)]
pub finished_at: Option<OffsetDateTime>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub network: Option<TaskNetwork>,
pub network: Option<DbTaskNetwork>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub custom_metadata: Option<String>,
}
@@ -163,6 +164,7 @@ pub enum KindDump {
IndexCompaction {
index_uid: String,
},
NetworkTopologyChange(NetworkTopologyChange),
}
impl From<Task> for TaskDump {
@@ -249,6 +251,9 @@ impl From<KindWithContent> for KindDump {
KindWithContent::IndexCompaction { index_uid } => {
KindDump::IndexCompaction { index_uid }
}
KindWithContent::NetworkTopologyChange(network_topology_change) => {
KindDump::NetworkTopologyChange(network_topology_change)
}
}
}
}
@@ -560,7 +565,8 @@ pub(crate) mod test {
Network {
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()) }},
sharding: false,
leader: None,
version: Default::default(),
}
}

View File

@@ -24,6 +24,7 @@ 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" }

View File

@@ -238,6 +238,9 @@ impl<'a> Dump<'a> {
KindDump::IndexCompaction { index_uid } => {
KindWithContent::IndexCompaction { index_uid }
}
KindDump::NetworkTopologyChange(network_topology_change) => {
KindWithContent::NetworkTopologyChange(network_topology_change)
}
},
};

View File

@@ -3,10 +3,13 @@ use std::fmt::Display;
use meilisearch_types::batches::BatchId;
use meilisearch_types::error::{Code, ErrorCode};
use meilisearch_types::milli::index::RollbackOutcome;
use meilisearch_types::milli::DocumentId;
use meilisearch_types::tasks::enterprise_edition::network::ReceiveTaskError;
use meilisearch_types::tasks::{Kind, Status};
use meilisearch_types::{heed, milli};
use reqwest::StatusCode;
use thiserror::Error;
use uuid::Uuid;
use crate::TaskId;
@@ -191,6 +194,15 @@ pub enum Error {
#[error(transparent)]
HeedTransaction(heed::Error),
#[error("No network topology change task is currently enqueued or processing")]
ImportTaskWithoutNetworkTask,
#[error("The network task version (`{network_task}`) does not match the import task version (`{import_task}`)")]
NetworkVersionMismatch { network_task: Uuid, import_task: Uuid },
#[error("The import task emanates from an unknown remote `{0}`")]
ImportTaskUnknownRemote(String),
#[error("The import task with key `{0}` was already received")]
ImportTaskAlreadyReceived(DocumentId),
#[cfg(test)]
#[error("Planned failure for tests.")]
PlannedFailure,
@@ -248,6 +260,10 @@ impl Error {
| Error::Persist(_)
| Error::FeatureNotEnabled(_)
| Error::Export(_)
| Error::ImportTaskWithoutNetworkTask
| Error::NetworkVersionMismatch { .. }
| Error::ImportTaskAlreadyReceived(_)
| Error::ImportTaskUnknownRemote(_)
| Error::Anyhow(_) => true,
Error::CreateBatch(_)
| Error::CorruptedTaskQueue
@@ -307,6 +323,10 @@ impl ErrorCode for Error {
Error::TaskDeletionWithEmptyQuery => Code::MissingTaskFilters,
Error::TaskCancelationWithEmptyQuery => Code::MissingTaskFilters,
Error::NoSpaceLeftInTaskQueue => Code::NoSpaceLeftOnDevice,
Error::ImportTaskWithoutNetworkTask => Code::ImportTaskWithoutNetworkTask,
Error::NetworkVersionMismatch { .. } => Code::NetworkVersionMismatch,
Error::ImportTaskAlreadyReceived(_) => Code::ImportTaskAlreadyReceived,
Error::ImportTaskUnknownRemote(_) => Code::ImportTaskUnknownRemote,
Error::S3Error { status, .. } if status.is_client_error() => {
Code::InvalidS3SnapshotRequest
}
@@ -345,3 +365,12 @@ impl ErrorCode for Error {
}
}
}
impl From<ReceiveTaskError> for Error {
fn from(value: ReceiveTaskError) -> Self {
match value {
ReceiveTaskError::UnknownRemote(unknown) => Error::ImportTaskUnknownRemote(unknown),
ReceiveTaskError::DuplicateTask(dup) => Error::ImportTaskAlreadyReceived(dup),
}
}
}

View File

@@ -38,6 +38,10 @@ impl RoFeatures {
Self { runtime }
}
pub fn from_runtime_features(features: RuntimeTogglableFeatures) -> Self {
Self { runtime: features }
}
pub fn runtime_features(&self) -> RuntimeTogglableFeatures {
self.runtime
}

View File

@@ -361,6 +361,11 @@ impl IndexMapper {
Ok(())
}
/// The number of indexes in the database
pub fn index_count(&self, rtxn: &RoTxn) -> Result<u64> {
Ok(self.index_mapping.len(rtxn)?)
}
/// Return an index, may open it if it wasn't already opened.
pub fn index(&self, rtxn: &RoTxn, name: &str) -> Result<Index> {
if let Some((current_name, current_index)) =

View File

@@ -325,6 +325,9 @@ fn snapshot_details(d: &Details) -> String {
Details::IndexCompaction { index_uid, pre_compaction_size, post_compaction_size } => {
format!("{{ index_uid: {index_uid:?}, pre_compaction_size: {pre_compaction_size:?}, post_compaction_size: {post_compaction_size:?} }}")
}
Details::NetworkTopologyChange { moved_documents, message } => {
format!("{{ moved_documents: {moved_documents:?}, message: {message:?}")
}
}
}

View File

@@ -68,10 +68,12 @@ use meilisearch_types::milli::vector::{
};
use meilisearch_types::milli::{self, Index};
use meilisearch_types::task_view::TaskView;
use meilisearch_types::tasks::{KindWithContent, Task, TaskNetwork};
use meilisearch_types::tasks::enterprise_edition::network::{
DbTaskNetwork, ImportData, ImportMetadata, Origin, TaskNetwork,
};
use meilisearch_types::tasks::{KindWithContent, Task};
use meilisearch_types::webhooks::{Webhook, WebhooksDumpView, WebhooksView};
use milli::vector::db::IndexEmbeddingConfig;
use processing::ProcessingTasks;
pub use queue::Query;
use queue::Queue;
use roaring::RoaringBitmap;
@@ -82,6 +84,7 @@ use uuid::Uuid;
use versioning::Versioning;
use crate::index_mapper::IndexMapper;
use crate::processing::ProcessingTasks;
use crate::utils::clamp_to_page_size;
pub(crate) type BEI128 = I128<BE>;
@@ -700,14 +703,14 @@ impl IndexScheduler {
self.queue.get_task_ids_from_authorized_indexes(&rtxn, query, filters, &processing)
}
pub fn set_task_network(&self, task_id: TaskId, network: TaskNetwork) -> Result<()> {
pub fn set_task_network(&self, task_id: TaskId, network: DbTaskNetwork) -> Result<Task> {
let mut wtxn = self.env.write_txn()?;
let mut task =
self.queue.tasks.get_task(&wtxn, task_id)?.ok_or(Error::TaskNotFound(task_id))?;
task.network = Some(network);
self.queue.tasks.all_tasks.put(&mut wtxn, &task_id, &task)?;
wtxn.commit()?;
Ok(())
Ok(task)
}
/// Return the batches matching the query from the user's point of view along
@@ -757,18 +760,30 @@ impl IndexScheduler {
task_id: Option<TaskId>,
dry_run: bool,
) -> Result<Task> {
self.register_with_custom_metadata(kind, task_id, None, dry_run)
self.register_with_custom_metadata(kind, task_id, None, dry_run, None)
}
/// Register a new task in the scheduler, with metadata.
///
/// If it fails and data was associated with the task, it tries to delete the associated data.
///
/// # Parameters
///
/// - task_network: network of the task to check.
///
/// If the task is an import task, only accept it if:
///
/// 1. There is an ongoing network topology change task
/// 2. The task to register matches the network version of the network topology change task
///
/// Always accept the task if it is not an import task.
pub fn register_with_custom_metadata(
&self,
kind: KindWithContent,
task_id: Option<TaskId>,
custom_metadata: Option<String>,
dry_run: bool,
task_network: Option<TaskNetwork>,
) -> Result<Task> {
// if the task doesn't delete or cancel anything and 40% of the task queue is full, we must refuse to enqueue the incoming task
if !matches!(&kind, KindWithContent::TaskDeletion { tasks, .. } | KindWithContent::TaskCancelation { tasks, .. } if !tasks.is_empty())
@@ -779,7 +794,19 @@ impl IndexScheduler {
}
let mut wtxn = self.env.write_txn()?;
let task = self.queue.register(&mut wtxn, &kind, task_id, custom_metadata, dry_run)?;
if let Some(TaskNetwork::Import { import_from, network_change, metadata }) = &task_network {
self.update_network_task(&mut wtxn, import_from, network_change, metadata)?;
}
let task = self.queue.register(
&mut wtxn,
&kind,
task_id,
custom_metadata,
dry_run,
task_network.map(DbTaskNetwork::from),
)?;
// If the registered task is a task cancelation
// we inform the processing tasks to stop (if necessary).
@@ -801,6 +828,91 @@ impl IndexScheduler {
Ok(task)
}
pub fn network_no_index_for_remote(
&self,
remote_name: String,
origin: Origin,
) -> Result<(), Error> {
let mut wtxn = self.env.write_txn()?;
self.update_network_task(
&mut wtxn,
&ImportData { remote_name, index_name: None, document_count: 0 },
&origin,
&ImportMetadata { index_count: 0, task_key: None, total_index_documents: 0 },
)?;
wtxn.commit()?;
// wake up the scheduler as the task state has changed
self.scheduler.wake_up.signal();
Ok(())
}
fn update_network_task(
&self,
wtxn: &mut heed::RwTxn<'_>,
import_from: &ImportData,
network_change: &Origin,
metadata: &ImportMetadata,
) -> Result<(), Error> {
let mut network_tasks = self
.queue
.tasks
.get_kind(&*wtxn, meilisearch_types::tasks::Kind::NetworkTopologyChange)?;
if network_tasks.is_empty() {
return Err(Error::ImportTaskWithoutNetworkTask);
}
let network_task = {
let processing = self.processing_tasks.read().unwrap().processing.clone();
if processing.is_disjoint(&network_tasks) {
let enqueued = self
.queue
.tasks
.get_status(&*wtxn, meilisearch_types::tasks::Status::Enqueued)?;
network_tasks &= enqueued;
if let Some(network_task) = network_tasks.into_iter().next() {
network_task
} else {
return Err(Error::ImportTaskWithoutNetworkTask);
}
} else {
network_tasks &= &*processing;
network_tasks.into_iter().next().unwrap()
}
};
let mut network_task = self.queue.tasks.get_task(&*wtxn, network_task)?.unwrap();
let network_task_version = network_task
.network
.as_ref()
.map(|network| network.network_version())
.unwrap_or_default();
if network_task_version != network_change.network_version {
return Err(Error::NetworkVersionMismatch {
network_task: network_task_version,
import_task: network_change.network_version,
});
}
let KindWithContent::NetworkTopologyChange(network_topology_change) =
&mut network_task.kind
else {
tracing::error!("unexpected network kind for network task while registering task");
return Err(Error::CorruptedTaskQueue);
};
network_topology_change.receive_remote_task(
&import_from.remote_name,
import_from.index_name.as_deref(),
metadata.task_key,
import_from.document_count,
metadata.index_count,
metadata.total_index_documents,
)?;
self.queue.tasks.update_task(wtxn, &mut network_task)?;
Ok(())
}
/// Register a new task coming from a dump in the scheduler.
/// By taking a mutable ref we're pretty sure no one will ever import a dump while actix is running.
pub fn register_dumped_task(&mut self) -> Result<Dump<'_>> {

View File

@@ -42,12 +42,10 @@ impl ProcessingTasks {
/// Set the processing tasks to an empty list
pub fn stop_processing(&mut self) -> Self {
self.progress = None;
Self {
batch: std::mem::take(&mut self.batch),
processing: std::mem::take(&mut self.processing),
progress: None,
progress: std::mem::take(&mut self.progress),
}
}

View File

@@ -15,6 +15,7 @@ use file_store::FileStore;
use meilisearch_types::batches::BatchId;
use meilisearch_types::heed::{Database, Env, RoTxn, RwTxn, WithoutTls};
use meilisearch_types::milli::{CboRoaringBitmapCodec, BEU32};
use meilisearch_types::tasks::enterprise_edition::network::DbTaskNetwork;
use meilisearch_types::tasks::{Kind, KindWithContent, Status, Task};
use roaring::RoaringBitmap;
use time::format_description::well_known::Rfc3339;
@@ -259,6 +260,7 @@ impl Queue {
task_id: Option<TaskId>,
custom_metadata: Option<String>,
dry_run: bool,
network: Option<DbTaskNetwork>,
) -> Result<Task> {
let next_task_id = self.tasks.next_task_id(wtxn)?;
@@ -280,7 +282,7 @@ impl Queue {
details: kind.default_details(),
status: Status::Enqueued,
kind: kind.clone(),
network: None,
network,
custom_metadata,
};
// For deletion and cancelation tasks, we want to make extra sure that they
@@ -348,6 +350,7 @@ impl Queue {
None,
None,
false,
None,
)?;
Ok(())

View File

@@ -3,7 +3,8 @@ use std::ops::{Bound, RangeBounds};
use meilisearch_types::heed::types::{DecodeIgnore, SerdeBincode, SerdeJson, Str};
use meilisearch_types::heed::{Database, Env, RoTxn, RwTxn, WithoutTls};
use meilisearch_types::milli::{CboRoaringBitmapCodec, RoaringBitmapCodec, BEU32};
use meilisearch_types::tasks::{Kind, Status, Task};
use meilisearch_types::tasks::enterprise_edition::network::DbTaskNetwork;
use meilisearch_types::tasks::{Kind, KindWithContent, Status, Task};
use roaring::{MultiOps, RoaringBitmap};
use time::OffsetDateTime;
@@ -114,14 +115,16 @@ impl TaskQueue {
/// - CorruptedTaskQueue: The task doesn't exist in the database
pub(crate) fn update_task(&self, wtxn: &mut RwTxn, task: &mut Task) -> Result<()> {
let old_task = self.get_task(wtxn, task.uid)?.ok_or(Error::CorruptedTaskQueue)?;
let reprocessing = old_task.status != Status::Enqueued;
// network topology tasks may be processed multiple times.
let maybe_reprocessing = old_task.status != Status::Enqueued
|| task.kind.as_kind() == Kind::NetworkTopologyChange;
debug_assert!(old_task != *task);
debug_assert_eq!(old_task.uid, task.uid);
// If we're processing a task that failed it may already contains a batch_uid
debug_assert!(
reprocessing || (old_task.batch_uid.is_none() && task.batch_uid.is_some()),
maybe_reprocessing || (old_task.batch_uid.is_none() && task.batch_uid.is_some()),
"\n==> old: {old_task:?}\n==> new: {task:?}"
);
@@ -143,13 +146,24 @@ impl TaskQueue {
})?;
}
// Avoids rewriting part of the network topology change because of TOCTOU errors
if let (
KindWithContent::NetworkTopologyChange(old_state),
KindWithContent::NetworkTopologyChange(new_state),
) = (old_task.kind, &mut task.kind)
{
new_state.merge(old_state);
// the state possibly just changed, rewrite the details
task.details = Some(new_state.to_details());
}
assert_eq!(
old_task.enqueued_at, task.enqueued_at,
"Cannot update a task's enqueued_at time"
);
if old_task.started_at != task.started_at {
assert!(
reprocessing || old_task.started_at.is_none(),
maybe_reprocessing || old_task.started_at.is_none(),
"Cannot update a task's started_at time"
);
if let Some(started_at) = old_task.started_at {
@@ -161,7 +175,7 @@ impl TaskQueue {
}
if old_task.finished_at != task.finished_at {
assert!(
reprocessing || old_task.finished_at.is_none(),
maybe_reprocessing || old_task.finished_at.is_none(),
"Cannot update a task's finished_at time"
);
if let Some(finished_at) = old_task.finished_at {
@@ -175,7 +189,16 @@ impl TaskQueue {
task.network = match (old_task.network, task.network.take()) {
(None, None) => None,
(None, Some(network)) | (Some(network), None) => Some(network),
(Some(_), Some(network)) => Some(network),
(Some(left), Some(right)) => Some(match (left, right) {
(
DbTaskNetwork::Remotes { remote_tasks: mut left, network_version: _ },
DbTaskNetwork::Remotes { remote_tasks: mut right, network_version },
) => {
left.append(&mut right);
DbTaskNetwork::Remotes { remote_tasks: left, network_version }
}
(_, right) => right,
}),
};
self.all_tasks.put(wtxn, &task.uid, task)?;

View File

@@ -203,26 +203,30 @@ fn test_disable_auto_deletion_of_tasks() {
)
.unwrap();
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks =
index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]" }), name: "task_queue_is_full");
drop(rtxn);
drop(proc);
{
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks = index_scheduler
.queue
.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc)
.unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]" }), name: "task_queue_is_full");
}
// now we're above the max number of tasks
// and if we try to advance in the tick function no new task deletion should be enqueued
handle.advance_till([Start, BatchCreated]);
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks =
index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "task_deletion_have_not_been_enqueued");
drop(rtxn);
drop(proc);
{
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks = index_scheduler
.queue
.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc)
.unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "task_deletion_have_not_been_enqueued");
}
}
#[test]
@@ -267,59 +271,69 @@ fn test_auto_deletion_of_tasks() {
)
.unwrap();
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks =
index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]" }), name: "task_queue_is_full");
drop(rtxn);
drop(proc);
{
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks = index_scheduler
.queue
.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc)
.unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]" }), name: "task_queue_is_full");
}
// now we're above the max number of tasks
// and if we try to advance in the tick function a new task deletion should be enqueued
handle.advance_till([Start, BatchCreated]);
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks =
index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "task_deletion_have_been_enqueued");
drop(rtxn);
drop(proc);
{
// now we're above the max number of tasks
// and if we try to advance in the tick function a new task deletion should be enqueued
handle.advance_till([Start, BatchCreated]);
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks = index_scheduler
.queue
.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc)
.unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "task_deletion_have_been_enqueued");
}
handle.advance_till([InsideProcessBatch, ProcessBatchSucceeded, AfterProcessing]);
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks =
index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "task_deletion_have_been_processed");
drop(rtxn);
drop(proc);
{
handle.advance_till([InsideProcessBatch, ProcessBatchSucceeded, AfterProcessing]);
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks = index_scheduler
.queue
.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc)
.unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "task_deletion_have_been_processed");
}
handle.advance_one_failed_batch();
// a new task deletion has been enqueued
handle.advance_one_successful_batch();
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks =
index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "after_the_second_task_deletion");
drop(rtxn);
drop(proc);
{
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks = index_scheduler
.queue
.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc)
.unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "after_the_second_task_deletion");
}
handle.advance_one_failed_batch();
handle.advance_one_successful_batch();
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks =
index_scheduler.queue.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc).unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "everything_has_been_processed");
drop(rtxn);
drop(proc);
{
let rtxn = index_scheduler.env.read_txn().unwrap();
let proc = index_scheduler.processing_tasks.read().unwrap();
let tasks = index_scheduler
.queue
.get_task_ids(&rtxn, &Query { ..Default::default() }, &proc)
.unwrap();
let tasks = index_scheduler.queue.tasks.get_existing_tasks(&rtxn, tasks).unwrap();
snapshot!(json_string!(tasks, { "[].enqueuedAt" => "[date]", "[].startedAt" => "[date]", "[].finishedAt" => "[date]", ".**.original_filter" => "[filter]", ".**.query" => "[query]" }), name: "everything_has_been_processed");
}
}
#[test]

View File

@@ -74,6 +74,7 @@ impl From<KindWithContent> for AutobatchKind {
| KindWithContent::DumpCreation { .. }
| KindWithContent::Export { .. }
| KindWithContent::UpgradeDatabase { .. }
| KindWithContent::NetworkTopologyChange(_)
| KindWithContent::SnapshotCreation => {
panic!("The autobatcher should never be called with tasks with special priority or that don't apply to an index.")
}

View File

@@ -4,6 +4,7 @@ use std::io::ErrorKind;
use meilisearch_types::heed::RoTxn;
use meilisearch_types::milli::update::IndexDocumentsMethod;
use meilisearch_types::settings::{Settings, Unchecked};
use meilisearch_types::tasks::enterprise_edition::network::NetworkTopologyState;
use meilisearch_types::tasks::{BatchStopReason, Kind, KindWithContent, Status, Task};
use roaring::RoaringBitmap;
use uuid::Uuid;
@@ -59,6 +60,14 @@ pub(crate) enum Batch {
index_uid: String,
task: Task,
},
#[allow(clippy::enum_variant_names)] // warranted because we are executing an inner index batch
NetworkIndexBatch {
network_task: Task,
inner_batch: Box<Batch>,
},
NetworkReady {
task: Task,
},
}
#[derive(Debug)]
@@ -140,9 +149,14 @@ impl Batch {
..
} => RoaringBitmap::from_iter(tasks.iter().chain(other).map(|task| task.uid)),
},
Batch::IndexSwap { task } => {
Batch::IndexSwap { task } | Batch::NetworkReady { task } => {
RoaringBitmap::from_sorted_iter(std::iter::once(task.uid)).unwrap()
}
Batch::NetworkIndexBatch { network_task, inner_batch } => {
let mut tasks = inner_batch.ids();
tasks.insert(network_task.uid);
tasks
}
}
}
@@ -156,12 +170,14 @@ impl Batch {
| Dump(_)
| Export { .. }
| UpgradeDatabase { .. }
| NetworkReady { .. }
| IndexSwap { .. } => None,
IndexOperation { op, .. } => Some(op.index_uid()),
IndexCreation { index_uid, .. }
| IndexUpdate { index_uid, .. }
| IndexDeletion { index_uid, .. }
| IndexCompaction { index_uid, .. } => Some(index_uid),
NetworkIndexBatch { network_task: _, inner_batch } => inner_batch.index_uid(),
}
}
}
@@ -184,6 +200,8 @@ impl fmt::Display for Batch {
Batch::IndexCompaction { .. } => f.write_str("IndexCompaction")?,
Batch::Export { .. } => f.write_str("Export")?,
Batch::UpgradeDatabase { .. } => f.write_str("UpgradeDatabase")?,
Batch::NetworkIndexBatch { .. } => f.write_str("NetworkTopologyChange")?,
Batch::NetworkReady { .. } => f.write_str("NetworkTopologyChange")?,
};
match index_uid {
Some(name) => f.write_fmt(format_args!(" on {name:?} from tasks: {tasks:?}")),
@@ -452,6 +470,7 @@ impl IndexScheduler {
pub(crate) fn create_next_batch(
&self,
rtxn: &RoTxn,
processing_network_tasks: &RoaringBitmap,
) -> Result<Option<(Batch, ProcessingBatch)>> {
#[cfg(test)]
self.maybe_fail(crate::test_utils::FailureLocation::InsideCreateBatch)?;
@@ -460,7 +479,6 @@ impl IndexScheduler {
let mut current_batch = ProcessingBatch::new(batch_id);
let enqueued = &self.queue.tasks.get_status(rtxn, Status::Enqueued)?;
let count_total_enqueued = enqueued.len();
let failed = &self.queue.tasks.get_status(rtxn, Status::Failed)?;
// 0. we get the last task to cancel.
@@ -509,7 +527,15 @@ impl IndexScheduler {
)));
}
// 2. we get the next task to delete
// 2. Check for enqueued network topology changes
let network_changes = self.queue.tasks.get_kind(rtxn, Kind::NetworkTopologyChange)?
& (enqueued | processing_network_tasks);
if let Some(task_id) = network_changes.iter().next() {
let task = self.queue.tasks.get_task(rtxn, task_id)?.unwrap();
return self.start_processing_network(rtxn, task, enqueued, current_batch);
}
// 3. we get the next task to delete
let to_delete = self.queue.tasks.get_kind(rtxn, Kind::TaskDeletion)? & enqueued;
if !to_delete.is_empty() {
let mut tasks = self.queue.tasks.get_existing_tasks(rtxn, to_delete)?;
@@ -519,7 +545,7 @@ impl IndexScheduler {
return Ok(Some((Batch::TaskDeletions(tasks), current_batch)));
}
// 3. we get the next task to compact
// 4. we get the next task to compact
let to_compact = self.queue.tasks.get_kind(rtxn, Kind::IndexCompaction)? & enqueued;
if let Some(task_id) = to_compact.min() {
let mut task =
@@ -534,7 +560,7 @@ impl IndexScheduler {
return Ok(Some((Batch::IndexCompaction { index_uid, task }, current_batch)));
}
// 4. we batch the export.
// 5. we batch the export.
let to_export = self.queue.tasks.get_kind(rtxn, Kind::Export)? & enqueued;
if !to_export.is_empty() {
let task_id = to_export.iter().next().expect("There must be at least one export task");
@@ -545,7 +571,7 @@ impl IndexScheduler {
return Ok(Some((Batch::Export { task }, current_batch)));
}
// 5. we batch the snapshot.
// 6. we batch the snapshot.
let to_snapshot = self.queue.tasks.get_kind(rtxn, Kind::SnapshotCreation)? & enqueued;
if !to_snapshot.is_empty() {
let mut tasks = self.queue.tasks.get_existing_tasks(rtxn, to_snapshot)?;
@@ -555,7 +581,7 @@ impl IndexScheduler {
return Ok(Some((Batch::SnapshotCreation(tasks), current_batch)));
}
// 6. we batch the dumps.
// 7. we batch the dumps.
let to_dump = self.queue.tasks.get_kind(rtxn, Kind::DumpCreation)? & enqueued;
if let Some(to_dump) = to_dump.min() {
let mut task =
@@ -568,25 +594,63 @@ impl IndexScheduler {
return Ok(Some((Batch::Dump(task), 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)?;
let network = self.network();
// If the task is not associated with any index, verify that it is an index swap and
// create the batch directly. Otherwise, get the index name associated with the task
// and use the autobatcher to batch the enqueued tasks associated with it
// 8. We make a batch from the unprioritised tasks.
let (batch, current_batch) =
self.create_next_batch_unprioritized(rtxn, enqueued, current_batch, |task| {
let is_task_from_the_future = task
.network
.as_ref()
.map(|task_network| task_network.network_version() > network.version)
// tasks without versions are not from the future
.unwrap_or_default();
let index_name = if let Some(&index_name) = task.indexes().first() {
index_name
} else {
assert!(matches!(&task.kind, KindWithContent::IndexSwap { swaps } if swaps.is_empty()));
current_batch.processing(Some(&mut task));
current_batch.reason(BatchStopReason::TaskCannotBeBatched {
kind: Kind::IndexSwap,
id: task.uid,
});
return Ok(Some((Batch::IndexSwap { task }, current_batch)));
is_task_from_the_future
})?;
Ok(batch.map(|batch| (batch, current_batch)))
}
fn create_next_batch_unprioritized<F>(
&self,
rtxn: &RoTxn,
enqueued: &RoaringBitmap,
mut current_batch: ProcessingBatch,
mut skip_if: F,
) -> Result<(Option<Batch>, ProcessingBatch)>
where
F: FnMut(&Task) -> bool,
{
let count_total_enqueued = enqueued.len();
let mut enqueued_it = enqueued.iter();
let mut task;
let index_name = loop {
let Some(task_id) = enqueued_it.next() else {
return Ok((None, current_batch));
};
task = self.queue.tasks.get_task(rtxn, task_id)?.ok_or(Error::CorruptedTaskQueue)?;
if skip_if(&task) {
continue;
}
// If the task is not associated with any index, verify that it is an index swap and
// create the batch directly. Otherwise, get the index name associated with the task
// and use the autobatcher to batch the enqueued tasks associated with it
if let Some(&index_name) = task.indexes().first() {
break index_name;
} else {
assert!(
matches!(&task.kind, KindWithContent::IndexSwap { swaps } if swaps.is_empty())
);
current_batch.processing(Some(&mut task));
current_batch.reason(BatchStopReason::TaskCannotBeBatched {
kind: Kind::IndexSwap,
id: task.uid,
});
return Ok((Some(Batch::IndexSwap { task }), current_batch));
};
};
let index_already_exists = self.index_mapper.exists(rtxn, index_name)?;
@@ -621,6 +685,10 @@ impl IndexScheduler {
.get_task(rtxn, task_id)
.and_then(|task| task.ok_or(Error::CorruptedTaskQueue))?;
if skip_if(&task) {
continue;
}
if let Some(uuid) = task.content_uuid() {
let content_size = match self.queue.file_store.compute_size(uuid) {
Ok(content_size) => content_size,
@@ -651,19 +719,109 @@ impl IndexScheduler {
autobatcher::autobatch(enqueued, index_already_exists, primary_key.as_deref())
{
current_batch.reason(autobatch_stop_reason.unwrap_or(stop_reason));
return Ok(self
.create_next_batch_index(
rtxn,
index_name.to_string(),
batchkind,
&mut current_batch,
create_index,
)?
.map(|batch| (batch, current_batch)));
let batch = self.create_next_batch_index(
rtxn,
index_name.to_string(),
batchkind,
&mut current_batch,
create_index,
)?;
return Ok((batch, current_batch));
}
// If we found no tasks then we were notified for something that got autobatched
// somehow and there is nothing to do.
Ok(None)
Ok((None, current_batch))
}
fn start_processing_network(
&self,
rtxn: &RoTxn,
mut task: Task,
enqueued: &RoaringBitmap,
mut current_batch: ProcessingBatch,
) -> Result<Option<(Batch, ProcessingBatch)>> {
current_batch.processing(Some(&mut task));
let change_version =
task.network.as_ref().map(|network| network.network_version()).unwrap_or_default();
let KindWithContent::NetworkTopologyChange(network_topology_change) = &task.kind else {
panic!("inconsistent kind with content")
};
match network_topology_change.state() {
NetworkTopologyState::WaitingForOlderTasks => {
let res =
self.create_next_batch_unprioritized(rtxn, enqueued, current_batch, |task| {
let has_index = task.index_uid().is_some();
if !has_index {
return true;
}
let has_older_network_version = task
.network
.as_ref()
.map(|network| network.network_version() < change_version)
// if there is no version, we never retain the task
.unwrap_or_default();
!has_older_network_version
});
let (batch, current_batch) = res?;
let batch = match batch {
Some(batch) => {
let inner_batch = Box::new(batch);
Batch::NetworkIndexBatch { network_task: task, inner_batch }
}
None => Batch::NetworkReady { task },
};
Ok(Some((batch, current_batch)))
}
NetworkTopologyState::ImportingDocuments => {
// if the import is done we need to go to the next state
if network_topology_change.is_import_finished() {
return Ok(Some((Batch::NetworkReady { task }, current_batch)));
}
let res =
self.create_next_batch_unprioritized(rtxn, enqueued, current_batch, |task| {
let has_index = task.index_uid().is_some();
if !has_index {
return true;
}
let is_import_task = task
.network
.as_ref()
.map(|network| {
network.network_version() == change_version
&& network.import_data().is_some()
})
// if there is no version, we never retain the task
.unwrap_or_default();
!is_import_task
});
let (batch, current_batch) = res?;
let batch = batch.map(|batch| {
let inner_batch = Box::new(batch);
(Batch::NetworkIndexBatch { network_task: task, inner_batch }, current_batch)
});
Ok(batch)
}
NetworkTopologyState::ExportingDocuments | NetworkTopologyState::Finished => {
Ok(Some((Batch::NetworkReady { task }, current_batch)))
}
}
}
}

View File

@@ -0,0 +1,281 @@
use std::collections::BTreeMap;
use std::time::Duration;
use bumpalo::Bump;
use meilisearch_types::enterprise_edition::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::new::indexer::enterprise_edition::sharding::Shards;
use meilisearch_types::milli::{self};
use meilisearch_types::tasks::enterprise_edition::network::{NetworkTopologyState, Origin};
use meilisearch_types::tasks::{KindWithContent, Status, Task};
use roaring::RoaringBitmap;
use super::create_batch::Batch;
use crate::scheduler::process_batch::ProcessBatchInfo;
use crate::scheduler::process_export::{ExportContext, ExportOptions, TargetInstance};
use crate::utils::ProcessingBatch;
use crate::{Error, IndexScheduler, Result};
impl IndexScheduler {
pub(super) fn process_network_index_batch(
&self,
mut network_task: Task,
inner_batch: Box<Batch>,
current_batch: &mut ProcessingBatch,
progress: Progress,
) -> Result<(Vec<Task>, ProcessBatchInfo)> {
let (mut tasks, info) = self.process_batch(*inner_batch, current_batch, progress)?;
let KindWithContent::NetworkTopologyChange(network_topology_change) =
&mut network_task.kind
else {
tracing::error!("unexpected network kind for network task while processing batch");
return Err(Error::CorruptedTaskQueue);
};
for task in &tasks {
let Some(network) = task.network.as_ref() else {
continue;
};
let Some(import) = network.import_data() else {
continue;
};
if let Some(index_name) = import.index_name.as_deref() {
network_topology_change.process_remote_tasks(
&import.remote_name,
index_name,
import.document_count,
);
}
}
network_task.details = Some(network_topology_change.to_details());
tasks.push(network_task);
Ok((tasks, info))
}
pub(super) fn process_network_ready(
&self,
mut task: Task,
progress: Progress,
) -> Result<(Vec<Task>, ProcessBatchInfo)> {
let KindWithContent::NetworkTopologyChange(network_topology_change) = &mut task.kind else {
tracing::error!("network topology change task has the wrong kind with content");
return Err(Error::CorruptedTaskQueue);
};
let Some(task_network) = &task.network else {
tracing::error!("network topology change task has no network");
return Err(Error::CorruptedTaskQueue);
};
let origin;
let origin = match task_network.origin() {
Some(origin) => origin,
None => {
let myself = network_topology_change.in_name().expect("origin is not the leader");
origin = Origin {
remote_name: myself.to_string(),
task_uid: task.uid,
network_version: task_network.network_version(),
};
&origin
}
};
if let Some((remotes, out_name)) = network_topology_change.export_to_process() {
network_topology_change.set_moved(self.balance_documents(
remotes,
out_name,
network_topology_change.in_name(),
origin,
&progress,
&self.scheduler.must_stop_processing,
)?);
}
network_topology_change.update_state();
if network_topology_change.state() == NetworkTopologyState::Finished {
task.status = Status::Succeeded;
}
task.details = Some(network_topology_change.to_details());
Ok((vec![task], Default::default()))
}
fn balance_documents(
&self,
remotes: &BTreeMap<String, Remote>,
out_name: &str,
in_name: Option<&str>,
network_change_origin: &Origin,
progress: &Progress,
must_stop_processing: &crate::scheduler::MustStopProcessing,
) -> crate::Result<u64> {
let new_shards =
Shards::from_remotes_local(remotes.keys().map(String::as_str).chain(in_name), in_name);
// 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();
let scheduler_rtxn = self.env.read_txn()?;
let index_count = self.index_mapper.index_count(&scheduler_rtxn)?;
// when the instance is empty, we still need to tell that to remotes, as they cannot know of that fact and will be waiting for
// data
if index_count == 0 {
for (remote_name, remote) in remotes {
let target = TargetInstance {
base_url: &remote.url,
api_key: remote.write_api_key.as_deref(),
};
let res = self.export_no_index(
target,
out_name,
network_change_origin,
&agent,
must_stop_processing,
);
match res {
Ok(_) => {}
Err(err) => {
tracing::warn!("Could not signal not to wait documents to `{remote_name}` due to error: {err}");
}
}
}
return Ok(0);
}
let moved_documents: Vec<u64> = self.index_mapper.try_for_each_index(
&scheduler_rtxn,
|index_uid, index| -> crate::Result<u64> {
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).or_default().insert(docid);
}
None => {
documents_to_delete.insert(docid);
}
}
}
let fields_ids_map = index.fields_ids_map(&index_rtxn)?;
for (remote_name, remote) in remotes {
let documents_to_move =
documents_to_move_to.remove(remote_name).unwrap_or_default();
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,
export_mode: super::process_export::ExportMode::NetworkBalancing {
index_count,
export_old_remote_name: out_name,
network_change_origin,
},
};
let ctx = ExportContext {
index,
index_rtxn: &index_rtxn,
universe: &documents_to_move,
progress,
agent: &agent,
must_stop_processing,
};
let res = self.export_one_index(target, options, ctx);
match res {
Ok(_) =>{ documents_to_delete |= documents_to_move;}
Err(err) => {
tracing::warn!("Could not export documents to `{remote_name}` due to error: {err}\n - Note: Documents will be kept");
}
}
}
if documents_to_delete.is_empty() {
return Ok(0);
}
let moved_count = documents_to_delete.len();
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.to_string(), 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)?;
// update stats
let mut mapper_wtxn = self.env.write_txn()?;
let stats =
crate::index_mapper::IndexStats::new(index, &index_wtxn).map_err(err)?;
self.index_mapper.store_stats_of(&mut mapper_wtxn, index_uid, &stats)?;
index_wtxn.commit()?;
// update stats after committing changes to index
mapper_wtxn.commit()?;
Ok(moved_count)
},
)?;
let moved_documents: u64 = moved_documents.into_iter().sum();
Ok(moved_documents)
}
}

View File

@@ -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;
@@ -21,7 +22,6 @@ use std::path::PathBuf;
use std::sync::atomic::{AtomicBool, AtomicU32, Ordering};
use std::sync::Arc;
use convert_case::{Case, Casing as _};
use meilisearch_types::error::ResponseError;
use meilisearch_types::heed::{Env, WithoutTls};
use meilisearch_types::milli;
@@ -178,6 +178,8 @@ impl IndexScheduler {
self.breakpoint(crate::test_utils::Breakpoint::Start);
}
let previous_processing_batch = self.processing_tasks.write().unwrap().stop_processing();
if self.cleanup_enabled {
let mut wtxn = self.env.write_txn()?;
self.queue.cleanup_task_queue(&mut wtxn)?;
@@ -185,11 +187,16 @@ impl IndexScheduler {
}
let rtxn = self.env.read_txn().map_err(Error::HeedTransaction)?;
let (batch, mut processing_batch) =
match self.create_next_batch(&rtxn).map_err(|e| Error::CreateBatch(Box::new(e)))? {
Some(batch) => batch,
None => return Ok(TickOutcome::WaitForSignal),
};
let (batch, mut processing_batch) = match self
.create_next_batch(&rtxn, &previous_processing_batch.processing)
.map_err(|e| Error::CreateBatch(Box::new(e)))?
{
Some(batch) => batch,
None => {
*self.processing_tasks.write().unwrap() = previous_processing_batch;
return Ok(TickOutcome::WaitForSignal);
}
};
let index_uid = batch.index_uid().map(ToOwned::to_owned);
drop(rtxn);
@@ -260,7 +267,14 @@ impl IndexScheduler {
self.maybe_fail(crate::test_utils::FailureLocation::AcquiringWtxn)?;
progress.update_progress(BatchProgress::WritingTasksToDisk);
processing_batch.finished();
// whether the batch made progress.
// a batch make progress if it failed or if it contains at least one fully processed (or cancelled) task.
//
// if a batch did not make progress, it means that all of its tasks are waiting on the scheduler to make progress,
// and so we must wait for new tasks. Such a batch is not persisted to DB, and is resumed on the next tick.
let mut batch_made_progress = false;
let mut stop_scheduler_forever = false;
let mut wtxn = self.env.write_txn().map_err(Error::HeedTransaction)?;
let mut canceled = RoaringBitmap::new();
@@ -281,7 +295,11 @@ impl IndexScheduler {
#[allow(unused_variables)]
for (i, mut task) in tasks.into_iter().enumerate() {
task_progress.fetch_add(1, Ordering::Relaxed);
processing_batch.update(&mut task);
processing_batch.update_from_task(&task);
if !matches!(task.status, Status::Processing | Status::Enqueued) {
batch_made_progress = true;
processing_batch.finish_task(&mut task);
}
if task.status == Status::Canceled {
canceled.insert(task.uid);
canceled_by = task.canceled_by;
@@ -348,6 +366,9 @@ impl IndexScheduler {
}
// In case of a failure we must get back and patch all the tasks with the error.
Err(err) => {
// always persist failed batches
batch_made_progress = true;
#[cfg(test)]
self.breakpoint(crate::test_utils::Breakpoint::ProcessBatchFailed);
let (task_progress, task_progress_obj) = AtomicTaskStep::new(ids.len() as u32);
@@ -371,7 +392,10 @@ impl IndexScheduler {
task.status = Status::Failed;
task.error = Some(error.clone());
task.details = task.details.map(|d| d.to_failed());
processing_batch.update(&mut task);
processing_batch.update_from_task(&task);
if !matches!(task.status, Status::Processing | Status::Enqueued) {
processing_batch.finish_task(&mut task);
}
#[cfg(test)]
self.maybe_fail(
@@ -394,44 +418,12 @@ impl IndexScheduler {
let ProcessBatchInfo { congestion, pre_commit_dabases_sizes, post_commit_dabases_sizes } =
process_batch_info;
processing_batch.stats.progress_trace =
progress.accumulated_durations().into_iter().map(|(k, v)| (k, v.into())).collect();
processing_batch.stats.write_channel_congestion = congestion.map(|congestion| {
let mut congestion_info = serde_json::Map::new();
congestion_info.insert("attempts".into(), congestion.attempts.into());
congestion_info.insert("blocking_attempts".into(), congestion.blocking_attempts.into());
congestion_info.insert("blocking_ratio".into(), congestion.congestion_ratio().into());
congestion_info
});
processing_batch.stats.internal_database_sizes = pre_commit_dabases_sizes
.iter()
.flat_map(|(dbname, pre_size)| {
post_commit_dabases_sizes
.get(dbname)
.map(|post_size| {
use std::cmp::Ordering::{Equal, Greater, Less};
use byte_unit::Byte;
use byte_unit::UnitType::Binary;
let post = Byte::from_u64(*post_size as u64).get_appropriate_unit(Binary);
let diff_size = post_size.abs_diff(*pre_size) as u64;
let diff = Byte::from_u64(diff_size).get_appropriate_unit(Binary);
let sign = match post_size.cmp(pre_size) {
Equal => return None,
Greater => "+",
Less => "-",
};
Some((
dbname.to_case(Case::Camel),
format!("{post:#.2} ({sign}{diff:#.2})").into(),
))
})
.into_iter()
.flatten()
})
.collect();
processing_batch.write_stats(
&progress,
congestion,
pre_commit_dabases_sizes,
post_commit_dabases_sizes,
);
if let Some(congestion) = congestion {
tracing::debug!(
@@ -444,46 +436,49 @@ impl IndexScheduler {
tracing::debug!("call trace: {:?}", progress.accumulated_durations());
self.queue.write_batch(&mut wtxn, processing_batch, &ids)?;
if batch_made_progress {
self.queue.write_batch(&mut wtxn, processing_batch, &ids)?;
}
#[cfg(test)]
self.maybe_fail(crate::test_utils::FailureLocation::CommittingWtxn)?;
wtxn.commit().map_err(Error::HeedTransaction)?;
// We should stop processing AFTER everything is processed and written to disk otherwise, a batch (which only lives in RAM) may appear in the processing task
// and then become « not found » for some time until the commit everything is written and the final commit is made.
self.processing_tasks.write().unwrap().stop_processing();
if batch_made_progress {
// We should stop processing AFTER everything is processed and written to disk otherwise, a batch (which only lives in RAM) may appear in the processing task
// and then become « not found » for some time until the commit everything is written and the final commit is made.
self.processing_tasks.write().unwrap().stop_processing();
// Once the tasks are committed, we should delete all the update files associated ASAP to avoid leaking files in case of a restart
tracing::debug!("Deleting the update files");
// Once the tasks are committed, we should delete all the update files associated ASAP to avoid leaking files in case of a restart
tracing::debug!("Deleting the update files");
//We take one read transaction **per thread**. Then, every thread is going to pull out new IDs from the roaring bitmap with the help of an atomic shared index into the bitmap
let idx = AtomicU32::new(0);
(0..current_num_threads()).into_par_iter().try_for_each(|_| -> Result<()> {
let rtxn = self.read_txn()?;
while let Some(id) = ids.select(idx.fetch_add(1, Ordering::Relaxed)) {
let task = self
.queue
.tasks
.get_task(&rtxn, id)
.map_err(|e| Error::UnrecoverableError(Box::new(e)))?
.ok_or(Error::CorruptedTaskQueue)?;
if let Err(e) = self.queue.delete_persisted_task_data(&task) {
tracing::error!(
//We take one read transaction **per thread**. Then, every thread is going to pull out new IDs from the roaring bitmap with the help of an atomic shared index into the bitmap
let idx = AtomicU32::new(0);
(0..current_num_threads()).into_par_iter().try_for_each(|_| -> Result<()> {
let rtxn = self.read_txn()?;
while let Some(id) = ids.select(idx.fetch_add(1, Ordering::Relaxed)) {
let task = self
.queue
.tasks
.get_task(&rtxn, id)
.map_err(|e| Error::UnrecoverableError(Box::new(e)))?
.ok_or(Error::CorruptedTaskQueue)?;
if let Err(e) = self.queue.delete_persisted_task_data(&task) {
tracing::error!(
"Failure to delete the content files associated with task {}. Error: {e}",
task.uid
);
}
}
}
Ok(())
})?;
Ok(())
})?;
self.notify_webhooks(ids);
self.notify_webhooks(ids);
}
#[cfg(test)]
self.breakpoint(crate::test_utils::Breakpoint::AfterProcessing);
if stop_scheduler_forever {
Ok(TickOutcome::StopProcessingForever)
} else {

View File

@@ -539,6 +539,10 @@ impl IndexScheduler {
Ok((tasks, ProcessBatchInfo::default()))
}
Batch::NetworkIndexBatch { network_task, inner_batch } => {
self.process_network_index_batch(network_task, inner_batch, current_batch, progress)
}
Batch::NetworkReady { task } => self.process_network_ready(task, progress),
}
}

View File

@@ -1,5 +1,6 @@
use std::collections::BTreeMap;
use std::io::{self, Write as _};
use std::ops::ControlFlow;
use std::sync::atomic;
use std::time::Duration;
@@ -7,6 +8,7 @@ use backoff::ExponentialBackoff;
use byte_unit::Byte;
use flate2::write::GzEncoder;
use flate2::Compression;
use meilisearch_types::error::Code;
use meilisearch_types::index_uid_pattern::IndexUidPattern;
use meilisearch_types::milli::constants::RESERVED_VECTORS_FIELD_NAME;
use meilisearch_types::milli::index::EmbeddingsWithMetadata;
@@ -15,7 +17,11 @@ use meilisearch_types::milli::update::{request_threads, Setting};
use meilisearch_types::milli::vector::parsed_vectors::{ExplicitVectors, VectorOrArrayOfVectors};
use meilisearch_types::milli::{self, obkv_to_json, Filter, InternalError};
use meilisearch_types::settings::{self, SecretPolicy};
use meilisearch_types::tasks::enterprise_edition::network::{
headers, ImportData, ImportMetadata, Origin,
};
use meilisearch_types::tasks::{DetailsExportIndexSettings, ExportIndexSettings};
use roaring::RoaringBitmap;
use serde::Deserialize;
use ureq::{json, Response};
@@ -50,6 +56,7 @@ 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);
}
@@ -61,261 +68,426 @@ 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,
export_mode: ExportMode::ExportRoute,
};
let total_documents = self.export_one_index(target, options, ctx)?;
output.insert(
IndexUidPattern::new_unchecked(uid.clone()),
DetailsExportIndexSettings {
settings: (*export_settings).clone(),
matched_documents: Some(total_documents as u64),
matched_documents: Some(total_documents),
},
);
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 total_index_documents = ctx.universe.len();
let task_network = options.task_network(total_index_documents);
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 == Code::IndexNotFound.name() =>
{
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
);
let _ = handle_response(retry(ctx.must_stop_processing, || {
let mut request = ctx.agent.patch(&url);
if let Some((import_data, origin, metadata)) = &task_network {
request = set_network_ureq_headers(request, import_data, origin, metadata);
}
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
);
// no document to send, but we must still send a task when performing network balancing
if ctx.universe.is_empty() {
if let Some((import_data, network_change_origin, metadata)) = task_network {
let mut compressed_buffer = Vec::new();
// ignore control flow, we're returning anyway
let _ = send_buffer(
b" ", // needs something otherwise meili complains about missing payload
&mut compressed_buffer,
ctx.must_stop_processing,
ctx.agent,
&documents_url,
bearer.as_deref(),
Some(&(import_data, network_change_origin.clone(), metadata)),
&err,
)?;
}
return Ok(0);
}
let results = request_threads()
.broadcast(|broadcast| {
let mut task_network = options.task_network(total_index_documents);
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;
}
if let Some((import_data, _, metadata)) = &mut task_network {
import_data.document_count += 1;
metadata.task_key = Some(docid);
}
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 {
let control_flow = send_buffer(
&buffer,
&mut compressed_buffer,
ctx.must_stop_processing,
ctx.agent,
&documents_url,
bearer.as_deref(),
task_network.as_ref(),
&err,
)?;
buffer.clear();
compressed_buffer.clear();
if let Some((import_data, _, metadata)) = &mut task_network {
import_data.document_count = 0;
metadata.task_key = None;
}
if control_flow.is_break() {
return Ok(());
}
}
buffer.extend_from_slice(&tmp_buffer);
if i > 0 && i % 100 == 0 {
step.fetch_add(100, atomic::Ordering::Relaxed);
}
}
// send the last buffered documents if any
if !buffer.is_empty() {
// ignore control flow here
let _ = send_buffer(
&buffer,
&mut compressed_buffer,
ctx.must_stop_processing,
ctx.agent,
&documents_url,
bearer.as_deref(),
task_network.as_ref(),
&err,
)?;
}
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)
}
pub(super) fn export_no_index(
&self,
target: TargetInstance<'_>,
export_old_remote_name: &str,
network_change_origin: &Origin,
agent: &ureq::Agent,
must_stop_processing: &MustStopProcessing,
) -> Result<(), Error> {
let bearer = target.api_key.map(|api_key| format!("Bearer {api_key}"));
let url = format!("{base_url}/network", base_url = target.base_url,);
{
let _ = handle_response(retry(must_stop_processing, || {
let request = agent.patch(&url);
let mut request = set_network_ureq_headers(
request,
&ImportData {
remote_name: export_old_remote_name.to_string(),
index_name: None,
document_count: 0,
},
network_change_origin,
&ImportMetadata { index_count: 0, task_key: None, total_index_documents: 0 },
);
request = request.set("Content-Type", "application/json");
if let Some(bearer) = &bearer {
request = request.set("Authorization", bearer);
}
request
.send_json(
// empty payload that will be disregarded
serde_json::Value::Object(Default::default()),
)
.map_err(into_backoff_error)
}))?;
}
Ok(())
}
}
fn set_network_ureq_headers(
request: ureq::Request,
import_data: &ImportData,
origin: &Origin,
metadata: &ImportMetadata,
) -> ureq::Request {
let request = request
.set(headers::PROXY_ORIGIN_REMOTE_HEADER, &origin.remote_name)
.set(headers::PROXY_ORIGIN_TASK_UID_HEADER, &origin.task_uid.to_string())
.set(headers::PROXY_ORIGIN_NETWORK_VERSION_HEADER, &origin.network_version.to_string())
.set(headers::PROXY_IMPORT_REMOTE_HEADER, &import_data.remote_name)
.set(headers::PROXY_IMPORT_DOCS_HEADER, &import_data.document_count.to_string())
.set(headers::PROXY_IMPORT_INDEX_COUNT_HEADER, &metadata.index_count.to_string())
.set(
headers::PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER,
&metadata.total_index_documents.to_string(),
);
let request = if let Some(index_name) = import_data.index_name.as_deref() {
request.set(headers::PROXY_IMPORT_INDEX_HEADER, index_name)
} else {
request
};
if let Some(task_key) = metadata.task_key {
request.set(headers::PROXY_IMPORT_TASK_KEY_HEADER, &task_key.to_string())
} else {
request
}
}
#[allow(clippy::too_many_arguments)]
fn send_buffer<'a>(
buffer: &'a [u8],
mut compressed_buffer: &'a mut Vec<u8>,
must_stop_processing: &MustStopProcessing,
agent: &ureq::Agent,
documents_url: &'a str,
bearer: Option<&'a str>,
task_network: Option<&(ImportData, Origin, ImportMetadata)>,
err: &'a impl Fn(milli::Error) -> crate::Error,
) -> Result<ControlFlow<(), ()>> {
// We compress the documents before sending them
let mut encoder: GzEncoder<&mut &mut Vec<u8>> =
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)?;
let res = 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);
}
if let Some((import_data, origin, metadata)) = task_network {
request = set_network_ureq_headers(request, import_data, origin, metadata);
}
request.send_bytes(compressed_buffer).map_err(into_backoff_error)
});
handle_response(res)
}
fn handle_response(res: Result<Response>) -> Result<ControlFlow<()>> {
match res {
Ok(_response) => Ok(ControlFlow::Continue(())),
Err(Error::FromRemoteWhenExporting { code, .. })
if code == Code::ImportTaskAlreadyReceived.name() =>
{
Ok(ControlFlow::Continue(()))
}
Err(Error::FromRemoteWhenExporting { code, message, .. })
if code == Code::ImportTaskUnknownRemote.name() =>
{
tracing::warn!("remote answered with: {message}");
Ok(ControlFlow::Break(()))
}
// note: there has already been many attempts to get this due to exponential backoff
Err(Error::FromRemoteWhenExporting { code, message, .. })
if code == Code::ImportTaskWithoutNetworkTask.name() =>
{
tracing::warn!("remote answered with: {message}");
Ok(ControlFlow::Break(()))
}
Err(e) => {
tracing::warn!("error while exporting: {e}");
Err(e)
}
}
}
fn retry<F>(must_stop_processing: &MustStopProcessing, send_request: F) -> Result<ureq::Response>
@@ -374,4 +546,63 @@ 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) export_mode: ExportMode<'a>,
}
impl ExportOptions<'_> {
fn task_network(
&self,
total_index_documents: u64,
) -> Option<(ImportData, Origin, ImportMetadata)> {
if let ExportMode::NetworkBalancing {
index_count,
export_old_remote_name,
network_change_origin,
} = self.export_mode
{
Some((
ImportData {
remote_name: export_old_remote_name.to_string(),
index_name: Some(self.index_uid.to_string()),
document_count: 0,
},
network_change_origin.clone(),
ImportMetadata { index_count, task_key: None, total_index_documents },
))
} else {
None
}
}
}
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,
}
pub(super) enum ExportMode<'a> {
ExportRoute,
NetworkBalancing {
index_count: u64,
export_old_remote_name: &'a str,
network_change_origin: &'a Origin,
},
}
// progress related
enum ExportIndex {}

View File

@@ -747,6 +747,7 @@ fn basic_get_stats() {
"indexDeletion": 0,
"indexSwap": 0,
"indexUpdate": 0,
"networkTopologyChange": 0,
"settingsUpdate": 0,
"snapshotCreation": 0,
"taskCancelation": 0,
@@ -782,6 +783,7 @@ fn basic_get_stats() {
"indexDeletion": 0,
"indexSwap": 0,
"indexUpdate": 0,
"networkTopologyChange": 0,
"settingsUpdate": 0,
"snapshotCreation": 0,
"taskCancelation": 0,
@@ -824,6 +826,7 @@ fn basic_get_stats() {
"indexDeletion": 0,
"indexSwap": 0,
"indexUpdate": 0,
"networkTopologyChange": 0,
"settingsUpdate": 0,
"snapshotCreation": 0,
"taskCancelation": 0,
@@ -867,6 +870,7 @@ fn basic_get_stats() {
"indexDeletion": 0,
"indexSwap": 0,
"indexUpdate": 0,
"networkTopologyChange": 0,
"settingsUpdate": 0,
"snapshotCreation": 0,
"taskCancelation": 0,

View File

@@ -4,9 +4,11 @@ use std::collections::{BTreeSet, HashSet};
use std::ops::Bound;
use std::sync::Arc;
use convert_case::{Case, Casing as _};
use meilisearch_types::batches::{Batch, BatchEnqueuedAt, BatchId, BatchStats};
use meilisearch_types::heed::{Database, RoTxn, RwTxn};
use meilisearch_types::milli::CboRoaringBitmapCodec;
use meilisearch_types::milli::progress::Progress;
use meilisearch_types::milli::{CboRoaringBitmapCodec, ChannelCongestion};
use meilisearch_types::task_view::DetailsView;
use meilisearch_types::tasks::{
BatchStopReason, Details, IndexSwap, Kind, KindWithContent, Status,
@@ -119,17 +121,8 @@ impl ProcessingBatch {
self.stats.total_nb_tasks = 0;
}
/// Update the timestamp of the tasks and the inner structure of this structure.
pub fn update(&mut self, task: &mut Task) {
// We must re-set this value in case we're dealing with a task that has been added between
// the `processing` and `finished` state
// We must re-set this value in case we're dealing with a task that has been added between
// the `processing` and `finished` state or that failed.
task.batch_uid = Some(self.uid);
// Same
task.started_at = Some(self.started_at);
task.finished_at = self.finished_at;
/// Update batch task from a processed task
pub fn update_from_task(&mut self, task: &Task) {
self.statuses.insert(task.status);
// Craft an aggregation of the details of all the tasks encountered in this batch.
@@ -144,6 +137,63 @@ impl ProcessingBatch {
}
}
/// Update the timestamp of the tasks after they're done
pub fn finish_task(&self, task: &mut Task) {
// We must re-set this value in case we're dealing with a task that has been added between
// the `processing` and `finished` state or that failed.
task.batch_uid = Some(self.uid);
// Same
task.started_at = Some(self.started_at);
task.finished_at = self.finished_at;
}
pub fn write_stats(
&mut self,
progress: &Progress,
congestion: Option<ChannelCongestion>,
pre_commit_dabases_sizes: indexmap::IndexMap<&'static str, usize>,
post_commit_dabases_sizes: indexmap::IndexMap<&'static str, usize>,
) {
self.stats.progress_trace =
progress.accumulated_durations().into_iter().map(|(k, v)| (k, v.into())).collect();
self.stats.write_channel_congestion = congestion.map(|congestion| {
let mut congestion_info = serde_json::Map::new();
congestion_info.insert("attempts".into(), congestion.attempts.into());
congestion_info.insert("blocking_attempts".into(), congestion.blocking_attempts.into());
congestion_info.insert("blocking_ratio".into(), congestion.congestion_ratio().into());
congestion_info
});
self.stats.internal_database_sizes = pre_commit_dabases_sizes
.iter()
.flat_map(|(dbname, pre_size)| {
post_commit_dabases_sizes
.get(dbname)
.map(|post_size| {
use std::cmp::Ordering::{Equal, Greater, Less};
use byte_unit::Byte;
use byte_unit::UnitType::Binary;
let post = Byte::from_u64(*post_size as u64).get_appropriate_unit(Binary);
let diff_size = post_size.abs_diff(*pre_size) as u64;
let diff = Byte::from_u64(diff_size).get_appropriate_unit(Binary);
let sign = match post_size.cmp(pre_size) {
Equal => return None,
Greater => "+",
Less => "-",
};
Some((
dbname.to_case(Case::Camel),
format!("{post:#.2} ({sign}{diff:#.2})").into(),
))
})
.into_iter()
.flatten()
})
.collect();
}
pub fn to_batch(&self) -> Batch {
Batch {
uid: self.uid,
@@ -286,6 +336,7 @@ 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 {
@@ -627,6 +678,9 @@ impl crate::IndexScheduler {
} => {
assert_eq!(kind.as_kind(), Kind::IndexCompaction);
}
Details::NetworkTopologyChange { moved_documents: _, message: _ } => {
assert_eq!(kind.as_kind(), Kind::NetworkTopologyChange);
}
}
}

View File

@@ -24,6 +24,7 @@ enum-iterator = "2.1.0"
file-store = { path = "../file-store" }
flate2 = "1.1.2"
fst = "0.4.7"
itertools = "0.14.0"
memmap2 = "0.9.7"
milli = { path = "../milli" }
roaring = { version = "0.10.12", features = ["serde"] }

View File

@@ -7,6 +7,7 @@ use std::collections::BTreeMap;
use milli::update::new::indexer::enterprise_edition::sharding::Shards;
use serde::{Deserialize, Serialize};
use uuid::Uuid;
#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq, Default)]
#[serde(rename_all = "camelCase")]
@@ -16,20 +17,18 @@ pub struct Network {
#[serde(default)]
pub remotes: BTreeMap<String, Remote>,
#[serde(default)]
pub sharding: bool,
pub leader: Option<String>,
#[serde(default)]
pub version: Uuid,
}
impl Network {
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 })
if self.leader.is_some() {
Some(Shards::from_remotes_local(
self.remotes.keys().map(String::as_str),
self.local.as_deref(),
))
} else {
None
}

View File

@@ -156,7 +156,7 @@ macro_rules! make_error_codes {
}
/// return error name, used as error code
fn name(&self) -> String {
pub fn name(&self) -> String {
match self {
$(
Code::$code_ident => stringify!($code_ident).to_case(convert_case::Case::Snake)
@@ -214,6 +214,9 @@ ImmutableApiKeyUid , InvalidRequest , BAD_REQU
ImmutableApiKeyUpdatedAt , InvalidRequest , BAD_REQUEST;
ImmutableIndexCreatedAt , InvalidRequest , BAD_REQUEST;
ImmutableIndexUpdatedAt , InvalidRequest , BAD_REQUEST;
ImportTaskAlreadyReceived , InvalidRequest , PRECONDITION_FAILED;
ImportTaskUnknownRemote , InvalidRequest , PRECONDITION_FAILED;
ImportTaskWithoutNetworkTask , InvalidRequest , SERVICE_UNAVAILABLE;
IndexAlreadyExists , InvalidRequest , CONFLICT ;
IndexCreationFailed , Internal , INTERNAL_SERVER_ERROR;
IndexNotFound , InvalidRequest , NOT_FOUND;
@@ -270,9 +273,9 @@ InvalidMultiSearchQueryRankingRules , InvalidRequest , BAD_REQU
InvalidMultiSearchQueryPosition , InvalidRequest , BAD_REQUEST ;
InvalidMultiSearchRemote , InvalidRequest , BAD_REQUEST ;
InvalidMultiSearchWeight , InvalidRequest , BAD_REQUEST ;
InvalidNetworkLeader , InvalidRequest , BAD_REQUEST ;
InvalidNetworkRemotes , InvalidRequest , BAD_REQUEST ;
InvalidNetworkSelf , InvalidRequest , BAD_REQUEST ;
InvalidNetworkSharding , InvalidRequest , BAD_REQUEST ;
InvalidNetworkSearchApiKey , InvalidRequest , BAD_REQUEST ;
InvalidNetworkWriteApiKey , InvalidRequest , BAD_REQUEST ;
InvalidNetworkUrl , InvalidRequest , BAD_REQUEST ;
@@ -377,7 +380,9 @@ MissingPayload , InvalidRequest , BAD_REQU
MissingSearchHybrid , InvalidRequest , BAD_REQUEST ;
MissingSwapIndexes , InvalidRequest , BAD_REQUEST ;
MissingTaskFilters , InvalidRequest , BAD_REQUEST ;
NetworkVersionMismatch , InvalidRequest , PRECONDITION_FAILED ;
NoSpaceLeftOnDevice , System , UNPROCESSABLE_ENTITY;
NotLeader , InvalidRequest , BAD_REQUEST ;
PayloadTooLarge , InvalidRequest , PAYLOAD_TOO_LARGE ;
RemoteBadResponse , System , BAD_GATEWAY ;
RemoteBadRequest , InvalidRequest , BAD_REQUEST ;
@@ -391,6 +396,9 @@ TaskFileNotFound , InvalidRequest , NOT_FOUN
BatchNotFound , InvalidRequest , NOT_FOUND ;
TooManyOpenFiles , System , UNPROCESSABLE_ENTITY ;
TooManyVectors , InvalidRequest , BAD_REQUEST ;
UnexpectedNetworkPreviousRemotes , InvalidRequest , BAD_REQUEST ;
NetworkVersionTooOld , InvalidRequest , BAD_REQUEST ;
UnprocessedNetworkTask , InvalidRequest , BAD_REQUEST ;
UnretrievableDocument , Internal , BAD_REQUEST ;
UnretrievableErrorCode , InvalidRequest , BAD_REQUEST ;
UnsupportedMediaType , InvalidRequest , UNSUPPORTED_MEDIA_TYPE ;

View File

@@ -9,12 +9,12 @@ use utoipa::ToSchema;
use crate::batches::BatchId;
use crate::error::ResponseError;
use crate::settings::{Settings, Unchecked};
use crate::tasks::enterprise_edition::network::DbTaskNetwork;
use crate::tasks::{
serialize_duration, Details, DetailsExportIndexSettings, IndexSwap, Kind, Status, Task, TaskId,
TaskNetwork,
};
#[derive(Debug, Clone, PartialEq, Serialize, ToSchema)]
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize, ToSchema)]
#[serde(rename_all = "camelCase")]
#[schema(rename_all = "camelCase")]
pub struct TaskView {
@@ -54,7 +54,7 @@ pub struct TaskView {
pub finished_at: Option<OffsetDateTime>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub network: Option<TaskNetwork>,
pub network: Option<DbTaskNetwork>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub custom_metadata: Option<String>,
@@ -151,6 +151,11 @@ pub struct DetailsView {
pub pre_compaction_size: Option<String>,
#[serde(skip_serializing_if = "Option::is_none")]
pub post_compaction_size: Option<String>,
// network topology change
#[serde(skip_serializing_if = "Option::is_none")]
pub moved_documents: Option<u64>,
#[serde(skip_serializing_if = "Option::is_none")]
pub message: Option<String>,
}
impl DetailsView {
@@ -161,6 +166,17 @@ impl DetailsView {
(None, Some(doc)) | (Some(doc), None) => Some(doc),
(Some(left), Some(right)) => Some(left + right),
},
moved_documents: match (self.moved_documents, other.moved_documents) {
(None, None) => None,
(None, Some(doc)) | (Some(doc), None) => Some(doc),
(Some(left), Some(right)) => Some(left + right),
},
message: match (&mut self.message, &other.message) {
(None, None) => None,
(None, Some(message)) => Some(message.clone()),
(Some(message), None) => Some(std::mem::take(message)),
(Some(message), Some(_)) => Some(std::mem::take(message)),
},
indexed_documents: match (self.indexed_documents, other.indexed_documents) {
(None, None) => None,
(None, Some(None)) | (Some(None), None) | (Some(None), Some(None)) => Some(None),
@@ -451,6 +467,11 @@ impl From<Details> for DetailsView {
..Default::default()
}
}
Details::NetworkTopologyChange { moved_documents, message } => DetailsView {
moved_documents: Some(moved_documents),
message: Some(message),
..Default::default()
},
}
}
}

View File

@@ -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>
pub mod network;

View File

@@ -0,0 +1,681 @@
// 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, BTreeSet};
use itertools::{EitherOrBoth, Itertools as _};
use milli::DocumentId;
use serde::{Deserialize, Serialize};
use utoipa::ToSchema;
use uuid::Uuid;
use crate::enterprise_edition::network::{Network, Remote};
use crate::error::ResponseError;
use crate::tasks::{Details, TaskId};
#[derive(Debug, PartialEq, Clone, Serialize, Deserialize, ToSchema)]
#[serde(untagged, rename_all = "camelCase")]
// This type is used in the database, care should be taken when modifying it.
pub enum DbTaskNetwork {
/// Tasks that were duplicated from `origin`
Origin { origin: Origin },
/// Tasks that were duplicated as `remote_tasks`
Remotes {
remote_tasks: BTreeMap<String, RemoteTask>,
#[serde(default)]
network_version: Uuid,
},
/// Document import tasks sent in the context of `network_change`
Import { import_from: ImportData, network_change: Origin },
}
impl DbTaskNetwork {
pub fn network_version(&self) -> Uuid {
match self {
DbTaskNetwork::Origin { origin } => origin.network_version,
DbTaskNetwork::Remotes { remote_tasks: _, network_version } => *network_version,
DbTaskNetwork::Import { import_from: _, network_change } => {
network_change.network_version
}
}
}
pub fn import_data(&self) -> Option<&ImportData> {
match self {
DbTaskNetwork::Origin { .. } | DbTaskNetwork::Remotes { .. } => None,
DbTaskNetwork::Import { import_from, .. } => Some(import_from),
}
}
pub fn origin(&self) -> Option<&Origin> {
match self {
DbTaskNetwork::Origin { origin } => Some(origin),
DbTaskNetwork::Remotes { .. } => None,
DbTaskNetwork::Import { network_change, .. } => Some(network_change),
}
}
}
#[derive(Debug, PartialEq, Clone)]
pub enum TaskNetwork {
/// Tasks that were duplicated from `origin`
Origin { origin: Origin },
/// Tasks that were duplicated as `remote_tasks`
Remotes { remote_tasks: BTreeMap<String, RemoteTask>, network_version: Uuid },
/// Document import tasks sent in the context of `network_change`
Import { import_from: ImportData, network_change: Origin, metadata: ImportMetadata },
}
impl TaskNetwork {
pub fn network_version(&self) -> Uuid {
match self {
TaskNetwork::Origin { origin } => origin.network_version,
TaskNetwork::Remotes { remote_tasks: _, network_version } => *network_version,
TaskNetwork::Import { import_from: _, network_change, metadata: _ } => {
network_change.network_version
}
}
}
}
impl From<TaskNetwork> for DbTaskNetwork {
fn from(value: TaskNetwork) -> Self {
match value {
TaskNetwork::Origin { origin } => DbTaskNetwork::Origin { origin },
TaskNetwork::Remotes { remote_tasks, network_version } => {
DbTaskNetwork::Remotes { remote_tasks, network_version }
}
TaskNetwork::Import { import_from, network_change, metadata: _ } => {
DbTaskNetwork::Import { import_from, network_change }
}
}
}
}
#[derive(Debug, PartialEq, Clone, Serialize, Deserialize, ToSchema)]
#[serde(rename_all = "camelCase")]
pub struct Origin {
pub remote_name: String,
pub task_uid: u32,
pub network_version: Uuid,
}
/// Import data stored in a task
#[derive(Debug, PartialEq, Clone, Serialize, Deserialize, ToSchema)]
#[serde(rename_all = "camelCase")]
pub struct ImportData {
/// Remote that this task is imported from
pub remote_name: String,
/// Index relevant to this task
pub index_name: Option<String>,
/// Number of documents in this task
pub document_count: u64,
}
/// Import metadata associated with a task but not stored in the task
#[derive(Debug, PartialEq, Clone)]
pub struct ImportMetadata {
/// Total number of indexes to import from this host
pub index_count: u64,
/// Key unique to this (network_change, index, host, key).
///
/// In practice, an internal document id of one of the documents to import.
pub task_key: Option<DocumentId>,
/// Total number of documents to import for this index from this host.
pub total_index_documents: u64,
}
#[derive(Debug, PartialEq, Clone, Serialize, Deserialize, ToSchema)]
#[serde(rename_all = "camelCase")]
pub struct RemoteTask {
#[serde(skip_serializing_if = "Option::is_none")]
task_uid: Option<TaskId>,
error: Option<ResponseError>,
}
impl From<Result<TaskId, ResponseError>> for RemoteTask {
fn from(res: Result<TaskId, ResponseError>) -> RemoteTask {
match res {
Ok(task_uid) => RemoteTask { task_uid: Some(task_uid), error: None },
Err(err) => RemoteTask { task_uid: None, error: Some(err) },
}
}
}
/// Contains the full state of a network topology change.
///
/// A network topology change task is unique in that it can be processed in multiple different batches, as its resolution
/// depends on various document additions tasks being processed.
///
/// A network topology task has 4 states:
///
/// 1. Processing any task that was meant for an earlier version of the network. This is necessary to know that we have the right version of
/// documents.
/// 2. Sending all documents that must be moved to other remotes.
/// 3. Processing any task coming from the remotes.
/// 4. Finished.
///
/// Furthermore, it maintains some stats
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct NetworkTopologyChange {
state: NetworkTopologyState,
// in name, `None` if the node is no longer part of the network
#[serde(skip_serializing_if = "Option::is_none")]
in_name: Option<String>,
// out name, `None` if the node is new to the network
#[serde(skip_serializing_if = "Option::is_none")]
out_name: Option<String>,
out_remotes: BTreeMap<String, Remote>,
in_remotes: BTreeMap<String, InRemote>,
stats: NetworkTopologyStats,
}
impl NetworkTopologyChange {
pub fn new(old_network: Network, new_network: Network) -> Self {
// we use our old name as export name
let out_name = old_network.local;
// we use our new name as import name
let in_name = new_network.local;
// we export to the new network
let mut out_remotes = new_network.remotes;
// don't export to ourselves
if let Some(in_name) = &in_name {
out_remotes.remove(in_name);
}
let in_remotes = old_network
.remotes
.into_keys()
// don't await imports from ourselves
.filter(|name| Some(name.as_str()) != out_name.as_deref())
.map(|name| (name, InRemote::new()))
.collect();
Self {
state: NetworkTopologyState::WaitingForOlderTasks,
in_name,
out_name,
out_remotes,
in_remotes,
stats: NetworkTopologyStats { moved_documents: 0 },
}
}
pub fn state(&self) -> NetworkTopologyState {
self.state
}
pub fn out_name(&self) -> Option<&str> {
// unwrap: one of out name or in_name must be defined
self.out_name.as_deref()
}
pub fn in_name(&self) -> Option<&str> {
self.in_name.as_deref()
}
pub fn export_to_process(&self) -> Option<(&BTreeMap<String, Remote>, &str)> {
if self.state != NetworkTopologyState::ExportingDocuments {
return None;
}
if self.out_remotes.is_empty() {
return None;
}
let out_name = self.out_name()?;
Some((&self.out_remotes, out_name))
}
pub fn set_moved(&mut self, moved_documents: u64) {
self.stats.moved_documents = moved_documents;
}
/// Compute the next state from the current state of the task.
pub fn update_state(&mut self) {
self.state = match self.state {
NetworkTopologyState::WaitingForOlderTasks => {
// no more older tasks, so finished waiting
NetworkTopologyState::ExportingDocuments
}
NetworkTopologyState::ExportingDocuments => {
// processed all exported documents
if self.is_import_finished() {
NetworkTopologyState::Finished
} else {
NetworkTopologyState::ImportingDocuments
}
}
NetworkTopologyState::ImportingDocuments => {
if self.is_import_finished() {
NetworkTopologyState::Finished
} else {
NetworkTopologyState::ImportingDocuments
}
}
NetworkTopologyState::Finished => NetworkTopologyState::Finished,
};
}
pub fn receive_remote_task(
&mut self,
remote_name: &str,
index_name: Option<&str>,
task_key: Option<DocumentId>,
document_count: u64,
total_indexes: u64,
total_index_documents: u64,
) -> Result<(), ReceiveTaskError> {
let remote = self
.in_remotes
.get_mut(remote_name)
.ok_or_else(|| ReceiveTaskError::UnknownRemote(remote_name.to_string()))?;
remote.import_state = match std::mem::take(&mut remote.import_state) {
ImportState::WaitingForInitialTask => {
if total_indexes == 0 {
ImportState::Finished { total_indexes, total_documents: 0 }
} else {
let mut task_keys = BTreeSet::new();
if let Some(index_name) = index_name {
if let Some(task_key) = task_key {
task_keys.insert(task_key);
}
let mut import_index_state = BTreeMap::new();
import_index_state.insert(
index_name.to_owned(),
ImportIndexState::Ongoing {
total_documents: total_index_documents,
received_documents: document_count,
task_keys,
processed_documents: 0,
},
);
ImportState::Ongoing { import_index_state, total_indexes }
} else {
ImportState::WaitingForInitialTask
}
}
}
ImportState::Ongoing { mut import_index_state, total_indexes } => {
if let Some(index_name) = index_name {
if let Some((index_name, mut index_state)) =
import_index_state.remove_entry(index_name)
{
index_state = match index_state {
ImportIndexState::Ongoing {
total_documents,
received_documents: previously_received,
processed_documents,
mut task_keys,
} => {
if let Some(task_key) = task_key {
if !task_keys.insert(task_key) {
return Err(ReceiveTaskError::DuplicateTask(task_key));
}
}
ImportIndexState::Ongoing {
total_documents,
received_documents: previously_received + document_count,
processed_documents,
task_keys,
}
}
ImportIndexState::Finished { total_documents } => {
ImportIndexState::Finished { total_documents }
}
};
import_index_state.insert(index_name, index_state);
} else {
let mut task_keys = BTreeSet::new();
if let Some(task_key) = task_key {
task_keys.insert(task_key);
}
let state = ImportIndexState::Ongoing {
total_documents: total_index_documents,
received_documents: document_count,
processed_documents: 0,
task_keys,
};
import_index_state.insert(index_name.to_string(), state);
}
ImportState::Ongoing { import_index_state, total_indexes }
} else {
ImportState::Ongoing { import_index_state, total_indexes }
}
}
ImportState::Finished { total_indexes, total_documents } => {
ImportState::Finished { total_indexes, total_documents }
}
};
Ok(())
}
pub fn process_remote_tasks(
&mut self,
remote_name: &str,
index_name: &str,
document_count: u64,
) {
let remote = self.in_remotes.get_mut(remote_name).unwrap();
remote.import_state = match std::mem::take(&mut remote.import_state) {
ImportState::WaitingForInitialTask => panic!("no task received yet one processed"),
ImportState::Ongoing { mut import_index_state, total_indexes } => {
let (index_name, mut index_state) =
import_index_state.remove_entry(index_name).unwrap();
index_state = match index_state {
ImportIndexState::Ongoing {
total_documents,
received_documents,
processed_documents: previously_processed,
task_keys,
} => {
let newly_processed_documents = previously_processed + document_count;
if newly_processed_documents >= total_documents {
ImportIndexState::Finished { total_documents }
} else {
ImportIndexState::Ongoing {
total_documents,
received_documents,
processed_documents: newly_processed_documents,
task_keys,
}
}
}
ImportIndexState::Finished { total_documents } => {
ImportIndexState::Finished { total_documents }
}
};
import_index_state.insert(index_name, index_state);
if import_index_state.len() as u64 == total_indexes
&& import_index_state.values().all(|index| index.is_finished())
{
let total_documents =
import_index_state.values().map(|index| index.total_documents()).sum();
ImportState::Finished { total_indexes, total_documents }
} else {
ImportState::Ongoing { import_index_state, total_indexes }
}
}
ImportState::Finished { total_indexes, total_documents } => {
ImportState::Finished { total_indexes, total_documents }
}
}
}
pub fn to_details(&self) -> Details {
let message = match self.state {
NetworkTopologyState::WaitingForOlderTasks => {
"Waiting for tasks enqueued before the network change to finish processing".into()
}
NetworkTopologyState::ExportingDocuments => "Exporting documents".into(),
NetworkTopologyState::ImportingDocuments => {
let mut finished_count = 0;
let mut first_ongoing = None;
let mut ongoing_total_indexes = 0;
let mut ongoing_processed_documents = 0;
let mut ongoing_missing_documents = 0;
let mut ongoing_total_documents = 0;
let mut other_ongoing_count = 0;
let mut first_waiting = None;
let mut other_waiting_count = 0;
for (remote_name, in_remote) in &self.in_remotes {
match &in_remote.import_state {
ImportState::WaitingForInitialTask => {
first_waiting = match first_waiting {
None => Some(remote_name),
first_waiting => {
other_waiting_count += 1;
first_waiting
}
};
}
ImportState::Ongoing { import_index_state, total_indexes } => {
first_ongoing = match first_ongoing {
None => {
ongoing_total_indexes = *total_indexes;
Some(remote_name)
}
first_ongoing => {
other_ongoing_count += 1;
first_ongoing
}
};
for import_state in import_index_state.values() {
match import_state {
ImportIndexState::Ongoing {
total_documents,
processed_documents,
received_documents,
task_keys: _,
} => {
ongoing_total_documents += total_documents;
ongoing_processed_documents += processed_documents;
ongoing_missing_documents +=
total_documents.saturating_sub(*received_documents);
}
ImportIndexState::Finished { total_documents } => {
ongoing_total_documents += total_documents;
ongoing_processed_documents += total_documents;
}
}
}
}
ImportState::Finished { total_indexes, total_documents } => {
finished_count += 1;
ongoing_total_indexes = *total_indexes;
ongoing_total_documents += *total_documents;
ongoing_processed_documents += *total_documents;
}
}
}
format!(
"Importing documents from {total} remotes{waiting}{ongoing}{finished}",
total = self.in_remotes.len(),
waiting = if let Some(first_waiting) = first_waiting {
&format!(
", waiting on first task from `{}`{others}",
first_waiting,
others = if other_waiting_count > 0 {
&format!(" and {other_waiting_count} other remotes")
} else {
""
}
)
} else {
""
},
ongoing = if let Some(first_ongoing) = first_ongoing {
&format!(", awaiting {ongoing_missing_documents} and processed {ongoing_processed_documents} out of {ongoing_total_documents} documents in {ongoing_total_indexes} indexes from `{first_ongoing}`{others}",
others=if other_ongoing_count > 0 {&format!(" and {other_ongoing_count} other remotes")} else {""})
} else {
""
},
finished = if finished_count >= 0 {
&format!(", {finished_count} remotes finished processing")
} else {
""
}
)
}
NetworkTopologyState::Finished => "Finished".into(),
};
Details::NetworkTopologyChange { moved_documents: self.stats.moved_documents, message }
}
pub fn is_import_finished(&self) -> bool {
self.in_remotes.values().all(|remote| remote.is_finished())
}
pub fn merge(&mut self, other: NetworkTopologyChange) {
// The topology change has a guarantee of forward progress, so for each field we're going to keep the "most advanced" values.
let Self { state, in_name: _, out_name: _, out_remotes: _, in_remotes, stats } = self;
*state = Ord::max(*state, other.state);
*stats = Ord::max(*stats, other.stats);
for (old_value, new_value) in other.in_remotes.into_values().zip(in_remotes.values_mut()) {
new_value.import_state = match (old_value.import_state, std::mem::take(&mut new_value.import_state)) {
// waiting for initial task is always older
(ImportState::WaitingForInitialTask, newer)
| (newer, ImportState::WaitingForInitialTask)
// finished is always newer
| (_, newer @ ImportState::Finished { .. })
| (newer @ ImportState::Finished { .. }, _) => newer,
(
ImportState::Ongoing { import_index_state: left_import, total_indexes: left_total_indexes },
ImportState::Ongoing { import_index_state: right_import, total_indexes: right_total_indexes },
) => {
let import_index_state = left_import.into_iter().merge_join_by(right_import.into_iter(), |(k,_), (x, _)|k.cmp(x)).map(|eob|
match eob {
EitherOrBoth::Both((name, left), (_, right)) => {
let newer = merge_import_index_state(left, right);
(name, newer)
},
EitherOrBoth::Left(import) |
EitherOrBoth::Right(import) => import,
}
).collect();
ImportState::Ongoing{ import_index_state, total_indexes : u64::max(left_total_indexes, right_total_indexes) }
},
}
}
}
}
fn merge_import_index_state(left: ImportIndexState, right: ImportIndexState) -> ImportIndexState {
match (left, right) {
(_, newer @ ImportIndexState::Finished { .. }) => newer,
(newer @ ImportIndexState::Finished { .. }, _) => newer,
(
ImportIndexState::Ongoing {
total_documents: left_total_documents,
received_documents: left_received_documents,
processed_documents: left_processed_documents,
task_keys: mut left_task_keys,
},
ImportIndexState::Ongoing {
total_documents: right_total_documents,
received_documents: right_received_documents,
processed_documents: right_processed_documents,
task_keys: mut right_task_keys,
},
) => {
let total_documents = u64::max(left_total_documents, right_total_documents);
let received_documents = u64::max(left_received_documents, right_received_documents);
let processed_documents = u64::max(left_processed_documents, right_processed_documents);
left_task_keys.append(&mut right_task_keys);
let task_keys = left_task_keys;
ImportIndexState::Ongoing {
total_documents,
received_documents,
processed_documents,
task_keys,
}
}
}
}
pub enum ReceiveTaskError {
UnknownRemote(String),
DuplicateTask(DocumentId),
}
#[derive(Debug, Clone, Copy, PartialEq, Serialize, Deserialize, Eq, PartialOrd, Ord)]
#[serde(rename_all = "camelCase")]
pub enum NetworkTopologyState {
WaitingForOlderTasks,
ExportingDocuments,
ImportingDocuments,
Finished,
}
#[derive(Debug, Clone, Copy, PartialEq, Serialize, Deserialize, Eq, PartialOrd, Ord)]
#[serde(rename_all = "camelCase")]
pub struct NetworkTopologyStats {
#[serde(default)]
pub moved_documents: u64,
}
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct InRemote {
import_state: ImportState,
}
impl InRemote {
pub fn new() -> Self {
Self { import_state: ImportState::WaitingForInitialTask }
}
pub fn is_finished(&self) -> bool {
matches!(self.import_state, ImportState::Finished { .. })
}
}
impl Default for InRemote {
fn default() -> Self {
Self::new()
}
}
#[derive(Default, Debug, Clone, PartialEq, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
enum ImportState {
/// Initially Meilisearch doesn't know how many documents it should expect from a remote.
/// The first task for each remote contains the information of how many indexes will be imported,
/// and the first task for each index contains the number of documents to import for that index.
#[default]
WaitingForInitialTask,
Ongoing {
import_index_state: BTreeMap<String, ImportIndexState>,
total_indexes: u64,
},
Finished {
total_indexes: u64,
total_documents: u64,
},
}
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
enum ImportIndexState {
Ongoing {
total_documents: u64,
received_documents: u64,
processed_documents: u64,
task_keys: BTreeSet<DocumentId>,
},
Finished {
total_documents: u64,
},
}
impl ImportIndexState {
pub fn is_finished(&self) -> bool {
matches!(self, ImportIndexState::Finished { .. })
}
fn total_documents(&self) -> u64 {
match *self {
ImportIndexState::Ongoing { total_documents, .. }
| ImportIndexState::Finished { total_documents } => total_documents,
}
}
}
pub mod headers {
pub const PROXY_ORIGIN_REMOTE_HEADER: &str = "Meili-Proxy-Origin-Remote";
pub const PROXY_ORIGIN_TASK_UID_HEADER: &str = "Meili-Proxy-Origin-TaskUid";
pub const PROXY_ORIGIN_NETWORK_VERSION_HEADER: &str = "Meili-Proxy-Origin-Network-Version";
pub const PROXY_IMPORT_REMOTE_HEADER: &str = "Meili-Proxy-Import-Remote";
pub const PROXY_IMPORT_INDEX_COUNT_HEADER: &str = "Meili-Proxy-Import-Index-Count";
pub const PROXY_IMPORT_INDEX_HEADER: &str = "Meili-Proxy-Import-Index";
pub const PROXY_IMPORT_TASK_KEY_HEADER: &str = "Meili-Proxy-Import-Task-Key";
pub const PROXY_IMPORT_DOCS_HEADER: &str = "Meili-Proxy-Import-Docs";
pub const PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER: &str = "Meili-Proxy-Import-Total-Index-Docs";
}

View File

@@ -23,6 +23,8 @@ use crate::{versioning, InstanceUid};
pub type TaskId = u32;
pub mod enterprise_edition;
#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub struct Task {
@@ -44,7 +46,7 @@ pub struct Task {
pub kind: KindWithContent,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub network: Option<TaskNetwork>,
pub network: Option<enterprise_edition::network::DbTaskNetwork>,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub custom_metadata: Option<String>,
@@ -61,6 +63,7 @@ impl Task {
| TaskDeletion { .. }
| Export { .. }
| UpgradeDatabase { .. }
| NetworkTopologyChange { .. }
| IndexSwap { .. } => None,
DocumentAdditionOrUpdate { index_uid, .. }
| DocumentEdition { index_uid, .. }
@@ -99,6 +102,7 @@ impl Task {
| KindWithContent::SnapshotCreation
| KindWithContent::Export { .. }
| KindWithContent::UpgradeDatabase { .. }
| KindWithContent::NetworkTopologyChange { .. }
| KindWithContent::IndexCompaction { .. } => None,
}
}
@@ -178,6 +182,7 @@ pub enum KindWithContent {
IndexCompaction {
index_uid: String,
},
NetworkTopologyChange(enterprise_edition::network::NetworkTopologyChange),
}
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, ToSchema)]
@@ -215,6 +220,7 @@ impl KindWithContent {
KindWithContent::Export { .. } => Kind::Export,
KindWithContent::UpgradeDatabase { .. } => Kind::UpgradeDatabase,
KindWithContent::IndexCompaction { .. } => Kind::IndexCompaction,
KindWithContent::NetworkTopologyChange { .. } => Kind::NetworkTopologyChange,
}
}
@@ -227,6 +233,7 @@ impl KindWithContent {
| TaskCancelation { .. }
| TaskDeletion { .. }
| Export { .. }
| NetworkTopologyChange { .. }
| UpgradeDatabase { .. } => vec![],
DocumentAdditionOrUpdate { index_uid, .. }
| DocumentEdition { index_uid, .. }
@@ -340,6 +347,10 @@ impl KindWithContent {
pre_compaction_size: None,
post_compaction_size: None,
}),
KindWithContent::NetworkTopologyChange { .. } => Some(Details::NetworkTopologyChange {
moved_documents: 0,
message: "processing tasks for previous network versions".into(),
}),
}
}
@@ -392,7 +403,7 @@ impl KindWithContent {
})
}
KindWithContent::IndexSwap { .. } => {
todo!()
unimplemented!("do not call `default_finished_details` for `IndexSwap` tasks")
}
KindWithContent::TaskCancelation { query, tasks } => Some(Details::TaskCancelation {
matched_tasks: tasks.len(),
@@ -427,6 +438,9 @@ impl KindWithContent {
pre_compaction_size: None,
post_compaction_size: None,
}),
KindWithContent::NetworkTopologyChange(network_topology_change) => {
Some(network_topology_change.to_details())
}
}
}
}
@@ -494,6 +508,9 @@ impl From<&KindWithContent> for Option<Details> {
pre_compaction_size: None,
post_compaction_size: None,
}),
KindWithContent::NetworkTopologyChange(network_topology_change) => {
Some(network_topology_change.to_details())
}
}
}
}
@@ -605,6 +622,7 @@ pub enum Kind {
Export,
UpgradeDatabase,
IndexCompaction,
NetworkTopologyChange,
}
impl Kind {
@@ -624,6 +642,7 @@ impl Kind {
| Kind::DumpCreation
| Kind::Export
| Kind::UpgradeDatabase
| Kind::NetworkTopologyChange
| Kind::SnapshotCreation => false,
}
}
@@ -646,6 +665,7 @@ impl Display for Kind {
Kind::Export => write!(f, "export"),
Kind::UpgradeDatabase => write!(f, "upgradeDatabase"),
Kind::IndexCompaction => write!(f, "indexCompaction"),
Kind::NetworkTopologyChange => write!(f, "networkTopologyChange"),
}
}
}
@@ -683,6 +703,8 @@ impl FromStr for Kind {
Ok(Kind::UpgradeDatabase)
} else if kind.eq_ignore_ascii_case("indexCompaction") {
Ok(Kind::IndexCompaction)
} else if kind.eq_ignore_ascii_case("networkTopologyChange") {
Ok(Kind::NetworkTopologyChange)
} else {
Err(ParseTaskKindError(kind.to_owned()))
}
@@ -773,36 +795,10 @@ pub enum Details {
pre_compaction_size: Option<Byte>,
post_compaction_size: Option<Byte>,
},
}
#[derive(Debug, PartialEq, Clone, Serialize, Deserialize, ToSchema)]
#[serde(untagged, rename_all = "camelCase")]
pub enum TaskNetwork {
Origin { origin: Origin },
Remotes { remote_tasks: BTreeMap<String, RemoteTask> },
}
#[derive(Debug, PartialEq, Clone, Serialize, Deserialize, ToSchema)]
#[serde(rename_all = "camelCase")]
pub struct Origin {
pub remote_name: String,
pub task_uid: usize,
}
#[derive(Debug, PartialEq, Clone, Serialize, Deserialize, ToSchema)]
#[serde(rename_all = "camelCase")]
pub struct RemoteTask {
#[serde(skip_serializing_if = "Option::is_none")]
task_uid: Option<TaskId>,
error: Option<ResponseError>,
}
impl From<Result<TaskId, ResponseError>> for RemoteTask {
fn from(res: Result<TaskId, ResponseError>) -> RemoteTask {
match res {
Ok(task_uid) => RemoteTask { task_uid: Some(task_uid), error: None },
Err(err) => RemoteTask { task_uid: None, error: Some(err) },
}
}
NetworkTopologyChange {
moved_documents: u64,
message: String,
},
}
#[derive(Debug, PartialEq, Clone, Serialize, Deserialize, ToSchema)]
@@ -845,6 +841,9 @@ impl Details {
| Self::Export { .. }
| Self::UpgradeDatabase { .. }
| Self::IndexSwap { .. } => (),
Self::NetworkTopologyChange { moved_documents: _, message } => {
*message = format!("Failed. Previous status: {}", message);
}
}
details

View File

@@ -6,8 +6,13 @@ use meilisearch_types::error::{Code, ErrorCode, ResponseError};
use meilisearch_types::index_uid::{IndexUid, IndexUidFormatError};
use meilisearch_types::milli;
use meilisearch_types::milli::OrderBy;
use meilisearch_types::tasks::enterprise_edition::network::headers::{
PROXY_IMPORT_DOCS_HEADER, PROXY_IMPORT_INDEX_COUNT_HEADER, PROXY_IMPORT_INDEX_HEADER,
PROXY_IMPORT_REMOTE_HEADER, PROXY_IMPORT_TASK_KEY_HEADER, PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER,
};
use serde_json::Value;
use tokio::task::JoinError;
use uuid::Uuid;
use crate::routes::indexes::{PROXY_ORIGIN_REMOTE_HEADER, PROXY_ORIGIN_TASK_UID_HEADER};
@@ -93,8 +98,58 @@ pub enum MeilisearchHttpError {
} else { PROXY_ORIGIN_TASK_UID_HEADER }
)]
InconsistentOriginHeaders { is_remote_missing: bool },
#[error("Inconsistent `Import` headers: {remote}: {remote_status}, {index}: {index_status}, {docs}: {docs_status}.\n - Hint: either all three headers should be provided, or none of them",
remote = PROXY_IMPORT_REMOTE_HEADER,
remote_status = if *is_remote_missing { "missing" } else{ "provided" },
index = PROXY_IMPORT_INDEX_HEADER,
index_status = if *is_index_missing { "missing" } else { "provided" },
docs = PROXY_IMPORT_DOCS_HEADER,
docs_status = if *is_docs_missing { "missing" } else { "provided" }
)]
InconsistentImportHeaders {
is_remote_missing: bool,
is_index_missing: bool,
is_docs_missing: bool,
},
#[error("Inconsistent `Import-Metadata` headers: {index_count}: {index_count_status}, {task_key}: {task_key_status}, {total_index_documents}: {total_index_documents_status}.\n - Hint: either all three headers should be provided, or none of them",
index_count = PROXY_IMPORT_INDEX_COUNT_HEADER,
index_count_status = if *is_index_count_missing { "missing" } else { "provided"},
task_key = PROXY_IMPORT_TASK_KEY_HEADER,
task_key_status = if *is_task_key_missing { "missing" } else { "provided"},
total_index_documents = PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER,
total_index_documents_status = if *is_total_index_documents_missing { "missing" } else { "provided"},
)]
InconsistentImportMetadataHeaders {
is_index_count_missing: bool,
is_task_key_missing: bool,
is_total_index_documents_missing: bool,
},
#[error(
"Inconsistent task network headers: origin headers: {origin_status}, import headers: {import_status}, import metadata: {import_metadata_status}",
origin_status = if *is_missing_origin { "missing"} else { "present" },
import_status = if *is_missing_import { "missing"} else { "present" },
import_metadata_status = if *is_missing_import_metadata { "missing"} else { "present" })]
InconsistentTaskNetworkHeaders {
is_missing_origin: bool,
is_missing_import: bool,
is_missing_import_metadata: bool,
},
#[error("Invalid value for header {header_name}: {msg}")]
InvalidHeaderValue { header_name: &'static str, msg: String },
#[error("This remote is not the leader of the network.\n - Note: only the leader `{leader}` can receive new tasks.")]
NotLeader { leader: String },
#[error("Unexpected `previousRemotes` in network call.\n - Note: `previousRemote` is reserved for internal use.")]
UnexpectedNetworkPreviousRemotes,
#[error("The network version in request is too old.\n - Received: {received}\n - Expected at least: {expected_at_least}")]
NetworkVersionTooOld { received: Uuid, expected_at_least: Uuid },
#[error("Remote `{remote}` encountered an error: {error}")]
RemoteIndexScheduler { remote: String, error: index_scheduler::Error },
#[error("{if_remote}Already has a pending network task with uid {task_uid}.\n - Note: No network task can be registered while any previous network task is not done processing.\n - Hint: Wait for task {task_uid} to complete or cancel it.",
if_remote=if let Some(remote) = remote {
format!("Remote `{remote}` encountered an error: ")
} else {"".into()} )]
UnprocessedNetworkTask { remote: Option<String>, task_uid: meilisearch_types::tasks::TaskId },
}
impl MeilisearchHttpError {
@@ -122,6 +177,7 @@ impl ErrorCode for MeilisearchHttpError {
MeilisearchHttpError::SerdeJson(_) => Code::Internal,
MeilisearchHttpError::HeedError(_) => Code::Internal,
MeilisearchHttpError::IndexScheduler(e) => e.error_code(),
MeilisearchHttpError::RemoteIndexScheduler { error, .. } => error.error_code(),
MeilisearchHttpError::Milli { error, .. } => error.error_code(),
MeilisearchHttpError::Payload(e) => e.error_code(),
MeilisearchHttpError::FileStore(_) => Code::Internal,
@@ -142,10 +198,19 @@ impl ErrorCode for MeilisearchHttpError {
MeilisearchHttpError::PersonalizationInFederatedQuery(_) => {
Code::InvalidMultiSearchQueryPersonalization
}
MeilisearchHttpError::InconsistentOriginHeaders { .. } => {
MeilisearchHttpError::InconsistentOriginHeaders { .. }
| MeilisearchHttpError::InconsistentImportHeaders { .. }
| MeilisearchHttpError::InconsistentImportMetadataHeaders { .. }
| MeilisearchHttpError::InconsistentTaskNetworkHeaders { .. } => {
Code::InconsistentDocumentChangeHeaders
}
MeilisearchHttpError::InvalidHeaderValue { .. } => Code::InvalidHeaderValue,
MeilisearchHttpError::NotLeader { .. } => Code::NotLeader,
MeilisearchHttpError::UnexpectedNetworkPreviousRemotes => {
Code::UnexpectedNetworkPreviousRemotes
}
MeilisearchHttpError::NetworkVersionTooOld { .. } => Code::NetworkVersionTooOld,
MeilisearchHttpError::UnprocessedNetworkTask { .. } => Code::UnprocessedNetworkTask,
}
}
}

View File

@@ -45,7 +45,9 @@ 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::{
proxy, task_network_and_check_leader_and_version, Body,
};
use crate::routes::indexes::search::fix_sort_query_parameters;
use crate::routes::{
get_task_id, is_dry_run, PaginationView, SummarizedTaskView, PAGINATION_DEFAULT_LIMIT,
@@ -342,6 +344,7 @@ 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 task_network = task_network_and_check_leader_and_version(&req, &network)?;
analytics.publish(
DocumentsDeletionAggregator {
@@ -359,16 +362,23 @@ pub async fn delete_document(
};
let uid = get_task_id(&req, &opt)?;
let dry_run = is_dry_run(&req, &opt)?;
let task = {
let mut task = {
let index_scheduler = index_scheduler.clone();
tokio::task::spawn_blocking(move || {
index_scheduler.register_with_custom_metadata(task, uid, custom_metadata, dry_run)
index_scheduler.register_with_custom_metadata(
task,
uid,
custom_metadata,
dry_run,
task_network,
)
})
.await??
};
if network.sharding && !dry_run {
proxy(&index_scheduler, &index_uid, &req, network, Body::none(), &task).await?;
if let Some(task_network) = task.network.take() {
proxy(&index_scheduler, Some(&index_uid), &req, task_network, network, Body::none(), &task)
.await?;
}
let task: SummarizedTaskView = task.into();
@@ -967,6 +977,7 @@ async fn document_addition(
) -> Result<SummarizedTaskView, MeilisearchHttpError> {
let mime_type = extract_mime_type(req)?;
let network = index_scheduler.network();
let task_network = task_network_and_check_leader_and_version(req, &network)?;
let format = match (
mime_type.as_ref().map(|m| (m.type_().as_str(), m.subtype().as_str())),
@@ -1085,9 +1096,16 @@ async fn document_addition(
index_uid: index_uid.to_string(),
};
// FIXME: not new to #6000, but _any_ error here will cause the payload to unduly persist
let scheduler = index_scheduler.clone();
let task = match tokio::task::spawn_blocking(move || {
scheduler.register_with_custom_metadata(task, task_id, custom_metadata, dry_run)
let mut task = match tokio::task::spawn_blocking(move || {
scheduler.register_with_custom_metadata(
task,
task_id,
custom_metadata,
dry_run,
task_network,
)
})
.await?
{
@@ -1098,12 +1116,13 @@ async fn document_addition(
}
};
if network.sharding {
if let Some(task_network) = task.network.take() {
if let Some(file) = file {
proxy(
&index_scheduler,
&index_uid,
Some(&index_uid),
req,
task_network,
network,
Body::with_ndjson_payload(file),
&task,
@@ -1194,6 +1213,7 @@ pub async fn delete_documents_batch(
let index_uid = IndexUid::try_from(index_uid.into_inner())?;
let network = index_scheduler.network();
let task_network = task_network_and_check_leader_and_version(&req, &network)?;
analytics.publish(
DocumentsDeletionAggregator {
@@ -1214,16 +1234,31 @@ pub async fn delete_documents_batch(
KindWithContent::DocumentDeletion { index_uid: index_uid.to_string(), documents_ids: ids };
let uid = get_task_id(&req, &opt)?;
let dry_run = is_dry_run(&req, &opt)?;
let task = {
let mut task = {
let index_scheduler = index_scheduler.clone();
tokio::task::spawn_blocking(move || {
index_scheduler.register_with_custom_metadata(task, uid, custom_metadata, dry_run)
index_scheduler.register_with_custom_metadata(
task,
uid,
custom_metadata,
dry_run,
task_network,
)
})
.await??
};
if network.sharding && !dry_run {
proxy(&index_scheduler, &index_uid, &req, network, Body::Inline(body), &task).await?;
if let Some(task_network) = task.network.take() {
proxy(
&index_scheduler,
Some(&index_uid),
&req,
task_network,
network,
Body::inline(body),
&task,
)
.await?;
}
let task: SummarizedTaskView = task.into();
@@ -1286,6 +1321,7 @@ 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 task_network = task_network_and_check_leader_and_version(&req, &network)?;
analytics.publish(
DocumentsDeletionAggregator {
@@ -1312,16 +1348,31 @@ pub async fn delete_documents_by_filter(
let uid = get_task_id(&req, &opt)?;
let dry_run = is_dry_run(&req, &opt)?;
let task = {
let mut task = {
let index_scheduler = index_scheduler.clone();
tokio::task::spawn_blocking(move || {
index_scheduler.register_with_custom_metadata(task, uid, custom_metadata, dry_run)
index_scheduler.register_with_custom_metadata(
task,
uid,
custom_metadata,
dry_run,
task_network,
)
})
.await??
};
if network.sharding && !dry_run {
proxy(&index_scheduler, &index_uid, &req, network, Body::Inline(filter), &task).await?;
if let Some(task_network) = task.network.take() {
proxy(
&index_scheduler,
Some(&index_uid),
&req,
task_network,
network,
Body::inline(filter),
&task,
)
.await?;
}
let task: SummarizedTaskView = task.into();
@@ -1421,6 +1472,7 @@ pub async fn edit_documents_by_function(
.check_edit_documents_by_function("Using the documents edit route")?;
let network = index_scheduler.network();
let task_network = task_network_and_check_leader_and_version(&req, &network)?;
let index_uid = IndexUid::try_from(index_uid.into_inner())?;
let index_uid = index_uid.into_inner();
@@ -1467,16 +1519,31 @@ pub async fn edit_documents_by_function(
let uid = get_task_id(&req, &opt)?;
let dry_run = is_dry_run(&req, &opt)?;
let task = {
let mut task = {
let index_scheduler = index_scheduler.clone();
tokio::task::spawn_blocking(move || {
index_scheduler.register_with_custom_metadata(task, uid, custom_metadata, dry_run)
index_scheduler.register_with_custom_metadata(
task,
uid,
custom_metadata,
dry_run,
task_network,
)
})
.await??
};
if network.sharding && !dry_run {
proxy(&index_scheduler, &index_uid, &req, network, Body::Inline(body), &task).await?;
if let Some(task_network) = task.network.take() {
proxy(
&index_scheduler,
Some(&index_uid),
&req,
task_network,
network,
Body::inline(body),
&task,
)
.await?;
}
let task: SummarizedTaskView = task.into();
@@ -1525,6 +1592,7 @@ pub async fn clear_all_documents(
let index_uid = IndexUid::try_from(index_uid.into_inner())?;
let network = index_scheduler.network();
let CustomMetadataQuery { custom_metadata } = params.into_inner();
let task_network = task_network_and_check_leader_and_version(&req, &network)?;
analytics.publish(
DocumentsDeletionAggregator {
@@ -1540,17 +1608,24 @@ pub async fn clear_all_documents(
let uid = get_task_id(&req, &opt)?;
let dry_run = is_dry_run(&req, &opt)?;
let task = {
let mut task = {
let index_scheduler = index_scheduler.clone();
tokio::task::spawn_blocking(move || {
index_scheduler.register_with_custom_metadata(task, uid, custom_metadata, dry_run)
index_scheduler.register_with_custom_metadata(
task,
uid,
custom_metadata,
dry_run,
task_network,
)
})
.await??
};
if network.sharding && !dry_run {
proxy(&index_scheduler, &index_uid, &req, network, Body::none(), &task).await?;
if let Some(task_network) = task.network.take() {
proxy(&index_scheduler, Some(&index_uid), &req, task_network, network, Body::none(), &task)
.await?;
}
let task: SummarizedTaskView = task.into();

View File

@@ -3,6 +3,7 @@
// 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::borrow::Cow;
use std::collections::BTreeMap;
use std::fs::File;
@@ -10,25 +11,41 @@ use actix_web::http::header::CONTENT_TYPE;
use actix_web::HttpRequest;
use bytes::Bytes;
use index_scheduler::IndexScheduler;
use meilisearch_types::enterprise_edition::network::Remote;
use meilisearch_types::error::ResponseError;
use meilisearch_types::tasks::{Origin, RemoteTask, TaskNetwork};
use meilisearch_types::milli::DocumentId;
use meilisearch_types::tasks::enterprise_edition::network::headers::{
PROXY_IMPORT_DOCS_HEADER, PROXY_IMPORT_INDEX_COUNT_HEADER, PROXY_IMPORT_INDEX_HEADER,
PROXY_IMPORT_REMOTE_HEADER, PROXY_IMPORT_TASK_KEY_HEADER, PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER,
PROXY_ORIGIN_NETWORK_VERSION_HEADER, PROXY_ORIGIN_REMOTE_HEADER, PROXY_ORIGIN_TASK_UID_HEADER,
};
use meilisearch_types::tasks::enterprise_edition::network::{
DbTaskNetwork, ImportData, ImportMetadata, Origin, TaskNetwork,
};
use meilisearch_types::tasks::Task;
use reqwest::StatusCode;
use serde::de::DeserializeOwned;
use serde_json::Value;
use uuid::Uuid;
use crate::error::MeilisearchHttpError;
use crate::routes::indexes::enterprise_edition::proxy::error::{
ProxyDocumentChangeError, ReqwestErrorWithoutUrl,
pub use crate::routes::indexes::enterprise_edition::proxy::error::{
ProxyError, ReqwestErrorWithoutUrl,
};
use crate::routes::SummarizedTaskView;
pub enum Body<T: serde::Serialize> {
pub enum Body<T, F>
where
T: serde::Serialize,
F: FnMut(&str, &Remote, &mut T),
{
NdJsonPayload(File),
Inline(T),
Generated(T, F),
None,
}
impl Body<()> {
impl Body<(), fn(&str, &Remote, &mut ())> {
pub fn with_ndjson_payload(file: File) -> Self {
Self::NdJsonPayload(file)
}
@@ -38,7 +55,146 @@ impl Body<()> {
}
}
/// If necessary, proxies the passed request to the network and update the task description.
impl<T> Body<T, fn(&str, &Remote, &mut T)>
where
T: serde::Serialize,
{
pub fn inline(payload: T) -> Self {
Self::Inline(payload)
}
}
impl<T, F> Body<T, F>
where
T: serde::Serialize,
F: FnMut(&str, &Remote, &mut T),
{
pub fn generated(initial: T, f: F) -> Self {
Self::Generated(initial, f)
}
}
impl<T, F> Body<T, F>
where
T: serde::Serialize,
F: FnMut(&str, &Remote, &mut T),
{
pub fn into_bytes_iter(
self,
remotes: impl IntoIterator<Item = (String, Remote)>,
) -> Result<
impl Iterator<Item = (Option<Bytes>, (String, Remote))>,
meilisearch_types::milli::Error,
> {
let bytes = match self {
Body::NdJsonPayload(file) => {
Some(Bytes::from_owner(unsafe { memmap2::Mmap::map(&file)? }))
}
Body::Inline(payload) => {
Some(Bytes::copy_from_slice(&serde_json::to_vec(&payload).unwrap()))
}
Body::None => None,
Body::Generated(mut initial, mut f) => {
return Ok(either::Right(remotes.into_iter().map(move |(name, remote)| {
f(&name, &remote, &mut initial);
let bytes =
Some(Bytes::copy_from_slice(&serde_json::to_vec(&initial).unwrap()));
(bytes, (name, remote))
})));
}
};
Ok(either::Left(std::iter::repeat(bytes).zip(remotes)))
}
pub fn into_bytes(
self,
remote_name: &str,
remote: &Remote,
) -> Result<Option<Bytes>, meilisearch_types::milli::Error> {
Ok(match self {
Body::NdJsonPayload(file) => {
Some(Bytes::from_owner(unsafe { memmap2::Mmap::map(&file)? }))
}
Body::Inline(payload) => {
Some(Bytes::copy_from_slice(&serde_json::to_vec(&payload).unwrap()))
}
Body::None => None,
Body::Generated(mut initial, mut f) => {
f(remote_name, remote, &mut initial);
Some(Bytes::copy_from_slice(&serde_json::to_vec(&initial).unwrap()))
}
})
}
}
/// Parses the header to determine if this task is a duplicate and originates with a remote.
///
/// If not, checks whether this remote is the leader and return `MeilisearchHttpError::NotLeader` if not.
///
/// If there is no leader, returns `Ok(None)`
///
/// # Errors
///
/// - `MeiliearchHttpError::NotLeader`: if the following are true simultaneously:
/// 1. The task originates with the current node
/// 2. There's a declared `leader`
/// 3. The declared leader is **not** the current node
/// - `MeilisearchHttpError::InvalidHeaderValue`: if headers cannot be parsed as a task network.
/// - `MeilisearchHttpError::InconsistentTaskNetwork`: if only some of the headers are present.
pub fn task_network_and_check_leader_and_version(
req: &HttpRequest,
network: &meilisearch_types::enterprise_edition::network::Network,
) -> Result<Option<TaskNetwork>, MeilisearchHttpError> {
let task_network =
match (origin_from_req(req)?, import_data_from_req(req)?, import_metadata_from_req(req)?) {
(Some(network_change), Some(import_from), Some(metadata)) => {
TaskNetwork::Import { import_from, network_change, metadata }
}
(Some(origin), None, None) => TaskNetwork::Origin { origin },
(None, None, None) => {
match (network.leader.as_deref(), network.local.as_deref()) {
// 1. Always allowed if there is no leader
(None, _) => return Ok(None),
// 2. Allowed if the leader is self
(Some(leader), Some(this)) if leader == this => (),
// 3. Any other change is disallowed
(Some(leader), _) => {
return Err(MeilisearchHttpError::NotLeader { leader: leader.to_string() })
}
}
TaskNetwork::Remotes {
remote_tasks: Default::default(),
network_version: network.version,
}
}
// all good cases were matched, so this is always an error
(origin, import_from, metadata) => {
return Err(MeilisearchHttpError::InconsistentTaskNetworkHeaders {
is_missing_origin: origin.is_none(),
is_missing_import: import_from.is_none(),
is_missing_import_metadata: metadata.is_none(),
})
}
};
if task_network.network_version() < network.version {
return Err(MeilisearchHttpError::NetworkVersionTooOld {
received: task_network.network_version(),
expected_at_least: network.version,
});
}
Ok(Some(task_network))
}
/// Updates the task description and, 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
/// has already been proxied.
@@ -48,152 +204,254 @@ impl Body<()> {
/// with the task ids from the task queues of the remotes.
/// - when the request has already been proxied, the custom headers contains information about the remote that created the initial task.
/// This information is copied to the passed task.
pub async fn proxy<T: serde::Serialize>(
///
/// # Returns
///
/// The updated task. The task is read back from the database to avoid erasing concurrent changes.
pub async fn proxy<T, F>(
index_scheduler: &IndexScheduler,
index_uid: &str,
index_uid: Option<&str>,
req: &HttpRequest,
mut task_network: DbTaskNetwork,
network: meilisearch_types::enterprise_edition::network::Network,
body: Body<T>,
body: Body<T, F>,
task: &meilisearch_types::tasks::Task,
) -> Result<(), MeilisearchHttpError> {
match origin_from_req(req)? {
Some(origin) => {
index_scheduler.set_task_network(task.uid, TaskNetwork::Origin { origin })?
) -> Result<Task, MeilisearchHttpError>
where
T: serde::Serialize,
F: FnMut(&str, &Remote, &mut T),
{
if let DbTaskNetwork::Remotes { remote_tasks, network_version } = &mut task_network {
let network_version = *network_version;
let this = network
.local
.as_deref()
.expect("inconsistent `network.sharding` and `network.self`")
.to_owned();
let content_type = match &body {
// for file bodies, force x-ndjson
Body::NdJsonPayload(_) => Some(b"application/x-ndjson".as_slice()),
// otherwise get content type from request
_ => req.headers().get(CONTENT_TYPE).map(|h| h.as_bytes()),
};
let mut in_flight_remote_queries = BTreeMap::new();
let client = reqwest::ClientBuilder::new()
.connect_timeout(std::time::Duration::from_secs(3))
.build()
.unwrap();
let method = from_old_http_method(req.method());
// send payload to all remotes
for (body, (node_name, node)) in body
.into_bytes_iter(network.remotes.into_iter().filter(|(name, _)| name.as_str() != this))
.map_err(|err| {
MeilisearchHttpError::from_milli(err, index_uid.map(ToOwned::to_owned))
})?
{
tracing::trace!(node_name, "proxying task to remote");
let client = client.clone();
let api_key = node.write_api_key;
let this = this.clone();
let method = method.clone();
let path_and_query = req.uri().path_and_query().map(|paq| paq.as_str()).unwrap_or("/");
in_flight_remote_queries.insert(
node_name,
tokio::spawn({
let url = format!("{}{}", node.url, path_and_query);
let url_encoded_this = urlencoding::encode(&this).into_owned();
let url_encoded_task_uid = task.uid.to_string(); // it's url encoded i promize
let content_type = content_type.map(|b| b.to_owned());
let backoff = backoff::ExponentialBackoffBuilder::new()
.with_max_elapsed_time(Some(std::time::Duration::from_secs(25)))
.build();
backoff::future::retry(backoff, move || {
let url = url.clone();
let client = client.clone();
let url_encoded_this = url_encoded_this.clone();
let url_encoded_task_uid = url_encoded_task_uid.clone();
let content_type = content_type.clone();
let body = body.clone();
let api_key = api_key.clone();
let method = method.clone();
async move {
try_proxy(
method,
&url,
content_type.as_deref(),
network_version,
api_key.as_deref(),
&client,
&url_encoded_this,
&url_encoded_task_uid,
body,
)
.await
}
})
}),
);
}
None => {
let this = network
.local
.as_deref()
.expect("inconsistent `network.sharding` and `network.self`")
.to_owned();
let content_type = match &body {
// for file bodies, force x-ndjson
Body::NdJsonPayload(_) => Some(b"application/x-ndjson".as_slice()),
// otherwise get content type from request
_ => req.headers().get(CONTENT_TYPE).map(|h| h.as_bytes()),
};
// wait for all in-flight queries to finish and collect their results
for (node_name, handle) in in_flight_remote_queries {
match handle.await {
Ok(Ok(res)) => {
let task_uid = res.task_uid;
let body = match body {
Body::NdJsonPayload(file) => Some(Bytes::from_owner(unsafe {
memmap2::Mmap::map(&file).map_err(|err| {
MeilisearchHttpError::from_milli(err.into(), Some(index_uid.to_owned()))
})?
})),
Body::Inline(payload) => {
Some(Bytes::copy_from_slice(&serde_json::to_vec(&payload).unwrap()))
remote_tasks.insert(node_name, Ok(task_uid).into());
}
Body::None => None,
};
let mut in_flight_remote_queries = BTreeMap::new();
let client = reqwest::ClientBuilder::new()
.connect_timeout(std::time::Duration::from_secs(3))
.build()
.unwrap();
let method = from_old_http_method(req.method());
// send payload to all remotes
for (node_name, node) in
network.remotes.into_iter().filter(|(name, _)| name.as_str() != this)
{
let body = body.clone();
let client = client.clone();
let api_key = node.write_api_key;
let this = this.clone();
let method = method.clone();
let path_and_query =
req.uri().path_and_query().map(|paq| paq.as_str()).unwrap_or("/");
in_flight_remote_queries.insert(
node_name,
tokio::spawn({
let url = format!("{}{}", node.url, path_and_query);
let url_encoded_this = urlencoding::encode(&this).into_owned();
let url_encoded_task_uid = task.uid.to_string(); // it's url encoded i promize
let content_type = content_type.map(|b| b.to_owned());
let backoff = backoff::ExponentialBackoffBuilder::new()
.with_max_elapsed_time(Some(std::time::Duration::from_secs(25)))
.build();
backoff::future::retry(backoff, move || {
let url = url.clone();
let client = client.clone();
let url_encoded_this = url_encoded_this.clone();
let url_encoded_task_uid = url_encoded_task_uid.clone();
let content_type = content_type.clone();
let body = body.clone();
let api_key = api_key.clone();
let method = method.clone();
async move {
try_proxy(
method,
&url,
content_type.as_deref(),
api_key.as_deref(),
&client,
&url_encoded_this,
&url_encoded_task_uid,
body,
)
.await
}
})
}),
);
}
// wait for all in-flight queries to finish and collect their results
let mut remote_tasks: BTreeMap<String, RemoteTask> = BTreeMap::new();
for (node_name, handle) in in_flight_remote_queries {
match handle.await {
Ok(Ok(res)) => {
let task_uid = res.task_uid;
remote_tasks.insert(node_name, Ok(task_uid).into());
}
Ok(Err(error)) => {
remote_tasks.insert(node_name, Err(error.as_response_error()).into());
}
Err(panic) => match panic.try_into_panic() {
Ok(panic) => {
let msg = match panic.downcast_ref::<&'static str>() {
Some(s) => *s,
None => match panic.downcast_ref::<String>() {
Some(s) => &s[..],
None => "Box<dyn Any>",
},
};
remote_tasks.insert(
node_name,
Err(ResponseError::from_msg(
msg.to_string(),
meilisearch_types::error::Code::Internal,
))
.into(),
);
}
Err(_) => {
tracing::error!("proxy task was unexpectedly cancelled")
}
},
Ok(Err(error)) => {
remote_tasks.insert(node_name, Err(error.as_response_error()).into());
}
Err(panic) => match panic.try_into_panic() {
Ok(panic) => {
let msg = match panic.downcast_ref::<&'static str>() {
Some(s) => *s,
None => match panic.downcast_ref::<String>() {
Some(s) => &s[..],
None => "Box<dyn Any>",
},
};
remote_tasks.insert(
node_name,
Err(ResponseError::from_msg(
msg.to_string(),
meilisearch_types::error::Code::Internal,
))
.into(),
);
}
Err(_) => {
tracing::error!("proxy task was unexpectedly cancelled")
}
},
}
// edit details to contain the return values from the remotes
index_scheduler.set_task_network(task.uid, TaskNetwork::Remotes { remote_tasks })?;
}
}
Ok(())
Ok(index_scheduler.set_task_network(task.uid, task_network)?)
}
pub async fn send_request<T, F, U>(
path_and_query: &str,
method: reqwest::Method,
content_type: Option<String>,
body: Body<T, F>,
remote_name: &str,
remote: &Remote,
) -> Result<U, ProxyError>
where
T: serde::Serialize,
F: FnMut(&str, &Remote, &mut T),
U: DeserializeOwned,
{
let content_type = match &body {
// for file bodies, force x-ndjson
Body::NdJsonPayload(_) => Some("application/x-ndjson".into()),
// otherwise get content type from request
_ => content_type,
};
let body = body.into_bytes(remote_name, remote).map_err(Box::new)?;
let client = reqwest::ClientBuilder::new()
.connect_timeout(std::time::Duration::from_secs(3))
.build()
.unwrap();
let url = format!("{}{}", remote.url, path_and_query);
// send payload to remote
tracing::trace!(remote_name, "sending request to remote");
let api_key = remote.write_api_key.clone();
let backoff = backoff::ExponentialBackoffBuilder::new()
.with_max_elapsed_time(Some(std::time::Duration::from_secs(25)))
.build();
backoff::future::retry(backoff, move || {
let url = url.clone();
let client = client.clone();
let content_type = content_type.clone();
let body = body.clone();
let api_key = api_key.clone();
let method = method.clone();
async move {
let request = client.request(method, url).timeout(std::time::Duration::from_secs(30));
let request = if let Some(body) = body { request.body(body) } else { request };
let request =
if let Some(api_key) = api_key { request.bearer_auth(api_key) } else { request };
let request = if let Some(content_type) = content_type {
request.header(CONTENT_TYPE.as_str(), content_type)
} else {
request
};
let response = request.send().await;
let response = match response {
Ok(response) => response,
Err(error) if error.is_timeout() => {
return Err(backoff::Error::transient(ProxyError::Timeout))
}
Err(error) => {
return Err(backoff::Error::transient(ProxyError::CouldNotSendRequest(
ReqwestErrorWithoutUrl::new(error),
)))
}
};
match response.status() {
status_code if status_code.is_success() => (),
StatusCode::UNAUTHORIZED | StatusCode::FORBIDDEN => {
return Err(backoff::Error::Permanent(ProxyError::AuthenticationError))
}
status_code if status_code.is_client_error() => {
let response = parse_error(response).await;
return Err(backoff::Error::Permanent(ProxyError::BadRequest {
status_code,
response,
}));
}
status_code if status_code.is_server_error() => {
let response = parse_error(response).await;
return Err(backoff::Error::transient(ProxyError::RemoteError {
status_code,
response,
}));
}
status_code => {
tracing::warn!(
status_code = status_code.as_u16(),
"remote replied with unexpected status code"
);
}
}
let response: U = match parse_response(response).await {
Ok(response) => response,
Err(response) => {
return Err(backoff::Error::transient(ProxyError::CouldNotParseResponse {
response,
}))
}
};
Ok(response)
}
})
.await
}
fn from_old_http_method(method: &actix_http::Method) -> reqwest::Method {
@@ -216,16 +474,18 @@ async fn try_proxy(
method: reqwest::Method,
url: &str,
content_type: Option<&[u8]>,
network_version: Uuid,
api_key: Option<&str>,
client: &reqwest::Client,
url_encoded_this: &str,
url_encoded_task_uid: &str,
body: Option<Bytes>,
) -> Result<SummarizedTaskView, backoff::Error<ProxyDocumentChangeError>> {
) -> Result<SummarizedTaskView, backoff::Error<ProxyError>> {
let request = client.request(method, url).timeout(std::time::Duration::from_secs(30));
let request = if let Some(body) = body { request.body(body) } else { request };
let request = if let Some(api_key) = api_key { request.bearer_auth(api_key) } else { request };
let request = request.header(PROXY_ORIGIN_TASK_UID_HEADER, url_encoded_task_uid);
let request = request.header(PROXY_ORIGIN_NETWORK_VERSION_HEADER, &network_version.to_string());
let request = request.header(PROXY_ORIGIN_REMOTE_HEADER, url_encoded_this);
let request = if let Some(content_type) = content_type {
request.header(CONTENT_TYPE.as_str(), content_type)
@@ -237,10 +497,10 @@ async fn try_proxy(
let response = match response {
Ok(response) => response,
Err(error) if error.is_timeout() => {
return Err(backoff::Error::transient(ProxyDocumentChangeError::Timeout))
return Err(backoff::Error::transient(ProxyError::Timeout))
}
Err(error) => {
return Err(backoff::Error::transient(ProxyDocumentChangeError::CouldNotSendRequest(
return Err(backoff::Error::transient(ProxyError::CouldNotSendRequest(
ReqwestErrorWithoutUrl::new(error),
)))
}
@@ -249,18 +509,18 @@ async fn try_proxy(
match response.status() {
status_code if status_code.is_success() => (),
StatusCode::UNAUTHORIZED | StatusCode::FORBIDDEN => {
return Err(backoff::Error::Permanent(ProxyDocumentChangeError::AuthenticationError))
return Err(backoff::Error::Permanent(ProxyError::AuthenticationError))
}
status_code if status_code.is_client_error() => {
let response = parse_error(response).await;
return Err(backoff::Error::Permanent(ProxyDocumentChangeError::BadRequest {
return Err(backoff::Error::Permanent(ProxyError::BadRequest {
status_code,
response,
}));
}
status_code if status_code.is_server_error() => {
let response = parse_error(response).await;
return Err(backoff::Error::transient(ProxyDocumentChangeError::RemoteError {
return Err(backoff::Error::transient(ProxyError::RemoteError {
status_code,
response,
}));
@@ -276,9 +536,7 @@ async fn try_proxy(
let response = match parse_response(response).await {
Ok(response) => response,
Err(response) => {
return Err(backoff::Error::transient(
ProxyDocumentChangeError::CouldNotParseResponse { response },
))
return Err(backoff::Error::transient(ProxyError::CouldNotParseResponse { response }))
}
};
@@ -316,11 +574,11 @@ async fn parse_response<T: DeserializeOwned>(
}
mod error {
use meilisearch_types::error::ResponseError;
use meilisearch_types::error::{ErrorCode as _, ResponseError};
use reqwest::StatusCode;
#[derive(Debug, thiserror::Error)]
pub enum ProxyDocumentChangeError {
pub enum ProxyError {
#[error("{0}")]
CouldNotSendRequest(ReqwestErrorWithoutUrl),
#[error("could not authenticate against the remote host\n - hint: check that the remote instance was registered with a valid API key having the `documents.add` action")]
@@ -336,19 +594,25 @@ mod error {
Timeout,
#[error("remote host responded with code {}{}", status_code.as_u16(), response_from_remote(response))]
RemoteError { status_code: StatusCode, response: Result<String, ReqwestErrorWithoutUrl> },
#[error("error while preparing the request: {error}")]
Milli {
#[from]
error: Box<meilisearch_types::milli::Error>,
},
}
impl ProxyDocumentChangeError {
impl ProxyError {
pub fn as_response_error(&self) -> ResponseError {
use meilisearch_types::error::Code;
let message = self.to_string();
let code = match self {
ProxyDocumentChangeError::CouldNotSendRequest(_) => Code::RemoteCouldNotSendRequest,
ProxyDocumentChangeError::AuthenticationError => Code::RemoteInvalidApiKey,
ProxyDocumentChangeError::BadRequest { .. } => Code::RemoteBadRequest,
ProxyDocumentChangeError::Timeout => Code::RemoteTimeout,
ProxyDocumentChangeError::RemoteError { .. } => Code::RemoteRemoteError,
ProxyDocumentChangeError::CouldNotParseResponse { .. } => Code::RemoteBadResponse,
ProxyError::CouldNotSendRequest(_) => Code::RemoteCouldNotSendRequest,
ProxyError::AuthenticationError => Code::RemoteInvalidApiKey,
ProxyError::BadRequest { .. } => Code::RemoteBadRequest,
ProxyError::Timeout => Code::RemoteTimeout,
ProxyError::RemoteError { .. } => Code::RemoteRemoteError,
ProxyError::CouldNotParseResponse { .. } => Code::RemoteBadResponse,
ProxyError::Milli { error } => error.error_code(),
};
ResponseError::from_msg(message, code)
}
@@ -375,25 +639,23 @@ mod error {
}
}
pub const PROXY_ORIGIN_REMOTE_HEADER: &str = "Meili-Proxy-Origin-Remote";
pub const PROXY_ORIGIN_TASK_UID_HEADER: &str = "Meili-Proxy-Origin-TaskUid";
pub fn origin_from_req(req: &HttpRequest) -> Result<Option<Origin>, MeilisearchHttpError> {
let (remote_name, task_uid) = match (
let (remote_name, task_uid, network_version) = match (
req.headers().get(PROXY_ORIGIN_REMOTE_HEADER),
req.headers().get(PROXY_ORIGIN_TASK_UID_HEADER),
req.headers().get(PROXY_ORIGIN_NETWORK_VERSION_HEADER),
) {
(None, None) => return Ok(None),
(None, Some(_)) => {
(None, None, _) => return Ok(None),
(None, Some(_), _) => {
return Err(MeilisearchHttpError::InconsistentOriginHeaders { is_remote_missing: true })
}
(Some(_), None) => {
(Some(_), None, _) => {
return Err(MeilisearchHttpError::InconsistentOriginHeaders {
is_remote_missing: false,
})
}
(Some(remote_name), Some(task_uid)) => (
urlencoding::decode(remote_name.to_str().map_err(|err| {
(Some(remote_name), Some(task_uid), network_version) => {
let remote_name = urlencoding::decode(remote_name.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_ORIGIN_REMOTE_HEADER,
msg: format!("while parsing remote name as UTF-8: {err}"),
@@ -402,8 +664,8 @@ pub fn origin_from_req(req: &HttpRequest) -> Result<Option<Origin>, MeilisearchH
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_ORIGIN_REMOTE_HEADER,
msg: format!("while URL-decoding remote name: {err}"),
})?,
urlencoding::decode(task_uid.to_str().map_err(|err| {
})?;
let task_uid = urlencoding::decode(task_uid.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_ORIGIN_TASK_UID_HEADER,
msg: format!("while parsing task UID as UTF-8: {err}"),
@@ -412,15 +674,235 @@ pub fn origin_from_req(req: &HttpRequest) -> Result<Option<Origin>, MeilisearchH
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_ORIGIN_TASK_UID_HEADER,
msg: format!("while URL-decoding task UID: {err}"),
})?,
),
})?;
let network_version = match network_version {
Some(network_version) => {
urlencoding::decode(network_version.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_ORIGIN_NETWORK_VERSION_HEADER,
msg: format!("while parsing network version as UTF-8: {err}"),
}
})?)
.map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_ORIGIN_NETWORK_VERSION_HEADER,
msg: format!("while URL-decoding network version: {err}"),
}
})?
}
None => Cow::Borrowed("0"),
};
(remote_name, task_uid, network_version)
}
};
let task_uid: usize =
let task_uid: u32 =
task_uid.parse().map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_ORIGIN_TASK_UID_HEADER,
msg: format!("while parsing the task UID as an integer: {err}"),
})?;
Ok(Some(Origin { remote_name: remote_name.into_owned(), task_uid }))
let network_version: Uuid = Uuid::parse_str(&network_version).map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_ORIGIN_NETWORK_VERSION_HEADER,
msg: format!("while parsing the network version as an UUID: {err}"),
}
})?;
Ok(Some(Origin { remote_name: remote_name.into_owned(), task_uid, network_version }))
}
pub fn import_data_from_req(req: &HttpRequest) -> Result<Option<ImportData>, MeilisearchHttpError> {
let (remote_name, index_name, documents) = match (
req.headers().get(PROXY_IMPORT_REMOTE_HEADER),
req.headers().get(PROXY_IMPORT_INDEX_HEADER),
req.headers().get(PROXY_IMPORT_DOCS_HEADER),
) {
(None, None, None) => return Ok(None),
(Some(remote_name), Some(index_name), Some(documents)) => {
let remote_name = urlencoding::decode(remote_name.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_REMOTE_HEADER,
msg: format!("while parsing import remote name as UTF-8: {err}"),
}
})?)
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_REMOTE_HEADER,
msg: format!("while URL-decoding import remote name: {err}"),
})?;
let index_name = urlencoding::decode(index_name.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_INDEX_HEADER,
msg: format!("while parsing import index name as UTF-8: {err}"),
}
})?)
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_INDEX_HEADER,
msg: format!("while URL-decoding import index name: {err}"),
})?;
let documents = urlencoding::decode(documents.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_DOCS_HEADER,
msg: format!("while parsing documents as UTF-8: {err}"),
}
})?)
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_DOCS_HEADER,
msg: format!("while URL-decoding documents: {err}"),
})?;
(remote_name, Some(index_name), documents)
}
(Some(remote_name), None, Some(documents)) => {
let remote_name = urlencoding::decode(remote_name.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_REMOTE_HEADER,
msg: format!("while parsing import remote name as UTF-8: {err}"),
}
})?)
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_REMOTE_HEADER,
msg: format!("while URL-decoding import remote name: {err}"),
})?;
let documents = urlencoding::decode(documents.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_DOCS_HEADER,
msg: format!("while parsing documents as UTF-8: {err}"),
}
})?)
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_DOCS_HEADER,
msg: format!("while URL-decoding documents: {err}"),
})?;
(remote_name, None, documents)
}
// catch-all pattern that has to contain an inconsistency since we already matched (None, None, None) and (Some, Some, Some)
(remote_name, index_name, documents) => {
return Err(MeilisearchHttpError::InconsistentImportHeaders {
is_remote_missing: remote_name.is_none(),
is_index_missing: index_name.is_none(),
is_docs_missing: documents.is_none(),
})
}
};
let document_count: u64 =
documents.parse().map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_DOCS_HEADER,
msg: format!("while parsing the documents as an integer: {err}"),
})?;
Ok(Some(ImportData {
remote_name: remote_name.to_string(),
index_name: index_name.map(|index_name| index_name.to_string()),
document_count,
}))
}
pub fn import_metadata_from_req(
req: &HttpRequest,
) -> Result<Option<ImportMetadata>, MeilisearchHttpError> {
let (index_count, task_key, total_index_documents) = match (
req.headers().get(PROXY_IMPORT_INDEX_COUNT_HEADER),
req.headers().get(PROXY_IMPORT_TASK_KEY_HEADER),
req.headers().get(PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER),
) {
(None, None, None) => return Ok(None),
(Some(index_count), Some(task_key), Some(total_index_documents)) => {
let index_count = urlencoding::decode(index_count.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_REMOTE_HEADER,
msg: format!("while parsing import index count as UTF-8: {err}"),
}
})?)
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_INDEX_COUNT_HEADER,
msg: format!("while URL-decoding import index count: {err}"),
})?;
let task_key = urlencoding::decode(task_key.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_TASK_KEY_HEADER,
msg: format!("while parsing import task key as UTF-8: {err}"),
}
})?)
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_TASK_KEY_HEADER,
msg: format!("while URL-decoding import task key: {err}"),
})?;
let total_index_documents =
urlencoding::decode(total_index_documents.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER,
msg: format!("while parsing total index documents as UTF-8: {err}"),
}
})?)
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER,
msg: format!("while URL-decoding total index documents: {err}"),
})?;
(index_count, Some(task_key), total_index_documents)
}
(Some(index_count), None, Some(total_index_documents)) => {
let index_count = urlencoding::decode(index_count.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_REMOTE_HEADER,
msg: format!("while parsing import index count as UTF-8: {err}"),
}
})?)
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_INDEX_COUNT_HEADER,
msg: format!("while URL-decoding import index count: {err}"),
})?;
let total_index_documents =
urlencoding::decode(total_index_documents.to_str().map_err(|err| {
MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER,
msg: format!("while parsing total index documents as UTF-8: {err}"),
}
})?)
.map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER,
msg: format!("while URL-decoding total index documents: {err}"),
})?;
(index_count, None, total_index_documents)
}
// catch-all pattern that has to contain an inconsistency since we already matched (None, None, None) and (Some, Some, Some)
(index_count, task_key, total_index_documents) => {
return Err(MeilisearchHttpError::InconsistentImportMetadataHeaders {
is_index_count_missing: index_count.is_none(),
is_task_key_missing: task_key.is_none(),
is_total_index_documents_missing: total_index_documents.is_none(),
})
}
};
let index_count: u64 =
index_count.parse().map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_INDEX_COUNT_HEADER,
msg: format!("while parsing the index count as an integer: {err}"),
})?;
let task_key = task_key
.map(|task_key| {
let task_key: Result<DocumentId, _> =
task_key.parse().map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_TASK_KEY_HEADER,
msg: format!("while parsing import task key as an integer: {err}"),
});
task_key
})
.transpose()?;
let total_index_documents: u64 =
total_index_documents.parse().map_err(|err| MeilisearchHttpError::InvalidHeaderValue {
header_name: PROXY_IMPORT_TOTAL_INDEX_DOCS_HEADER,
msg: format!("while parsing the total index documents as an integer: {err}"),
})?;
Ok(Some(ImportMetadata { index_count, task_key, total_index_documents }))
}

View File

@@ -30,7 +30,7 @@ use crate::Opt;
pub mod compact;
pub mod documents;
mod enterprise_edition;
pub mod enterprise_edition;
pub mod facet_search;
pub mod search;
mod search_analytics;
@@ -41,7 +41,9 @@ mod settings_analytics;
pub mod similar;
mod similar_analytics;
pub use enterprise_edition::proxy::{PROXY_ORIGIN_REMOTE_HEADER, PROXY_ORIGIN_TASK_UID_HEADER};
pub use meilisearch_types::tasks::enterprise_edition::network::headers::{
PROXY_ORIGIN_REMOTE_HEADER, PROXY_ORIGIN_TASK_UID_HEADER,
};
#[derive(OpenApi)]
#[openapi(

View File

@@ -17,6 +17,9 @@ use super::settings_analytics::*;
use crate::analytics::Analytics;
use crate::extractors::authentication::policies::*;
use crate::extractors::authentication::GuardedData;
use crate::routes::indexes::enterprise_edition::proxy::{
proxy, task_network_and_check_leader_and_version, Body,
};
use crate::routes::{get_task_id, is_dry_run, SummarizedTaskView};
use crate::Opt;
@@ -76,14 +79,13 @@ macro_rules! make_setting_route {
use meilisearch_types::index_uid::IndexUid;
use meilisearch_types::milli::update::Setting;
use meilisearch_types::settings::{settings, Settings};
use meilisearch_types::tasks::KindWithContent;
use tracing::debug;
use $crate::analytics::Analytics;
use $crate::extractors::authentication::policies::*;
use $crate::extractors::authentication::GuardedData;
use $crate::extractors::sequential_extractor::SeqHandler;
use $crate::Opt;
use $crate::routes::{is_dry_run, get_task_id, SummarizedTaskView};
use $crate::routes::SummarizedTaskView;
#[allow(unused_imports)]
use super::*;
@@ -130,21 +132,7 @@ macro_rules! make_setting_route {
let new_settings = Settings { $attr: Setting::Reset.into(), ..Default::default() };
let allow_index_creation =
index_scheduler.filters().allow_index_creation(&index_uid);
let task = KindWithContent::SettingsUpdate {
index_uid: index_uid.to_string(),
new_settings: Box::new(new_settings),
is_deletion: true,
allow_index_creation,
};
let uid = get_task_id(&req, &opt)?;
let dry_run = is_dry_run(&req, &opt)?;
let task: SummarizedTaskView =
tokio::task::spawn_blocking(move || index_scheduler.register(task, uid, dry_run))
.await??
.into();
let task = register_new_settings(new_settings, true, index_scheduler, &req, index_uid, opt).await?;
debug!(returns = ?task, "Delete settings");
Ok(HttpResponse::Accepted().json(task))
@@ -216,21 +204,7 @@ macro_rules! make_setting_route {
&index_scheduler,
)?;
let allow_index_creation =
index_scheduler.filters().allow_index_creation(&index_uid);
let task = KindWithContent::SettingsUpdate {
index_uid: index_uid.to_string(),
new_settings: Box::new(new_settings),
is_deletion: false,
allow_index_creation,
};
let uid = get_task_id(&req, &opt)?;
let dry_run = is_dry_run(&req, &opt)?;
let task: SummarizedTaskView =
tokio::task::spawn_blocking(move || index_scheduler.register(task, uid, dry_run))
.await??
.into();
let task = register_new_settings(new_settings, false, index_scheduler, &req, index_uid, opt).await?;
debug!(returns = ?task, "Update settings");
Ok(HttpResponse::Accepted().json(task))
@@ -571,12 +545,12 @@ pub async fn update_all(
index_uid: web::Path<String>,
body: AwebJson<Settings<Unchecked>, DeserrJsonError>,
req: HttpRequest,
opt: web::Data<Opt>,
analytics: web::Data<Analytics>,
opt: Data<Opt>,
analytics: Data<Analytics>,
) -> Result<HttpResponse, ResponseError> {
let index_uid = IndexUid::try_from(index_uid.into_inner())?;
let new_settings = body.into_inner();
let new_settings: Settings<Unchecked> = body.into_inner();
debug!(parameters = ?new_settings, "Update all settings");
let new_settings = validate_settings(new_settings, &index_scheduler)?;
@@ -626,25 +600,57 @@ pub async fn update_all(
&req,
);
let allow_index_creation = index_scheduler.filters().allow_index_creation(&index_uid);
let index_uid = IndexUid::try_from(index_uid.into_inner())?.into_inner();
let task = KindWithContent::SettingsUpdate {
index_uid,
new_settings: Box::new(new_settings),
is_deletion: false,
allow_index_creation,
};
let uid = get_task_id(&req, &opt)?;
let dry_run = is_dry_run(&req, &opt)?;
let task: SummarizedTaskView =
tokio::task::spawn_blocking(move || index_scheduler.register(task, uid, dry_run))
.await??
.into();
let task =
register_new_settings(new_settings, false, index_scheduler, &req, index_uid, opt).await?;
debug!(returns = ?task, "Update all settings");
Ok(HttpResponse::Accepted().json(task))
}
async fn register_new_settings(
new_settings: Settings<Unchecked>,
is_deletion: bool,
index_scheduler: GuardedData<ActionPolicy<{ actions::SETTINGS_UPDATE }>, Data<IndexScheduler>>,
req: &HttpRequest,
index_uid: IndexUid,
opt: Data<Opt>,
) -> Result<SummarizedTaskView, ResponseError> {
let network = index_scheduler.network();
let task_network = task_network_and_check_leader_and_version(req, &network)?;
let allow_index_creation = index_scheduler.filters().allow_index_creation(&index_uid);
let index_uid = IndexUid::try_from(index_uid.into_inner())?.into_inner();
let task = KindWithContent::SettingsUpdate {
index_uid: index_uid.clone(),
new_settings: Box::new(new_settings.clone()),
is_deletion,
allow_index_creation,
};
let uid = get_task_id(req, &opt)?;
let dry_run = is_dry_run(req, &opt)?;
let scheduler = index_scheduler.clone();
let mut task = tokio::task::spawn_blocking(move || {
scheduler.register_with_custom_metadata(task, uid, None, dry_run, task_network)
})
.await??;
if let Some(task_network) = task.network.take() {
proxy(
&index_scheduler,
Some(&index_uid),
req,
task_network,
network,
Body::inline(new_settings),
&task,
)
.await?;
}
Ok(task.into())
}
#[utoipa::path(
get,
path = "{indexUid}/settings",
@@ -731,20 +737,8 @@ pub async fn delete_all(
let new_settings = Settings::cleared().into_unchecked();
let allow_index_creation = index_scheduler.filters().allow_index_creation(&index_uid);
let index_uid = IndexUid::try_from(index_uid.into_inner())?.into_inner();
let task = KindWithContent::SettingsUpdate {
index_uid,
new_settings: Box::new(new_settings),
is_deletion: true,
allow_index_creation,
};
let uid = get_task_id(&req, &opt)?;
let dry_run = is_dry_run(&req, &opt)?;
let task: SummarizedTaskView =
tokio::task::spawn_blocking(move || index_scheduler.register(task, uid, dry_run))
.await??
.into();
let task =
register_new_settings(new_settings, true, index_scheduler, &req, index_uid, opt).await?;
debug!(returns = ?task, "Delete all settings");
Ok(HttpResponse::Accepted().json(task))

View File

@@ -4,25 +4,36 @@ use actix_web::web::{self, Data};
use actix_web::{HttpRequest, HttpResponse};
use deserr::actix_web::AwebJson;
use deserr::Deserr;
use index_scheduler::IndexScheduler;
use futures::TryStreamExt;
use index_scheduler::{IndexScheduler, Query, RoFeatures};
use itertools::{EitherOrBoth, Itertools};
use meilisearch_auth::AuthFilter;
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,
InvalidNetworkLeader, InvalidNetworkRemotes, InvalidNetworkSearchApiKey, InvalidNetworkSelf,
InvalidNetworkUrl, InvalidNetworkWriteApiKey,
};
use meilisearch_types::error::ResponseError;
use meilisearch_types::error::{Code, ResponseError};
use meilisearch_types::features::RuntimeTogglableFeatures;
use meilisearch_types::keys::actions;
use meilisearch_types::milli::update::Setting;
use meilisearch_types::tasks::enterprise_edition::network::{
headers, NetworkTopologyChange, Origin, TaskNetwork,
};
use meilisearch_types::tasks::KindWithContent;
use serde::Serialize;
use tracing::debug;
use utoipa::{OpenApi, ToSchema};
use crate::analytics::{Aggregate, Analytics};
use crate::error::MeilisearchHttpError;
use crate::extractors::authentication::policies::ActionPolicy;
use crate::extractors::authentication::GuardedData;
use crate::extractors::sequential_extractor::SeqHandler;
use crate::routes::indexes::enterprise_edition::proxy::{self, proxy, Body, ProxyError};
use crate::routes::tasks::AllTasks;
use crate::routes::SummarizedTaskView;
#[derive(OpenApi)]
#[openapi(
@@ -83,7 +94,7 @@ async fn get_network(
Ok(HttpResponse::Ok().json(network))
}
#[derive(Debug, Deserr, ToSchema, Serialize)]
#[derive(Clone, Debug, Deserr, ToSchema, Serialize)]
#[deserr(error = DeserrJsonError<InvalidNetworkRemotes>, rename_all = camelCase, deny_unknown_fields)]
#[serde(rename_all = "camelCase")]
#[schema(rename_all = "camelCase")]
@@ -106,12 +117,19 @@ pub struct Remote {
pub write_api_key: Setting<String>,
}
#[derive(Debug, Deserr, ToSchema, Serialize)]
#[derive(Clone, 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"))]
#[schema(value_type = Option<BTreeMap<String, Remote>>, example = json!({
"ms-00": {
"url": "http://localhost:7700"
},
"ms-01": {
"url": "http://localhost:7701"
}
}))]
#[deserr(default, error = DeserrJsonError<InvalidNetworkRemotes>)]
#[serde(default)]
pub remotes: Setting<BTreeMap<String, Option<Remote>>>,
@@ -119,10 +137,21 @@ pub struct Network {
#[serde(default, rename = "self")]
#[deserr(default, rename = "self", error = DeserrJsonError<InvalidNetworkSelf>)]
pub local: Setting<String>,
#[schema(value_type = Option<bool>, example = json!(true))]
#[schema(value_type = Option<String>, example = json!("ms-00"))]
#[serde(default)]
#[deserr(default, error = DeserrJsonError<InvalidNetworkSharding>)]
pub sharding: Setting<bool>,
#[deserr(default, error = DeserrJsonError<InvalidNetworkLeader>)]
pub leader: Setting<String>,
#[schema(value_type = Option<BTreeMap<String, Remote>>, example = json!({
"ms-00": {
"url": "http://localhost:7700"
},
"ms-01": {
"url": "http://localhost:7701"
}
}))]
#[deserr(default, error = DeserrJsonError<InvalidNetworkRemotes>)]
#[serde(default)]
pub previous_remotes: Setting<BTreeMap<String, Option<Remote>>>,
}
impl Remote {
@@ -207,29 +236,359 @@ async fn patch_network(
) -> Result<HttpResponse, ResponseError> {
index_scheduler.features().check_network("Using the /network route")?;
match (
proxy::origin_from_req(&req)?,
proxy::import_data_from_req(&req)?,
proxy::import_metadata_from_req(&req)?,
) {
(Some(origin), None, None) => {
patch_network_with_origin(index_scheduler, new_network, req, origin, analytics).await
}
(None, None, None) => {
patch_network_without_origin(index_scheduler, new_network, req, analytics).await
}
(Some(origin), Some(import_data), Some(metadata)) => {
if metadata.index_count == 0 {
tokio::task::spawn_blocking(move || {
index_scheduler.network_no_index_for_remote(import_data.remote_name, origin)
})
.await
.map_err(|e| ResponseError::from_msg(e.to_string(), Code::Internal))??;
Ok(HttpResponse::Ok().finish())
} else {
Err(MeilisearchHttpError::InvalidHeaderValue {
header_name: headers::PROXY_IMPORT_INDEX_COUNT_HEADER,
msg: format!("Expected 0 indexes, got `{}`", metadata.index_count),
}
.into())
}
}
(origin, import_data, metadata) => {
Err(MeilisearchHttpError::InconsistentTaskNetworkHeaders {
is_missing_origin: origin.is_none(),
is_missing_import: import_data.is_none(),
is_missing_import_metadata: metadata.is_none(),
}
.into())
}
}
}
async fn patch_network_without_origin(
index_scheduler: GuardedData<ActionPolicy<{ actions::NETWORK_UPDATE }>, Data<IndexScheduler>>,
new_network: AwebJson<Network, DeserrJsonError>,
req: HttpRequest,
analytics: Data<Analytics>,
) -> Result<HttpResponse, ResponseError> {
let new_network = new_network.0;
let old_network = index_scheduler.network();
debug!(parameters = ?new_network, "Patch network");
if !matches!(new_network.previous_remotes, Setting::NotSet) {
return Err(MeilisearchHttpError::UnexpectedNetworkPreviousRemotes.into());
}
let merged_network = merge_networks(old_network.clone(), new_network)?;
// When a network task must be created, perform some sanity checks against common errors:
// - missing experimental feature on an host from the network
// - a network task is already enqueued
//
// These checks are by no mean perfect (they are not atomic since the network is involved), but they should
// help preventing a bad situation.
if merged_network.leader.is_some() {
let query = Query {
statuses: Some(vec![
meilisearch_types::tasks::Status::Enqueued,
meilisearch_types::tasks::Status::Processing,
]),
types: Some(vec![meilisearch_types::tasks::Kind::NetworkTopologyChange]),
..Default::default()
};
let filters = AuthFilter::default();
let (tasks, _) = index_scheduler.get_task_ids_from_authorized_indexes(&query, &filters)?;
if let Some(first) = tasks.min() {
return Err(MeilisearchHttpError::UnprocessedNetworkTask {
remote: None,
task_uid: first,
}
.into());
}
futures::stream::iter(
old_network
.remotes
.iter()
.merge_join_by(merged_network.remotes.iter(), |(left, _), (right, _)| {
left.cmp(right)
})
.map(|eob| -> Result<_, ResponseError> {
Ok(async move {
let (remote_name, remote, allow_unreachable) = match eob {
EitherOrBoth::Both(_, (remote_name, remote))
| EitherOrBoth::Right((remote_name, remote)) => {
(remote_name, remote, false)
}
EitherOrBoth::Left((remote_name, remote)) => {
(remote_name, remote, true)
}
};
{
// 1. check that the experimental feature is enabled
let remote_features: RuntimeTogglableFeatures = match proxy::send_request(
"/experimental-features",
reqwest::Method::GET,
None,
Body::none(),
remote_name,
remote,
)
.await {
Ok(remote_features) => remote_features,
Err(ProxyError::Timeout | ProxyError::CouldNotSendRequest(_)) if allow_unreachable => {
return Ok(())
},
Err(err) => return Err(err.as_response_error()),
};
let remote_features =
RoFeatures::from_runtime_features(remote_features);
remote_features
.check_network("receiving a proxied network task")
.map_err(|error| MeilisearchHttpError::RemoteIndexScheduler {
remote: remote_name.to_owned(),
error,
})?;
// 2. check whether there are any unfinished network task
let network_tasks: AllTasks = match proxy::send_request(
"/tasks?types=networkTopologyChange&statuses=enqueued,processing&limit=1",
reqwest::Method::GET,
None,
Body::none(),
remote_name,
remote).await {
Ok(network_tasks) => network_tasks,
Err(ProxyError::Timeout | ProxyError::CouldNotSendRequest(_)) if allow_unreachable => {
return Ok(())
},
Err(err) => return Err(err.as_response_error()),
};
if let [first, ..] = network_tasks.results.as_slice() {
return Err(ResponseError::from(
MeilisearchHttpError::UnprocessedNetworkTask {
remote: Some(remote_name.to_owned()),
task_uid: first.uid,
},
));
}
}
Ok(())
})
}),
)
.try_buffer_unordered(40)
.try_collect::<()>()
.await?;
}
index_scheduler.put_network(merged_network.clone())?;
analytics.publish(
PatchNetworkAnalytics {
network_size: merged_network.remotes.len(),
network_has_self: merged_network.local.is_some(),
},
&req,
);
if merged_network.leader.is_some() {
let network_topology_change =
NetworkTopologyChange::new(old_network.clone(), merged_network.clone());
let task = KindWithContent::NetworkTopologyChange(network_topology_change);
let mut task = {
let index_scheduler = index_scheduler.clone();
tokio::task::spawn_blocking(move || {
index_scheduler.register_with_custom_metadata(
task,
None,
None,
false,
Some(TaskNetwork::Remotes {
remote_tasks: Default::default(),
network_version: merged_network.version,
}),
)
})
.await??
};
let mut proxied_network = Network {
remotes: Setting::Set(to_settings_remotes(&merged_network.remotes)),
local: Setting::NotSet,
leader: Setting::some_or_not_set(merged_network.leader.clone()),
previous_remotes: Setting::Set(to_settings_remotes(&old_network.remotes)),
};
let mut deleted_network = old_network;
let deleted_remotes = &mut deleted_network.remotes;
deleted_remotes.retain(|node, _| !merged_network.remotes.contains_key(node));
// proxy network change to the remaining remotes.
let updated_task = proxy(
&index_scheduler,
None,
&req,
task.network.take().unwrap(), // set in register
merged_network,
Body::generated(proxied_network.clone(), |name, _remote, network| {
network.local = Setting::Set(name.to_string());
}),
&task,
)
.await?;
// unwrap: network was set by `proxy`
let task_network = updated_task.network.unwrap();
proxied_network.previous_remotes = Setting::NotSet;
if deleted_network.leader.is_some() {
// proxy network change to the deleted remotes
proxy(
&index_scheduler,
None,
&req,
task_network,
deleted_network,
Body::generated(proxied_network.clone(), |_name, _remote, network| {
network.local = Setting::Reset;
}),
&task,
)
.await?;
}
let task: SummarizedTaskView = task.into();
debug!("returns: {:?}", task);
Ok(HttpResponse::Accepted().json(task))
} else {
Ok(HttpResponse::Ok().json(merged_network))
}
}
async fn patch_network_with_origin(
index_scheduler: GuardedData<ActionPolicy<{ actions::NETWORK_UPDATE }>, Data<IndexScheduler>>,
merged_network: AwebJson<Network, DeserrJsonError>,
req: HttpRequest,
origin: Origin,
analytics: Data<Analytics>,
) -> Result<HttpResponse, ResponseError> {
let merged_network = merged_network.into_inner();
debug!(parameters = ?merged_network, ?origin, "Patch network");
let mut remotes = BTreeMap::new();
let mut old_network = index_scheduler.network();
for (name, remote) in merged_network.remotes.set().into_iter().flat_map(|x| x.into_iter()) {
let Some(remote) = remote else { continue };
let remote = remote.try_into_db_node(&name)?;
remotes.insert(name, remote);
}
let mut previous_remotes = BTreeMap::new();
for (name, remote) in
merged_network.previous_remotes.set().into_iter().flat_map(|x| x.into_iter())
{
let Some(remote) = remote else {
continue;
};
let remote = remote.try_into_db_node(&name)?;
previous_remotes.insert(name, remote);
}
old_network.remotes = previous_remotes;
let new_network = DbNetwork {
local: merged_network.local.set(),
remotes,
leader: merged_network.leader.set(),
version: origin.network_version,
};
index_scheduler.put_network(new_network.clone())?;
analytics.publish(
PatchNetworkAnalytics {
network_size: new_network.remotes.len(),
network_has_self: new_network.local.is_some(),
},
&req,
);
let network_topology_change = NetworkTopologyChange::new(old_network, new_network);
let task = KindWithContent::NetworkTopologyChange(network_topology_change);
let task = {
let index_scheduler = index_scheduler.clone();
tokio::task::spawn_blocking(move || {
index_scheduler.register_with_custom_metadata(
task,
None,
None,
false,
Some(TaskNetwork::Origin { origin }),
)
})
.await??
};
let task: SummarizedTaskView = task.into();
debug!("returns: {:?}", task);
Ok(HttpResponse::Accepted().json(task))
}
fn to_settings_remotes(
db_remotes: &BTreeMap<String, DbRemote>,
) -> BTreeMap<String, Option<Remote>> {
db_remotes
.iter()
.map(|(name, remote)| {
(
name.clone(),
Some(Remote {
url: Setting::Set(remote.url.clone()),
search_api_key: Setting::some_or_not_set(remote.search_api_key.clone()),
write_api_key: Setting::some_or_not_set(remote.write_api_key.clone()),
}),
)
})
.collect()
}
fn merge_networks(
old_network: DbNetwork,
new_network: Network,
) -> Result<DbNetwork, ResponseError> {
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,
let merged_leader = match new_network.leader {
Setting::Set(new_leader) => Some(new_leader),
Setting::Reset => None,
Setting::NotSet => old_network.leader,
};
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,
));
match (merged_leader.as_deref(), merged_self.as_deref()) {
// 1. Always allowed if there is no leader
(None, _) => (),
// 2. Allowed if the leader is self
(Some(leader), Some(this)) if leader == this => (),
// 3. Any other change is disallowed
(Some(leader), _) => {
return Err(MeilisearchHttpError::NotLeader { leader: leader.to_string() }.into())
}
}
let new_version = uuid::Uuid::now_v7();
let merged_remotes = match new_network.remotes {
Setting::Set(new_remotes) => {
let mut merged_remotes = BTreeMap::new();
@@ -301,18 +660,11 @@ async fn patch_network(
Setting::Reset => BTreeMap::new(),
Setting::NotSet => old_network.remotes,
};
analytics.publish(
PatchNetworkAnalytics {
network_size: merged_remotes.len(),
network_has_self: merged_self.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 merged_network = DbNetwork {
local: merged_self,
remotes: merged_remotes,
leader: merged_leader,
version: new_version,
};
Ok(merged_network)
}

View File

@@ -14,7 +14,7 @@ use meilisearch_types::index_uid::IndexUid;
use meilisearch_types::star_or::{OptionStarOr, OptionStarOrList};
use meilisearch_types::task_view::TaskView;
use meilisearch_types::tasks::{Kind, KindWithContent, Status};
use serde::Serialize;
use serde::{Deserialize, Serialize};
use time::format_description::well_known::Rfc3339;
use time::macros::format_description;
use time::{Date, Duration, OffsetDateTime, Time};
@@ -488,18 +488,18 @@ async fn delete_tasks(
Ok(HttpResponse::Ok().json(task))
}
#[derive(Debug, Serialize, ToSchema)]
#[derive(Debug, Serialize, Deserialize, ToSchema)]
pub struct AllTasks {
/// The list of tasks that matched the filter.
results: Vec<TaskView>,
pub results: Vec<TaskView>,
/// Total number of browsable results using offset/limit parameters for the given resource.
total: u64,
pub total: u64,
/// Limit given for the query. If limit is not provided as a query parameter, this parameter displays the default limit value.
limit: u32,
pub limit: u32,
/// The first task uid returned.
from: Option<u32>,
pub from: Option<u32>,
/// Represents the value to send in from to fetch the next slice of the results. The first item for the next slice starts at this exact number. When the returned value is null, it means that all the data have been browsed in the given order.
next: Option<u32>,
pub next: Option<u32>,
}
/// Get all tasks

View File

@@ -228,7 +228,7 @@ mod tests {
let err = deserr_query_params::<TaskDeletionOrCancelationQuery>(params).unwrap_err();
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`, `indexCompaction`.",
"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`, `indexCompaction`, `networkTopologyChange`.",
"code": "invalid_task_types",
"type": "invalid_request",
"link": "https://docs.meilisearch.com/errors#invalid_task_types"

View File

@@ -42,7 +42,7 @@ async fn batch_bad_types() {
snapshot!(code, @"400 Bad Request");
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`, `indexCompaction`.",
"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`, `indexCompaction`, `networkTopologyChange`.",
"code": "invalid_task_types",
"type": "invalid_request",
"link": "https://docs.meilisearch.com/errors#invalid_task_types"

View File

@@ -143,6 +143,8 @@ impl Display for Value {
".processingTimeMs" => "[duration]",
".details.embedders.*.url" => "[url]",
".details.dumpUid" => "[dump_uid]",
".network.network_version" => "[version]",
".network.origin.networkVersion" => "[version]",
})
)
}

View File

@@ -93,6 +93,20 @@ impl Service {
self.request(req).await
}
pub async fn patch_str(
&self,
url: impl AsRef<str>,
body: impl AsRef<str>,
headers: Vec<(&str, &str)>,
) -> (Value, StatusCode) {
let mut req =
test::TestRequest::patch().uri(url.as_ref()).set_payload(body.as_ref().to_string());
for header in headers {
req = req.insert_header(header);
}
self.request(req).await
}
pub async fn patch(&self, url: impl AsRef<str>, body: Value) -> (Value, StatusCode) {
self.patch_encoded(url, body, Encoder::Plain).await
}

View File

@@ -46,7 +46,7 @@ async fn errors_on_param() {
meili_snap::snapshot!(code, @"400 Bad Request");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
{
"message": "Unknown field `selfie`: expected one of `remotes`, `self`, `sharding`",
"message": "Unknown field `selfie`: expected one of `remotes`, `self`, `leader`, `previousRemotes`",
"code": "bad_request",
"type": "invalid_request",
"link": "https://docs.meilisearch.com/errors#bad_request"
@@ -186,7 +186,7 @@ async fn errors_on_param() {
.await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": null,
"remotes": {
@@ -196,7 +196,8 @@ async fn errors_on_param() {
"writeApiKey": null
}
},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = server
@@ -265,22 +266,24 @@ async fn auth() {
.await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "master",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = server.get_network().await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "master",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -289,11 +292,12 @@ async fn auth() {
let (response, code) = server.get_network().await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "master",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -307,11 +311,12 @@ async fn auth() {
.await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "api_key",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -390,18 +395,20 @@ async fn get_and_set_network() {
{
"self": null,
"remotes": {},
"sharding": false
"leader": null,
"version": "00000000-0000-0000-0000-000000000000"
}
"###);
// adding self
let (response, code) = server.set_network(json!({"self": "myself"})).await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "myself",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -419,7 +426,7 @@ async fn get_and_set_network() {
.await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "myself",
"remotes": {
@@ -434,7 +441,8 @@ async fn get_and_set_network() {
"writeApiKey": null
}
},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -448,7 +456,7 @@ async fn get_and_set_network() {
.await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "myself",
"remotes": {
@@ -463,7 +471,8 @@ async fn get_and_set_network() {
"writeApiKey": null
}
},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -478,7 +487,7 @@ async fn get_and_set_network() {
.await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "myself",
"remotes": {
@@ -498,7 +507,8 @@ async fn get_and_set_network() {
"writeApiKey": null
}
},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -510,7 +520,7 @@ async fn get_and_set_network() {
.await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "myself",
"remotes": {
@@ -525,7 +535,8 @@ async fn get_and_set_network() {
"writeApiKey": null
}
},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -533,7 +544,7 @@ async fn get_and_set_network() {
let (response, code) = server.set_network(json!({"self": Null})).await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": null,
"remotes": {
@@ -548,7 +559,8 @@ async fn get_and_set_network() {
"writeApiKey": null
}
},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -556,7 +568,7 @@ async fn get_and_set_network() {
let (response, code) = server.set_network(json!({"self": "thy"})).await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "thy",
"remotes": {
@@ -571,7 +583,8 @@ async fn get_and_set_network() {
"writeApiKey": null
}
},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -579,7 +592,7 @@ async fn get_and_set_network() {
let (response, code) = server.set_network(json!({})).await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "thy",
"remotes": {
@@ -594,7 +607,8 @@ async fn get_and_set_network() {
"writeApiKey": null
}
},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -602,7 +616,7 @@ async fn get_and_set_network() {
let (response, code) = server.set_network(json!({"remotes": {}})).await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "thy",
"remotes": {
@@ -617,7 +631,8 @@ async fn get_and_set_network() {
"writeApiKey": null
}
},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -625,7 +640,7 @@ async fn get_and_set_network() {
let (response, code) = server.get_network().await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "thy",
"remotes": {
@@ -640,7 +655,8 @@ async fn get_and_set_network() {
"writeApiKey": null
}
},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -652,11 +668,12 @@ async fn get_and_set_network() {
.await;
meili_snap::snapshot!(code, @"200 OK");
meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
meili_snap::snapshot!(meili_snap::json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "thy",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
}

View File

@@ -128,29 +128,32 @@ async fn remote_sharding() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms2.set_network(json!({"self": "ms2"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms2",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -189,8 +192,6 @@ 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");
let (_response, status_code) = ms1.set_network(network.clone()).await;
@@ -446,29 +447,32 @@ async fn remote_sharding_retrieve_vectors() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms2.set_network(json!({"self": "ms2"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms2",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -944,20 +948,22 @@ async fn error_unregistered_remote() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -1064,20 +1070,22 @@ async fn error_no_weighted_score() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -1200,20 +1208,22 @@ async fn error_bad_response() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -1340,20 +1350,22 @@ async fn error_bad_request() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -1473,20 +1485,22 @@ async fn error_bad_request_facets_by_index() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -1617,20 +1631,22 @@ async fn error_bad_request_facets_by_index_facet() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -1770,7 +1786,7 @@ async fn error_remote_does_not_answer() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
@@ -1779,7 +1795,7 @@ async fn error_remote_does_not_answer() {
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
@@ -1973,20 +1989,22 @@ async fn error_remote_404() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -2172,20 +2190,22 @@ async fn error_remote_sharding_auth() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -2335,20 +2355,22 @@ async fn remote_sharding_auth() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -2493,20 +2515,22 @@ async fn error_remote_500() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -2676,20 +2700,22 @@ async fn error_remote_500_once() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": false
"leader": null,
"version": "[version]"
}
"###);
@@ -2863,7 +2889,7 @@ async fn error_remote_timeout() {
let (response, code) = ms0.set_network(json!({"self": "ms0"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms0",
"remotes": {},
@@ -2872,7 +2898,7 @@ async fn error_remote_timeout() {
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1"})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
snapshot!(json_string!(response, {".version" => "[version]"}), @r###"
{
"self": "ms1",
"remotes": {},
@@ -3082,8 +3108,8 @@ impl LocalMeili {
let (value, code) = rt.block_on(async {
match req.method.as_str() {
"POST" => server.service.post_str(&req.url, body, headers.clone()).await,
"PUT" => server.service.put_str(&req.url, body, headers).await,
"PATCH" => server.service.patch(&req.url, req.body_json().unwrap()).await,
"PUT" => server.service.put_str(&req.url, body, headers.clone()).await,
"PATCH" => server.service.patch_str(&req.url, body, headers).await,
"GET" => server.service.get(&req.url).await,
"DELETE" => server.service.delete(&req.url).await,
_ => unimplemented!(),
@@ -3160,36 +3186,6 @@ async fn remote_auto_sharding() {
snapshot!(code, @"200 OK");
snapshot!(json_string!(response["network"]), @"true");
// set self & sharding
let (response, code) = ms0.set_network(json!({"self": "ms0", "sharding": true})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": true
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1", "sharding": true})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": true
}
"###);
let (response, code) = ms2.set_network(json!({"self": "ms2", "sharding": true})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
{
"self": "ms2",
"remotes": {},
"sharding": true
}
"###);
// wrap servers
let ms0 = Arc::new(ms0);
let ms1 = Arc::new(ms1);
@@ -3200,7 +3196,10 @@ async fn remote_auto_sharding() {
let rms2 = LocalMeili::new(ms2.clone()).await;
// set network
let network = json!({"remotes": {
let network = json!({
"self": "ms0",
"leader": "ms0",
"remotes": {
"ms0": {
"url": rms0.url()
},
@@ -3214,12 +3213,99 @@ async fn remote_auto_sharding() {
println!("{}", serde_json::to_string_pretty(&network).unwrap());
let (_response, status_code) = ms0.set_network(network.clone()).await;
let (task, status_code) = ms0.set_network(network.clone()).await;
snapshot!(status_code, @"202 Accepted");
let t0 = task.uid();
let (t, _) = ms0.get_task(t0).await;
let t1 = t["network"]["remote_tasks"]["ms1"]["taskUid"].as_u64().unwrap();
let t2 = t["network"]["remote_tasks"]["ms2"]["taskUid"].as_u64().unwrap();
ms0.wait_task(t0).await.succeeded();
ms1.wait_task(t1).await.succeeded();
ms2.wait_task(t2).await.succeeded();
let (response, status_code) = ms0.get_network().await;
snapshot!(status_code, @"200 OK");
let (_response, status_code) = ms1.set_network(network.clone()).await;
snapshot!(json_string!(response, {".version" => "[version]", ".remotes.*.url" => "[url]"}), @r###"
{
"self": "ms0",
"remotes": {
"ms0": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms1": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms2": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
}
},
"leader": "ms0",
"version": "[version]"
}
"###);
let (response, status_code) = ms1.get_network().await;
snapshot!(status_code, @"200 OK");
let (_response, status_code) = ms2.set_network(network.clone()).await;
snapshot!(json_string!(response, {".version" => "[version]", ".remotes.*.url" => "[url]"}), @r###"
{
"self": "ms1",
"remotes": {
"ms0": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms1": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms2": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
}
},
"leader": "ms0",
"version": "[version]"
}
"###);
let (response, status_code) = ms2.get_network().await;
snapshot!(status_code, @"200 OK");
snapshot!(json_string!(response, {".version" => "[version]", ".remotes.*.url" => "[url]"}), @r###"
{
"self": "ms2",
"remotes": {
"ms0": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms1": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms2": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
}
},
"leader": "ms0",
"version": "[version]"
}
"###);
// add documents
let documents = SCORE_DOCUMENTS.clone();
@@ -3480,36 +3566,6 @@ async fn remote_auto_sharding_with_custom_metadata() {
snapshot!(code, @"200 OK");
snapshot!(json_string!(response["network"]), @"true");
// set self & sharding
let (response, code) = ms0.set_network(json!({"self": "ms0", "sharding": true})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
{
"self": "ms0",
"remotes": {},
"sharding": true
}
"###);
let (response, code) = ms1.set_network(json!({"self": "ms1", "sharding": true})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
{
"self": "ms1",
"remotes": {},
"sharding": true
}
"###);
let (response, code) = ms2.set_network(json!({"self": "ms2", "sharding": true})).await;
snapshot!(code, @"200 OK");
snapshot!(json_string!(response), @r###"
{
"self": "ms2",
"remotes": {},
"sharding": true
}
"###);
// wrap servers
let ms0 = Arc::new(ms0);
let ms1 = Arc::new(ms1);
@@ -3520,7 +3576,10 @@ async fn remote_auto_sharding_with_custom_metadata() {
let rms2 = LocalMeili::new(ms2.clone()).await;
// set network
let network = json!({"remotes": {
let network = json!({
"self": "ms0",
"leader": "ms0",
"remotes": {
"ms0": {
"url": rms0.url()
},
@@ -3534,12 +3593,99 @@ async fn remote_auto_sharding_with_custom_metadata() {
println!("{}", serde_json::to_string_pretty(&network).unwrap());
let (_response, status_code) = ms0.set_network(network.clone()).await;
let (task, status_code) = ms0.set_network(network.clone()).await;
snapshot!(status_code, @"202 Accepted");
let t0 = task.uid();
let (t, _) = ms0.get_task(t0).await;
let t1 = t["network"]["remote_tasks"]["ms1"]["taskUid"].as_u64().unwrap();
let t2 = t["network"]["remote_tasks"]["ms2"]["taskUid"].as_u64().unwrap();
ms0.wait_task(t0).await.succeeded();
ms1.wait_task(t1).await.succeeded();
ms2.wait_task(t2).await.succeeded();
let (response, status_code) = ms0.get_network().await;
snapshot!(status_code, @"200 OK");
let (_response, status_code) = ms1.set_network(network.clone()).await;
snapshot!(json_string!(response, {".version" => "[version]", ".remotes.*.url" => "[url]"}), @r###"
{
"self": "ms0",
"remotes": {
"ms0": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms1": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms2": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
}
},
"leader": "ms0",
"version": "[version]"
}
"###);
let (response, status_code) = ms1.get_network().await;
snapshot!(status_code, @"200 OK");
let (_response, status_code) = ms2.set_network(network.clone()).await;
snapshot!(json_string!(response, {".version" => "[version]", ".remotes.*.url" => "[url]"}), @r###"
{
"self": "ms1",
"remotes": {
"ms0": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms1": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms2": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
}
},
"leader": "ms0",
"version": "[version]"
}
"###);
let (response, status_code) = ms2.get_network().await;
snapshot!(status_code, @"200 OK");
snapshot!(json_string!(response, {".version" => "[version]", ".remotes.*.url" => "[url]"}), @r###"
{
"self": "ms2",
"remotes": {
"ms0": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms1": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
},
"ms2": {
"url": "[url]",
"searchApiKey": null,
"writeApiKey": null
}
},
"leader": "ms0",
"version": "[version]"
}
"###);
// add documents
let documents = SCORE_DOCUMENTS.clone();
@@ -3562,6 +3708,7 @@ async fn remote_auto_sharding_with_custom_metadata() {
let t2 = t["network"]["remote_tasks"]["ms2"]["taskUid"].as_u64().unwrap();
let t = ms0.wait_task(t0).await.succeeded();
snapshot!(t, @r###"
{
"uid": "[uid]",
@@ -3582,14 +3729,15 @@ async fn remote_auto_sharding_with_custom_metadata() {
"network": {
"remote_tasks": {
"ms1": {
"taskUid": 0,
"taskUid": 1,
"error": null
},
"ms2": {
"taskUid": 0,
"taskUid": 1,
"error": null
}
}
},
"network_version": "[version]"
},
"customMetadata": "remote_auto_sharding_with_custom_metadata"
}
@@ -3616,7 +3764,8 @@ async fn remote_auto_sharding_with_custom_metadata() {
"network": {
"origin": {
"remoteName": "ms0",
"taskUid": 0
"taskUid": 1,
"networkVersion": "[version]"
}
},
"customMetadata": "remote_auto_sharding_with_custom_metadata"
@@ -3644,7 +3793,8 @@ async fn remote_auto_sharding_with_custom_metadata() {
"network": {
"origin": {
"remoteName": "ms0",
"taskUid": 0
"taskUid": 1,
"networkVersion": "[version]"
}
},
"customMetadata": "remote_auto_sharding_with_custom_metadata"

View File

@@ -95,36 +95,36 @@ 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`, `indexCompaction`.",
"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`, `indexCompaction`, `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");
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`, `indexCompaction`.",
"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`, `indexCompaction`, `networkTopologyChange`.",
"code": "invalid_task_types",
"type": "invalid_request",
"link": "https://docs.meilisearch.com/errors#invalid_task_types"
}
"#);
"###);
let (response, code) = server.delete_tasks("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`, `indexCompaction`.",
"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`, `indexCompaction`, `networkTopologyChange`.",
"code": "invalid_task_types",
"type": "invalid_request",
"link": "https://docs.meilisearch.com/errors#invalid_task_types"
}
"#);
"###);
}
#[actix_rt::test]

View File

@@ -5,18 +5,38 @@
use std::hash::{BuildHasher as _, BuildHasherDefault};
pub struct Shards {
pub own: Vec<String>,
pub others: Vec<String>,
#[derive(Debug, Clone)]
pub struct Shards(pub Vec<Shard>);
#[derive(Debug, Clone)]
pub struct Shard {
pub is_own: bool,
pub name: String,
}
impl Shards {
pub fn from_remotes_local<'a>(
remotes: impl IntoIterator<Item = &'a str>,
local: Option<&str>,
) -> Self {
Shards(
remotes
.into_iter()
.map(|name| Shard { is_own: Some(name) == local, name: name.to_owned() })
.collect(),
)
}
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
}
}