Merge pull request #5956 from meilisearch/progress-trace-in-metrics

Expose batch progress traces on the metrics route

Cargo.lock

@@ -2931,6 +2931,12 @@ version = "1.0.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "df3b46402a9d5adb4c86a0cf463f42e19994e3ee891101b1841f30a545cb49a9"
 
+[[package]]
+name = "humantime"
+version = "2.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "135b12329e5e3ce057a9f972339ea52bc954fe1e9358ef27f95e89716fbc5424"
+
 [[package]]
 name = "hyper"
 version = "1.8.1"
@@ -4008,6 +4014,7 @@ dependencies = [
  "futures",
  "futures-util",
  "hex",
+ "humantime",
  "index-scheduler",
  "indexmap",
  "insta",

crates/meilisearch/Cargo.toml

@@ -117,7 +117,7 @@ secrecy = "0.10.3"
 actix-web-lab = { version = "0.24.3", default-features = false }
 urlencoding = "2.1.3"
 backoff = { version = "0.4.0", features = ["tokio"] }
-
+humantime = { version = "2.3.0", default-features = false }
 
 [dev-dependencies]
 actix-rt = "2.11.0"

crates/meilisearch/src/metrics.rs

@@ -1,7 +1,8 @@
 use lazy_static::lazy_static;
 use prometheus::{
-    opts, register_gauge, register_histogram_vec, register_int_counter_vec, register_int_gauge,
+    opts, register_gauge, register_gauge_vec, register_histogram_vec, register_int_counter_vec,
-    register_int_gauge_vec, Gauge, HistogramVec, IntCounterVec, IntGauge, IntGaugeVec,
+    register_int_gauge, register_int_gauge_vec, Gauge, GaugeVec, HistogramVec, IntCounterVec,
+    IntGauge, IntGaugeVec,
 };
 
 lazy_static! {
@@ -73,6 +74,20 @@ lazy_static! {
         &["kind", "value"]
     )
     .expect("Can't create a metric");
+    pub static ref MEILISEARCH_BATCH_RUNNING_PROGRESS_TRACE: GaugeVec = register_gauge_vec!(
+        opts!("meilisearch_batch_running_progress_trace", "The currently running progress trace"),
+        &["batch_uid", "step_name"]
+    )
+    .expect("Can't create a metric");
+    pub static ref MEILISEARCH_LAST_FINISHED_BATCHES_PROGRESS_TRACE_MS: IntGaugeVec =
+        register_int_gauge_vec!(
+            opts!(
+                "meilisearch_last_finished_batches_progress_trace_ms",
+                "The last few batches progress trace in milliseconds"
+            ),
+            &["batch_uid", "step_name"]
+        )
+        .expect("Can't create a metric");
     pub static ref MEILISEARCH_LAST_UPDATE: IntGauge =
         register_int_gauge!(opts!("meilisearch_last_update", "Meilisearch Last Update"))
             .expect("Can't create a metric");

crates/meilisearch/src/routes/metrics.rs

@@ -4,6 +4,7 @@ use index_scheduler::{IndexScheduler, Query};
 use meilisearch_auth::AuthController;
 use meilisearch_types::error::ResponseError;
 use meilisearch_types::keys::actions;
+use meilisearch_types::milli::progress::ProgressStepView;
 use meilisearch_types::tasks::Status;
 use prometheus::{Encoder, TextEncoder};
 use time::OffsetDateTime;
@@ -38,6 +39,12 @@ pub fn configure(config: &mut web::ServiceConfig) {
 # HELP meilisearch_db_size_bytes Meilisearch DB Size In Bytes
 # TYPE meilisearch_db_size_bytes gauge
 meilisearch_db_size_bytes 1130496
+# HELP meilisearch_batch_running_progress_trace The currently running progress trace
+# TYPE meilisearch_batch_running_progress_trace gauge
+meilisearch_batch_running_progress_trace{batch_uid="0",step_name="document"} 0.710618582519409
+meilisearch_batch_running_progress_trace{batch_uid="0",step_name="extracting word proximity"} 0.2222222222222222
+meilisearch_batch_running_progress_trace{batch_uid="0",step_name="indexing"} 0.6666666666666666
+meilisearch_batch_running_progress_trace{batch_uid="0",step_name="processing tasks"} 0
 # HELP meilisearch_http_requests_total Meilisearch HTTP requests total
 # TYPE meilisearch_http_requests_total counter
 meilisearch_http_requests_total{method="GET",path="/metrics",status="400"} 1
@@ -61,6 +68,13 @@ meilisearch_http_response_time_seconds_bucket{method="GET",path="/metrics",le="1
 meilisearch_http_response_time_seconds_bucket{method="GET",path="/metrics",le="+Inf"} 0
 meilisearch_http_response_time_seconds_sum{method="GET",path="/metrics"} 0
 meilisearch_http_response_time_seconds_count{method="GET",path="/metrics"} 0
+# HELP meilisearch_last_finished_batches_progress_trace_ms The last few batches progress trace in milliseconds
+# TYPE meilisearch_last_finished_batches_progress_trace_ms gauge
+meilisearch_last_finished_batches_progress_trace_ms{batch_uid="0",step_name="processing tasks"} 19360
+meilisearch_last_finished_batches_progress_trace_ms{batch_uid="0",step_name="processing tasks > computing document changes"} 368
+meilisearch_last_finished_batches_progress_trace_ms{batch_uid="0",step_name="processing tasks > computing document changes > preparing payloads"} 367
+meilisearch_last_finished_batches_progress_trace_ms{batch_uid="0",step_name="processing tasks > computing document changes > preparing payloads > payload"} 367
+meilisearch_last_finished_batches_progress_trace_ms{batch_uid="0",step_name="processing tasks > indexing"} 18970
 # HELP meilisearch_index_count Meilisearch Index Count
 # TYPE meilisearch_index_count gauge
 meilisearch_index_count 1
@@ -148,6 +162,46 @@ pub async fn get_metrics(
         }
     }
 
+    // Fetch and expose the current progressing step
+    crate::metrics::MEILISEARCH_BATCH_RUNNING_PROGRESS_TRACE.reset();
+    let (batches, _total) = index_scheduler.get_batches_from_authorized_indexes(
+        &Query { statuses: Some(vec![Status::Processing]), ..Query::default() },
+        auth_filters,
+    )?;
+    if let Some(batch) = batches.into_iter().next() {
+        let batch_uid = batch.uid.to_string();
+        if let Some(progress) = batch.progress {
+            for ProgressStepView { current_step, finished, total } in progress.steps {
+                crate::metrics::MEILISEARCH_BATCH_RUNNING_PROGRESS_TRACE
+                    .with_label_values(&[batch_uid.as_str(), current_step.as_ref()])
+                    // We return the completion ratio of the current step
+                    .set(finished as f64 / total as f64);
+            }
+        }
+    }
+
+    crate::metrics::MEILISEARCH_LAST_FINISHED_BATCHES_PROGRESS_TRACE_MS.reset();
+    let (batches, _total) = index_scheduler.get_batches_from_authorized_indexes(
+        // Fetch the finished batches...
+        &Query { statuses: Some(vec![Status::Succeeded, Status::Failed]), ..Query::default() },
+        auth_filters,
+    )?;
+    // ...and get the last batch only.
+    if let Some(batch) = batches.into_iter().next() {
+        let batch_uid = batch.uid.to_string();
+        for (step_name, duration_str) in batch.stats.progress_trace {
+            let Some(duration_str) = duration_str.as_str() else { continue };
+            match humantime::parse_duration(duration_str) {
+                Ok(duration) => {
+                    crate::metrics::MEILISEARCH_LAST_FINISHED_BATCHES_PROGRESS_TRACE_MS
+                        .with_label_values(&[&batch_uid, &step_name])
+                        .set(duration.as_millis() as i64);
+                }
+                Err(e) => tracing::error!("Failed to parse duration: {e}"),
+            }
+        }
+    }
+
     if let Some(last_update) = response.last_update {
         crate::metrics::MEILISEARCH_LAST_UPDATE.set(last_update.unix_timestamp());
     }