Merge #3789

3789: Improve the metrics r=dureuill a=irevoire

# Pull Request

## Related issue
Implements https://github.com/meilisearch/meilisearch/issues/3790
Associated specification: https://github.com/meilisearch/specifications/pull/242

## Be cautious; it's DB-breaking 😱 

While reviewing and after merging this PR, be cautious; if you already have a `data.ms` and run meilisearch with this code on it, it won't work because we need to cache a new information on the index stats (that are backed up on disk). You'll get internal errors.

### About the breaking-change label

We only break the API of the metrics route, which does not pose any problem since it's experimental.

## What does this PR do?
- Create a method to get the « facet distribution » of the task queue.
- Prefix all the metrics by `meilisearch_`
- Add the real database size used by meilisearch
- Add metrics on the task queue
- Update the grafana dashboard to these new changes
- Move the dashboard to the `assets` directory
- Provide a new prometheus file to scrape meilisearch easily

Co-authored-by: Tamo <tamo@meilisearch.com>

assets/prometheus-basic-scraper.yml

@@ -0,0 +1,19 @@
+global:
+  scrape_interval:     15s # By default, scrape targets every 15 seconds.
+
+  # Attach these labels to any time series or alerts when communicating with
+  # external systems (federation, remote storage, Alertmanager).
+  external_labels:
+    monitor: 'codelab-monitor'
+
+# A scrape configuration containing exactly one endpoint to scrape:
+# Here it's Prometheus itself.
+scrape_configs:
+  # The job name is added as a label `job=<job_name>` to any timeseries scraped from this config.
+  - job_name: 'meilisearch'
+
+    # Override the global default and scrape targets from this job every 5 seconds.
+    scrape_interval: 5s
+
+    static_configs:
+      - targets: ['localhost:7700']

index-scheduler/src/index_mapper/mod.rs

@@ -88,8 +88,17 @@ pub enum IndexStatus {
 pub struct IndexStats {
     /// Number of documents in the index.
     pub number_of_documents: u64,
-    /// Size of the index' DB, in bytes.
+    /// Size taken up by the index' DB, in bytes.
+    ///
+    /// This includes the size taken by both the used and free pages of the DB, and as the free pages
+    /// are not returned to the disk after a deletion, this number is typically larger than
+    /// `used_database_size` that only includes the size of the used pages.
     pub database_size: u64,
+    /// Size taken by the used pages of the index' DB, in bytes.
+    ///
+    /// As the DB backend does not return to the disk the pages that are not currently used by the DB,
+    /// this value is typically smaller than `database_size`.
+    pub used_database_size: u64,
     /// Association of every field name with the number of times it occurs in the documents.
     pub field_distribution: FieldDistribution,
     /// Creation date of the index.
@@ -105,10 +114,10 @@ impl IndexStats {
     ///
     /// - rtxn: a RO transaction for the index, obtained from `Index::read_txn()`.
     pub fn new(index: &Index, rtxn: &RoTxn) -> Result<Self> {
-        let database_size = index.on_disk_size()?;
         Ok(IndexStats {
             number_of_documents: index.number_of_documents(rtxn)?,
-            database_size,
+            database_size: index.on_disk_size()?,
+            used_database_size: index.used_size()?,
             field_distribution: index.field_distribution(rtxn)?,
             created_at: index.created_at(rtxn)?,
             updated_at: index.updated_at(rtxn)?,

index-scheduler/src/lib.rs

@@ -31,7 +31,7 @@ mod uuid_codec;
 pub type Result<T> = std::result::Result<T, Error>;
 pub type TaskId = u32;
 
-use std::collections::HashMap;
+use std::collections::{BTreeMap, HashMap};
 use std::ops::{Bound, RangeBounds};
 use std::path::{Path, PathBuf};
 use std::sync::atomic::AtomicBool;
@@ -554,10 +554,16 @@ impl IndexScheduler {
         &self.index_mapper.indexer_config
     }
 
+    /// Return the real database size (i.e.: The size **with** the free pages)
     pub fn size(&self) -> Result<u64> {
         Ok(self.env.real_disk_size()?)
     }
 
+    /// Return the used database size (i.e.: The size **without** the free pages)
+    pub fn used_size(&self) -> Result<u64> {
+        Ok(self.env.non_free_pages_size()?)
+    }
+
     /// Return the index corresponding to the name.
     ///
     /// * If the index wasn't opened before, the index will be opened.
@@ -737,6 +743,38 @@ impl IndexScheduler {
         Ok(tasks)
     }
 
+    /// The returned structure contains:
+    /// 1. The name of the property being observed can be `statuses`, `types`, or `indexes`.
+    /// 2. The name of the specific data related to the property can be `enqueued` for the `statuses`, `settingsUpdate` for the `types`, or the name of the index for the `indexes`, for example.
+    /// 3. The number of times the properties appeared.
+    pub fn get_stats(&self) -> Result<BTreeMap<String, BTreeMap<String, u64>>> {
+        let rtxn = self.read_txn()?;
+
+        let mut res = BTreeMap::new();
+
+        res.insert(
+            "statuses".to_string(),
+            enum_iterator::all::<Status>()
+                .map(|s| Ok((s.to_string(), self.get_status(&rtxn, s)?.len())))
+                .collect::<Result<BTreeMap<String, u64>>>()?,
+        );
+        res.insert(
+            "types".to_string(),
+            enum_iterator::all::<Kind>()
+                .map(|s| Ok((s.to_string(), self.get_kind(&rtxn, s)?.len())))
+                .collect::<Result<BTreeMap<String, u64>>>()?,
+        );
+        res.insert(
+            "indexes".to_string(),
+            self.index_tasks
+                .iter(&rtxn)?
+                .map(|res| Ok(res.map(|(name, bitmap)| (name.to_string(), bitmap.len()))?))
+                .collect::<Result<BTreeMap<String, u64>>>()?,
+        );
+
+        Ok(res)
+    }
+
     /// Return true iff there is at least one task associated with this index
     /// that is processing.
     pub fn is_index_processing(&self, index: &str) -> Result<bool> {

meilisearch-auth/src/lib.rs

@@ -45,6 +45,11 @@ impl AuthController {
         self.store.size()
     }
 
+    /// Return the used size of the `AuthController` database in bytes.
+    pub fn used_size(&self) -> Result<u64> {
+        self.store.used_size()
+    }
+
     pub fn create_key(&self, create_key: CreateApiKey) -> Result<Key> {
         match self.store.get_api_key(create_key.uid)? {
             Some(_) => Err(AuthControllerError::ApiKeyAlreadyExists(create_key.uid.to_string())),

meilisearch-auth/src/store.rs

@@ -73,6 +73,11 @@ impl HeedAuthStore {
         Ok(self.env.real_disk_size()?)
     }
 
+    /// Return the number of bytes actually used in the database
+    pub fn used_size(&self) -> Result<u64> {
+        Ok(self.env.non_free_pages_size()?)
+    }
+
     pub fn set_drop_on_close(&mut self, v: bool) {
         self.should_close_on_drop = v;
     }

meilisearch/src/metrics.rs

@@ -4,20 +4,32 @@ use prometheus::{
     register_int_gauge_vec, HistogramVec, IntCounterVec, IntGauge, IntGaugeVec,
 };
 
-const HTTP_RESPONSE_TIME_CUSTOM_BUCKETS: &[f64; 14] = &[
-    0.0005, 0.0008, 0.00085, 0.0009, 0.00095, 0.001, 0.00105, 0.0011, 0.00115, 0.0012, 0.0015,
-    0.002, 0.003, 1.0,
-];
+/// Create evenly distributed buckets
+fn create_buckets() -> [f64; 29] {
+    (0..10)
+        .chain((10..100).step_by(10))
+        .chain((100..=1000).step_by(100))
+        .map(|i| i as f64 / 1000.)
+        .collect::<Vec<_>>()
+        .try_into()
+        .unwrap()
+}
 
 lazy_static! {
-    pub static ref HTTP_REQUESTS_TOTAL: IntCounterVec = register_int_counter_vec!(
-        opts!("http_requests_total", "HTTP requests total"),
+    pub static ref HTTP_RESPONSE_TIME_CUSTOM_BUCKETS: [f64; 29] = create_buckets();
+    pub static ref MEILISEARCH_HTTP_REQUESTS_TOTAL: IntCounterVec = register_int_counter_vec!(
+        opts!("meilisearch_http_requests_total", "Meilisearch HTTP requests total"),
         &["method", "path"]
     )
     .expect("Can't create a metric");
     pub static ref MEILISEARCH_DB_SIZE_BYTES: IntGauge =
-        register_int_gauge!(opts!("meilisearch_db_size_bytes", "Meilisearch Db Size In Bytes"))
+        register_int_gauge!(opts!("meilisearch_db_size_bytes", "Meilisearch DB Size In Bytes"))
             .expect("Can't create a metric");
+    pub static ref MEILISEARCH_USED_DB_SIZE_BYTES: IntGauge = register_int_gauge!(opts!(
+        "meilisearch_used_db_size_bytes",
+        "Meilisearch Used DB Size In Bytes"
+    ))
+    .expect("Can't create a metric");
     pub static ref MEILISEARCH_INDEX_COUNT: IntGauge =
         register_int_gauge!(opts!("meilisearch_index_count", "Meilisearch Index Count"))
             .expect("Can't create a metric");
@@ -26,11 +38,16 @@ lazy_static! {
         &["index"]
     )
     .expect("Can't create a metric");
-    pub static ref HTTP_RESPONSE_TIME_SECONDS: HistogramVec = register_histogram_vec!(
+    pub static ref MEILISEARCH_HTTP_RESPONSE_TIME_SECONDS: HistogramVec = register_histogram_vec!(
         "http_response_time_seconds",
         "HTTP response times",
         &["method", "path"],
         HTTP_RESPONSE_TIME_CUSTOM_BUCKETS.to_vec()
     )
     .expect("Can't create a metric");
+    pub static ref MEILISEARCH_NB_TASKS: IntGaugeVec = register_int_gauge_vec!(
+        opts!("meilisearch_nb_tasks", "Meilisearch Number of tasks"),
+        &["kind", "value"]
+    )
+    .expect("Can't create a metric");
 }

meilisearch/src/middleware.rs

@@ -52,11 +52,11 @@ where
         if is_registered_resource {
             let request_method = req.method().to_string();
             histogram_timer = Some(
-                crate::metrics::HTTP_RESPONSE_TIME_SECONDS
+                crate::metrics::MEILISEARCH_HTTP_RESPONSE_TIME_SECONDS
                     .with_label_values(&[&request_method, request_path])
                     .start_timer(),
             );
-            crate::metrics::HTTP_REQUESTS_TOTAL
+            crate::metrics::MEILISEARCH_HTTP_REQUESTS_TOTAL
                 .with_label_values(&[&request_method, request_path])
                 .inc();
         }

meilisearch/src/routes/metrics.rs

@@ -17,7 +17,7 @@ pub fn configure(config: &mut web::ServiceConfig) {
 
 pub async fn get_metrics(
     index_scheduler: GuardedData<ActionPolicy<{ actions::METRICS_GET }>, Data<IndexScheduler>>,
-    auth_controller: GuardedData<ActionPolicy<{ actions::METRICS_GET }>, Data<AuthController>>,
+    auth_controller: Data<AuthController>,
 ) -> Result<HttpResponse, ResponseError> {
     let auth_filters = index_scheduler.filters();
     if !auth_filters.all_indexes_authorized() {
@@ -28,10 +28,10 @@ pub async fn get_metrics(
         return Err(error);
     }
 
-    let response =
-        create_all_stats((*index_scheduler).clone(), (*auth_controller).clone(), auth_filters)?;
+    let response = create_all_stats((*index_scheduler).clone(), auth_controller, auth_filters)?;
 
     crate::metrics::MEILISEARCH_DB_SIZE_BYTES.set(response.database_size as i64);
+    crate::metrics::MEILISEARCH_USED_DB_SIZE_BYTES.set(response.used_database_size as i64);
     crate::metrics::MEILISEARCH_INDEX_COUNT.set(response.indexes.len() as i64);
 
     for (index, value) in response.indexes.iter() {
@@ -40,6 +40,14 @@ pub async fn get_metrics(
             .set(value.number_of_documents as i64);
     }
 
+    for (kind, value) in index_scheduler.get_stats()? {
+        for (value, count) in value {
+            crate::metrics::MEILISEARCH_NB_TASKS
+                .with_label_values(&[&kind, &value])
+                .set(count as i64);
+        }
+    }
+
     let encoder = TextEncoder::new();
     let mut buffer = vec![];
     encoder.encode(&prometheus::gather(), &mut buffer).expect("Failed to encode metrics");

meilisearch/src/routes/mod.rs

@@ -231,6 +231,8 @@ pub async fn running() -> HttpResponse {
 #[serde(rename_all = "camelCase")]
 pub struct Stats {
     pub database_size: u64,
+    #[serde(skip)]
+    pub used_database_size: u64,
     #[serde(serialize_with = "time::serde::rfc3339::option::serialize")]
     pub last_update: Option<OffsetDateTime>,
     pub indexes: BTreeMap<String, indexes::IndexStats>,
@@ -259,6 +261,7 @@ pub fn create_all_stats(
     let mut last_task: Option<OffsetDateTime> = None;
     let mut indexes = BTreeMap::new();
     let mut database_size = 0;
+    let mut used_database_size = 0;
 
     for index_uid in index_scheduler.index_names()? {
         // Accumulate the size of all indexes, even unauthorized ones, so
@@ -266,6 +269,7 @@ pub fn create_all_stats(
         // See <https://github.com/meilisearch/meilisearch/pull/3541#discussion_r1126747643> for context.
         let stats = index_scheduler.index_stats(&index_uid)?;
         database_size += stats.inner_stats.database_size;
+        used_database_size += stats.inner_stats.used_database_size;
 
         if !filters.is_index_authorized(&index_uid) {
             continue;
@@ -278,10 +282,14 @@ pub fn create_all_stats(
     }
 
     database_size += index_scheduler.size()?;
+    used_database_size += index_scheduler.used_size()?;
     database_size += auth_controller.size()?;
-    database_size += index_scheduler.compute_update_file_size()?;
+    used_database_size += auth_controller.used_size()?;
+    let update_file_size = index_scheduler.compute_update_file_size()?;
+    database_size += update_file_size;
+    used_database_size += update_file_size;
 
-    let stats = Stats { database_size, last_update: last_task, indexes };
+    let stats = Stats { database_size, used_database_size, last_update: last_task, indexes };
     Ok(stats)
 }