feat( cluster): Add pg_stat_statements instrumentation (#805)

Signed-off-by: Philippe Noël <philippemnoel@gmail.com>
Signed-off-by: Itay Grudev <itay@verito.digital>
Co-authored-by: Itay Grudev <itay@verito.digital>

Author: philippemnoel

charts/cluster/README.md

@@ -174,6 +174,7 @@ Kubernetes: `>=1.29.0-0`
 | cluster.monitoring.disableDefaultQueries | bool | `false` | Whether the default queries should be injected. Set it to true if you don't want to inject default queries into the cluster. |
 | cluster.monitoring.enabled | bool | `false` | Whether to enable monitoring |
 | cluster.monitoring.instrumentation.logicalReplication | bool | `true` | Enable logical replication metrics |
+| cluster.monitoring.instrumentation.pgStatStatements | bool | `false` | Enable planning and execution statistics for all SQL statements. Increases shared memory usage. |
 | cluster.monitoring.podMonitor.enabled | bool | `true` | Whether to enable the PodMonitor |
 | cluster.monitoring.podMonitor.labels | object | `{}` | Additional labels to set on the generated PodMonitor resource. Add labels your monitoring stack requires (for example `team-name`). |
 | cluster.monitoring.podMonitor.metricRelabelings | list | `[]` | The list of metric relabelings for the PodMonitor. Applied to samples before ingestion. |

charts/cluster/templates/_bootstrap.tpl

@@ -10,7 +10,7 @@ bootstrap:
     {{- if .Values.cluster.initdb.owner }}
     owner: {{ tpl .Values.cluster.initdb.owner . }}
     {{- end }}
-    {{- if or (eq .Values.type "postgis") (eq .Values.type "timescaledb") (not (empty .Values.cluster.initdb.postInitApplicationSQL)) }}
+    {{- if or (eq .Values.type "postgis") (eq .Values.type "timescaledb") (not (empty .Values.cluster.initdb.postInitApplicationSQL)) .Values.cluster.monitoring.instrumentation.pgStatStatements }}
     postInitApplicationSQL:
       {{- if eq .Values.type "postgis" }}
       - CREATE EXTENSION IF NOT EXISTS postgis;
@@ -20,6 +20,9 @@ bootstrap:
       {{- else if eq .Values.type "timescaledb" }}
       - CREATE EXTENSION IF NOT EXISTS timescaledb;
       {{- end }}
+      {{- if .Values.cluster.monitoring.instrumentation.pgStatStatements }}
+      - CREATE EXTENSION IF NOT EXISTS pg_stat_statements;
+      {{- end }}
       {{- with .Values.cluster.initdb }}
           {{- range .postInitApplicationSQL }}
             {{- printf "- %s" . | nindent 6 }}

charts/cluster/templates/cluster.yaml

@@ -68,11 +68,14 @@ spec:
   {{ end }}
   enablePDB: {{ .Values.cluster.enablePDB }}
   postgresql:
-    {{- if or (eq .Values.type "timescaledb") (not (empty .Values.cluster.postgresql.shared_preload_libraries)) }}
+    {{- if or (eq .Values.type "timescaledb") (not (empty .Values.cluster.postgresql.shared_preload_libraries)) .Values.cluster.monitoring.instrumentation.pgStatStatements }}
     shared_preload_libraries:
       {{- if eq .Values.type "timescaledb" }}
       - timescaledb
       {{- end }}
+      {{- if .Values.cluster.monitoring.instrumentation.pgStatStatements }}
+      - pg_stat_statements
+      {{- end }}
       {{- with .Values.cluster.postgresql.shared_preload_libraries }}
       {{- toYaml . | nindent 6 }}
       {{- end }}

charts/cluster/templates/monitoring-pg-stat-statements.yaml

@@ -0,0 +1,52 @@
+{{- if .Values.cluster.monitoring.instrumentation.pgStatStatements }}
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: {{ include "cluster.fullname" . }}-monitoring-pg-stat-statements
+  namespace: {{ include "cluster.namespace" . }}
+  labels:
+    cnpg.io/reload: ""
+    {{- include "cluster.labels" . | nindent 4 }}
+data:
+  custom-queries: |
+    pg_stat_statements:
+      query: |
+        SELECT d.datname
+             , sum(s.calls) AS calls
+             , sum(s.total_exec_time) / 1000 AS total_exec_time_seconds
+             , sum(s.rows) AS rows
+             , sum(s.shared_blks_hit) AS shared_blks_hit
+             , sum(s.shared_blks_read) AS shared_blks_read
+             , sum(s.blk_read_time) / 1000 AS blk_read_time_seconds
+             , sum(s.blk_write_time) / 1000 AS blk_write_time_seconds
+        FROM pg_stat_statements s
+        JOIN pg_database d ON s.dbid = d.oid
+        GROUP BY d.datname;
+      target_databases: ["*"]
+      predicate_query: "SELECT 1 FROM pg_extension WHERE extname = 'pg_stat_statements';"
+      metrics:
+        - datname:
+            description: Name of the database
+            usage: LABEL
+        - calls:
+            description: Total number of query executions
+            usage: COUNTER
+        - total_exec_time_seconds:
+            description: Total execution time in seconds
+            usage: COUNTER
+        - rows:
+            description: Total number of rows returned or affected
+            usage: COUNTER
+        - shared_blks_hit:
+            description: Total number of shared block cache hits
+            usage: COUNTER
+        - shared_blks_read:
+            description: Total number of shared blocks read from disk
+            usage: COUNTER
+        - blk_read_time_seconds:
+            description: Total block read time in seconds
+            usage: COUNTER
+        - blk_write_time_seconds:
+            description: Total block write time in seconds
+            usage: COUNTER
+{{- end }}

charts/cluster/test/monitoring/01-monitoring_cluster-assert.yaml

@@ -19,6 +19,13 @@ spec:
         - name: monitoring-cluster-monitoring
           key: custom-queries
       enablePodMonitor: false
+  bootstrap:
+    initdb:
+      postInitApplicationSQL:
+        - CREATE EXTENSION IF NOT EXISTS pg_stat_statements;
+  postgresql:
+    shared_preload_libraries:
+      - pg_stat_statements
 ---
 apiVersion: monitoring.coreos.com/v1
 kind: PodMonitor
@@ -124,3 +131,14 @@ data:
         - ratio:
             description: Cache hit ratio
             usage: GAUGE
+---
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: monitoring-cluster-monitoring-logical-replication
+
+---
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: monitoring-cluster-monitoring-pg-stat-statements

charts/cluster/test/monitoring/01-monitoring_cluster.yaml

@@ -8,6 +8,9 @@ cluster:
   monitoring:
     enabled: true
     disableDefaultQueries: true
+    instrumentation:
+      logicalReplication: true
+      pgStatStatements: true
     tls:
       enabled: true
     customQueries:

charts/cluster/values.yaml

@@ -327,6 +327,8 @@ cluster:
     instrumentation:
       # -- Enable logical replication metrics
       logicalReplication: true
+      # -- Enable planning and execution statistics for all SQL statements. Increases shared memory usage.
+      pgStatStatements: false
     # -- Whether the default queries should be injected.
     # Set it to true if you don't want to inject default queries into the cluster.
     disableDefaultQueries: false