PG: Add recovery prefetch metrics (#20464)

Author: bonnefoa

postgres/changelog.d/20464.added

@@ -0,0 +1 @@
+PG: Add recovery prefetch metrics

postgres/datadog_checks/postgres/postgres.py

@@ -55,6 +55,7 @@
     QUERY_PG_REPLICATION_STATS_METRICS,
     QUERY_PG_STAT_DATABASE,
     QUERY_PG_STAT_DATABASE_CONFLICTS,
+    QUERY_PG_STAT_RECOVERY_PREFETCH,
     QUERY_PG_STAT_WAL_RECEIVER,
     QUERY_PG_UPTIME,
     REPLICATION_METRICS,
@@ -363,6 +364,7 @@ def dynamic_queries(self):
             queries.append(SUBSCRIPTION_STATE_METRICS)
         if self.version >= V15:
             queries.append(STAT_SUBSCRIPTION_STATS_METRICS)
+            queries.append(QUERY_PG_STAT_RECOVERY_PREFETCH)
         if self.version >= V16:
             if self._config.dbm_enabled:
                 queries.append(STAT_IO_METRICS)

postgres/datadog_checks/postgres/util.py

@@ -249,6 +249,34 @@ def trim_leading_set_stmts(sql):
     ],
 }
 
+QUERY_PG_STAT_RECOVERY_PREFETCH = {
+    'name': 'pg_stat_recovery_prefetch',
+    'query': """
+        SELECT
+            prefetch,
+            hit,
+            skip_init,
+            skip_new,
+            skip_fpw,
+            skip_rep,
+            wal_distance,
+            block_distance,
+            io_depth
+        FROM pg_stat_recovery_prefetch
+    """.strip(),
+    'columns': [
+        {'name': 'recovery_prefetch.prefetch', 'type': 'monotonic_count'},
+        {'name': 'recovery_prefetch.hit', 'type': 'monotonic_count'},
+        {'name': 'recovery_prefetch.skip_init', 'type': 'monotonic_count'},
+        {'name': 'recovery_prefetch.skip_new', 'type': 'monotonic_count'},
+        {'name': 'recovery_prefetch.skip_fpw', 'type': 'monotonic_count'},
+        {'name': 'recovery_prefetch.skip_rep', 'type': 'monotonic_count'},
+        {'name': 'recovery_prefetch.wal_distance', 'type': 'gauge'},
+        {'name': 'recovery_prefetch.block_distance', 'type': 'gauge'},
+        {'name': 'recovery_prefetch.io_depth', 'type': 'gauge'},
+    ],
+}
+
 QUERY_PG_UPTIME = {
     'name': 'pg_uptime',
     'query': "SELECT FLOOR(EXTRACT(EPOCH FROM current_timestamp - pg_postmaster_start_time()))",

postgres/metadata.csv

@@ -17,6 +17,7 @@
     QUERY_PG_REPLICATION_SLOTS,
     QUERY_PG_REPLICATION_SLOTS_STATS,
     QUERY_PG_REPLICATION_STATS_METRICS,
+    QUERY_PG_STAT_RECOVERY_PREFETCH,
     QUERY_PG_STAT_WAL_RECEIVER,
     QUERY_PG_UPTIME,
     SLRU_METRICS,
@@ -478,6 +479,14 @@ def check_subscription_state_metrics(aggregator, expected_tags, count=1):
         aggregator.assert_metric(metric_name, count=count, tags=expected_tags)
 
 
+def check_recovery_prefetch_metrics(aggregator, expected_tags, count=1):
+    if float(POSTGRES_VERSION) < 15.0:
+        return
+
+    for metric_name in _iterate_metric_name(QUERY_PG_STAT_RECOVERY_PREFETCH):
+        aggregator.assert_metric(metric_name, count=count, tags=expected_tags)
+
+
 def check_subscription_stats_metrics(aggregator, expected_tags, count=1):
     if float(POSTGRES_VERSION) < 15:
         return

postgres/tests/common.py

@@ -17,6 +17,7 @@
     check_db_count,
     check_file_wal_metrics,
     check_performance_metrics,
+    check_recovery_prefetch_metrics,
     check_slru_metrics,
     check_snapshot_txid_metrics,
     check_stat_wal_metrics,
@@ -51,6 +52,7 @@ def test_common_logical_replica_metrics(aggregator, integration_check, pg_replic
     check_snapshot_txid_metrics(aggregator, expected_tags=expected_tags)
     check_stat_wal_metrics(aggregator, expected_tags=expected_tags)
     check_file_wal_metrics(aggregator, expected_tags=expected_tags)
+    check_recovery_prefetch_metrics(aggregator, expected_tags=expected_tags)
 
     check_wal_receiver_metrics(aggregator, connected=False, expected_tags=expected_tags)
     check_subscription_metrics(aggregator, expected_tags=expected_tags + ['subscription_name:subscription_cities'])

postgres/tests/test_logical_replication.py

@@ -37,6 +37,7 @@
     check_metrics_metadata,
     check_performance_metrics,
     check_physical_replication_slots,
+    check_recovery_prefetch_metrics,
     check_slru_metrics,
     check_snapshot_txid_metrics,
     check_stat_io_metrics,
@@ -94,6 +95,7 @@ def test_common_metrics(aggregator, integration_check, pg_instance, is_aurora):
     check_logical_replication_slots(aggregator, expected_tags)
     check_physical_replication_slots(aggregator, expected_tags)
     check_snapshot_txid_metrics(aggregator, expected_tags=expected_tags)
+    check_recovery_prefetch_metrics(aggregator, expected_tags=expected_tags)
 
     check_performance_metrics(aggregator, expected_tags=check.debug_stats_kwargs()['tags'], is_aurora=is_aurora)
 

postgres/tests/test_pg_integration.py

@@ -19,6 +19,7 @@
     check_file_wal_metrics,
     check_metrics_metadata,
     check_performance_metrics,
+    check_recovery_prefetch_metrics,
     check_replication_delay,
     check_slru_metrics,
     check_snapshot_txid_metrics,
@@ -53,6 +54,7 @@ def test_common_replica_metrics(aggregator, integration_check, metrics_cache_rep
     check_snapshot_txid_metrics(aggregator, expected_tags=expected_tags)
     check_stat_wal_metrics(aggregator, expected_tags=expected_tags)
     check_file_wal_metrics(aggregator, expected_tags=expected_tags)
+    check_recovery_prefetch_metrics(aggregator, expected_tags=expected_tags)
 
     check_performance_metrics(aggregator, expected_tags=check.debug_stats_kwargs()['tags'])