[+] improve recommendation metrics, closes #764 (#921)

* [+] improve recommendation metrics, closes #764

Refactor PostgreSQL recommendation metrics to consolidate multiple
findings per metric type.

The changes consolidate all 8 recommendation metrics to use
an aggregation pattern that combines multiple database objects into
comma-separated strings instead of separate rows, limiting results to
50 items with overflow indicators, and actionable guidance templates
were added to extra_info fields.

* [-] fix `metric_storage_name` properties in YAML, fixes #920

* [+] improve "Recommendations" dashboard

* [-] fix object name aggregation by @0xgouda

* [-] tweak recommendations for `reco_add_index` metric

Latest versions of pg_qualstats return not only CREATE INDEX but
affected queries array. So use CREATE INDEX as is in recommendations,
and output `queryids` if available in `extra_info`.

Due to long recommendation text limit concatenated output to 10 items.
Reported by @0xgouda.

------

Co-authored-by: Ahmed Gouda <[email protected]>

Author: pashagolub

contrib/sample.metrics.yaml

@@ -19,7 +19,7 @@
 #        is_instance_level: true
 #        node_status: primary
 #        statement_timeout_seconds: 300
-#        metric_storage_name: optional_metric_storage_name
+#        storage_name: optional_metric_storage_name
 # presets:
 #    preset_name:
 #        description: custom preset for some metrics
@@ -61,7 +61,7 @@ metrics:
     gauges:
       - '*'
     is_instance_level: false
-    metric_storage_name: buffer_cache_hit_ratio
+    storage_name: buffer_cache_hit_ratio
 presets:
   default:
     description: minimal example preset

docker/compose.pgwatch.yml

@@ -8,13 +8,16 @@ services:
     container_name: pgwatch
     environment:
       PW_SOURCES: postgresql://pgwatch@postgres:5432/pgwatch
+      # PW_METRICS: /etc/pgwatch/metrics.yml
     command:
       - "--sink=postgresql://pgwatch@postgres:5432/pgwatch_metrics"
       - "--sink=prometheus://pgwatch:9187/pgwatch"
     ports:
       - "8080:8080"
       - "9187:9187"
       # - "6060:6060" # Uncomment for profiling
+    volumes:
+      - "../internal/metrics/metrics.yaml:/etc/pgwatch/metrics.yml"
     depends_on:
       postgres:
         condition: service_healthy

docker/compose.postgres.yml

@@ -11,7 +11,7 @@ services:
     image: &pgimage postgres-plpython3u:latest
     container_name: postgres
     command:
-      - "-cshared_preload_libraries=pg_stat_statements"
+      - "-cshared_preload_libraries=pg_stat_statements,pg_qualstats"
       - "-cpg_stat_statements.track=all"
       - "-cpg_stat_statements.track_planning=on"
       - "-ctrack_io_timing=on"

docs/reference/metric_definitions.md

@@ -136,7 +136,7 @@ Here is the structure of a metric definition in YAML format:
         is_instance_level: true
         node_status: primary/standby
         statement_timeout_seconds: 300
-        metric_storage_name: some_other_metric_name
+        storage_name: some_other_metric_name
 ```
 
 - *init_sql*
@@ -199,7 +199,7 @@ Here is the structure of a metric definition in YAML format:
     The maximum time in seconds the metric query is allowed to run
     before it's killed. The default is 5 seconds.
 
-- *metric_storage_name*
+- *storage_name*
 
     Enables dynamic "renaming" of metrics at the storage level, i.e.
     declaring almost similar metrics with different names, but the data