Merge branch 'fix/grafana-rc6-qa-bugs' into 'main'

Sarumyan · Sarumyan · commit ed9979b05057 · 2026-06-01T14:52:33.000Z
fix(grafana): RC6 demo QA pass — Dashboard 3 query text, Dashboard 6 title TODO, ASH legend dedup, default time range, version banner

Closes #217

See merge request postgres-ai/postgresai!281
diff --git a/config/grafana/dashboards/Dashboard_10_Index health.json b/config/grafana/dashboards/Dashboard_10_Index health.json
@@ -1215,7 +1215,7 @@
     ]
   },
   "time": {
-    "from": "now-6h",
+    "from": "now-1h",
     "to": "now"
   },
   "timepicker": {},
diff --git a/config/grafana/dashboards/Dashboard_11_Single_index_analysis.json b/config/grafana/dashboards/Dashboard_11_Single_index_analysis.json
@@ -839,7 +839,7 @@
     ]
   },
   "time": {
-    "from": "now-12h",
+    "from": "now-1h",
     "to": "now"
   },
   "timepicker": {},
diff --git a/config/grafana/dashboards/Dashboard_1_Node_performance_overview.json b/config/grafana/dashboards/Dashboard_1_Node_performance_overview.json
@@ -292,19 +292,6 @@
           "legendFormat": "Idle Internal - {{wait_event_type}}",
           "range": true,
           "refId": "C"
-        },
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "P7A0D6631BB10B34F"
-          },
-          "editorMode": "code",
-          "expr": "sum by (wait_event_type) (pgwatch_wait_events_total)>0",
-          "hide": false,
-          "interval": "20",
-          "legendFormat": " {{wait_event_type}}",
-          "range": true,
-          "refId": "D"
         }
       ],
       "title": "Active session history",
@@ -3569,7 +3556,7 @@
     ]
   },
   "time": {
-    "from": "now-24h",
+    "from": "now-1h",
     "to": "now"
   },
   "timepicker": {},
diff --git a/config/grafana/dashboards/Dashboard_2_Aggregated_query_analysis.json b/config/grafana/dashboards/Dashboard_2_Aggregated_query_analysis.json
@@ -3520,7 +3520,7 @@
     ]
   },
   "time": {
-    "from": "now-6h",
+    "from": "now-1h",
     "to": "now"
   },
   "timepicker": {},
diff --git a/config/grafana/dashboards/Dashboard_3_Single_query_analysis.json b/config/grafana/dashboards/Dashboard_3_Single_query_analysis.json
@@ -92,7 +92,7 @@
           "editorMode": "code",
           "format": "table",
           "rawQuery": true,
-          "rawSql": "WITH q AS ( SELECT data ->> 'query' AS qt FROM pgss_queryid_queries WHERE (data ->> 'queryid') = '$query_id' LIMIT 1 ) SELECT qt AS \"Query text\" FROM q UNION ALL SELECT '(query text not yet collected for this queryid — pgwatch writes it on the next scrape; refresh in ~30s)' WHERE NOT EXISTS (SELECT 1 FROM q)",
+          "rawSql": "WITH q AS ( SELECT data ->> 'query' AS qt FROM pgss_queryid_queries WHERE (data ->> 'queryid') = '$query_id' AND (data ? 'real_dbname') AND (data ->> 'real_dbname') = '${db_name}' ORDER BY time DESC LIMIT 1 ) SELECT qt AS \"Query text\" FROM q UNION ALL SELECT '(query text not yet collected for this queryid \u2014 pgwatch writes it on the next scrape; refresh in ~30s)' WHERE NOT EXISTS (SELECT 1 FROM q)",
           "refId": "A",
           "sql": {
             "columns": [
@@ -304,19 +304,6 @@
           "legendFormat": "{{wait_event_type}}:{{wait_event}}",
           "range": true,
           "refId": "A"
-        },
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "P7A0D6631BB10B34F"
-          },
-          "editorMode": "code",
-          "expr": "sum by (wait_event_type) (pgwatch_wait_events_total)>0",
-          "hide": false,
-          "interval": "20",
-          "legendFormat": " {{wait_event_type}}",
-          "range": true,
-          "refId": "D"
         }
       ],
       "title": "Active session history",
@@ -2226,7 +2213,7 @@
   },
   "timepicker": {},
   "timezone": "utc",
-      "title": "03. Single queryid analysis",
+  "title": "03. Single queryid analysis",
   "uid": "db52944d-b025-4e18-b70b-89c0af3e7e41",
   "version": 2
-}
+}
diff --git a/config/grafana/dashboards/Dashboard_4_Wait_Sampling_Dashboard.json b/config/grafana/dashboards/Dashboard_4_Wait_Sampling_Dashboard.json
@@ -323,19 +323,6 @@
           "legendFormat": "Idle Internal - {{wait_event_type}}",
           "range": true,
           "refId": "C"
-        },
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "P7A0D6631BB10B34F"
-          },
-          "editorMode": "code",
-          "expr": "sum by (wait_event_type) (pgwatch_wait_events_total)>0",
-          "hide": false,
-          "interval": "20",
-          "legendFormat": " {{wait_event_type}}",
-          "range": true,
-          "refId": "D"
         }
       ],
       "title": "Active session history",
@@ -603,19 +590,6 @@
           "legendFormat": "{{wait_event_type}} - {{wait_event}}",
           "range": true,
           "refId": "A"
-        },
-        {
-          "datasource": {
-            "type": "prometheus",
-            "uid": "P7A0D6631BB10B34F"
-          },
-          "editorMode": "code",
-          "expr": "sum by (wait_event_type) (pgwatch_wait_events_total)>0",
-          "hide": true,
-          "interval": "20",
-          "legendFormat": " {{wait_event_type}}",
-          "range": true,
-          "refId": "D"
         }
       ],
       "title": "Active session history by event type",
@@ -1050,7 +1024,7 @@
     ]
   },
   "time": {
-    "from": "now-6h",
+    "from": "now-1h",
     "to": "now"
   },
   "timepicker": {},
diff --git a/config/grafana/dashboards/Dashboard_6_Replication_and_HA.json b/config/grafana/dashboards/Dashboard_6_Replication_and_HA.json
@@ -72,12 +72,12 @@
       "list": []
     },
     "time": {
-      "from": "now-6h",
+      "from": "now-1h",
       "to": "now"
     },
     "timepicker": {},
     "timezone": "browser",
-    "title": "06. Replication and HA  -- \"Metrics are collected (part of health check); dashboard – TODO\"",
+    "title": "06. Replication and HA",
     "uid": "9086b3ba-a250-429e-a968-9eabee4fb68a",
     "version": 1
   }
diff --git a/config/grafana/dashboards/Dashboard_7_Autovacuum_and_xmin_horizon.json b/config/grafana/dashboards/Dashboard_7_Autovacuum_and_xmin_horizon.json
@@ -1767,7 +1767,7 @@
     ]
   },
   "time": {
-    "from": "now-6h",
+    "from": "now-1h",
     "to": "now"
   },
   "timepicker": {},
diff --git a/config/grafana/dashboards/Dashboard_8_Table_Stats.json b/config/grafana/dashboards/Dashboard_8_Table_Stats.json
@@ -3382,7 +3382,7 @@
     ]
   },
   "time": {
-    "from": "now-6h",
+    "from": "now-1h",
     "to": "now"
   },
   "timepicker": {},
diff --git a/config/grafana/dashboards/Dashboard_9_Single_table_analysis.json b/config/grafana/dashboards/Dashboard_9_Single_table_analysis.json
@@ -2058,7 +2058,7 @@
     ]
   },
   "time": {
-    "from": "now-6h",
+    "from": "now-1h",
     "to": "now"
   },
   "timepicker": {},
diff --git a/config/grafana/provisioning/grafana.ini b/config/grafana/provisioning/grafana.ini
@@ -1,3 +1,10 @@
+[analytics]
+# Suppress Grafana's update-check phone-home and the "New version available"
+# banner that renders on every dashboard. Demo deployments should look
+# polished; the banner is visually distracting and offers no operator value
+# in a pinned-version monitoring stack.
+check_for_updates = false
+
 [users]
 home_page = /d/f90500a0-a12e-4081-a2f0-07ed96f27915/1-postgres-node-performance-overview-high-level/
 default_theme = desertbloom
diff --git a/postgres_ai_helm/values.yaml b/postgres_ai_helm/values.yaml
@@ -238,6 +238,14 @@ grafana:
   plugins:
     - yesoreyeram-infinity-datasource
 
+  # Mirror config/grafana/provisioning/grafana.ini for the Helm-deployed
+  # Grafana. Without an explicit check_for_updates=false, Grafana phones
+  # home and renders a "New version available" banner across every
+  # dashboard — distracting for demo environments and offering no
+  # operator value when the Grafana image is pinned by this chart.
+  grafana.ini:
+    analytics:
+      check_for_updates: false
 
   persistence:
     enabled: true
diff --git a/quality/gitlab-ci-quality.yml b/quality/gitlab-ci-quality.yml
@@ -7,6 +7,31 @@
 # These jobs add quality-specific checks to the CI pipeline.
 # They complement (not replace) the existing test jobs.
 
+# --- Grafana Dashboard Lint ---
+# Fast, dependency-light JSON-only lint that catches common dashboard bugs
+# surfaced by manual QA on the rc.6 demo:
+#   - Postgres-source panels not pinning the database via ${db_name}
+#     (silent "No data" with a tiny pink warning triangle)
+#   - Dashboard titles leaking TODO / WIP / FIXME markers
+#   - Duplicate targets within a panel (legend duplicates)
+#   - Default time ranges wider than 1 hour (empty-looking graphs on demo)
+#   - Grafana auto-update / "new version available" banner not disabled
+# The tests live under tests/grafana_dashboards/ and run on every MR.
+quality:grafana-dashboards-lint:
+  stage: test
+  image: python:3.11-bullseye
+  variables:
+    GIT_STRATEGY: fetch
+    PIP_DISABLE_PIP_VERSION_CHECK: "1"
+    PIP_NO_CACHE_DIR: "1"
+  before_script:
+    - pip install pytest pyyaml
+  script:
+    - python -m pytest tests/grafana_dashboards -v --tb=short
+  rules:
+    - if: '$CI_PIPELINE_SOURCE == "merge_request_event"'
+    - if: '$CI_COMMIT_BRANCH == "main"'
+
 # --- Schema Validation ---
 # Ensures all health check JSON schemas are valid and consistent
 quality:schema-validation:
diff --git a/tests/grafana_dashboards/__init__.py b/tests/grafana_dashboards/__init__.py
diff --git a/tests/grafana_dashboards/conftest.py b/tests/grafana_dashboards/conftest.py
@@ -0,0 +1,65 @@
+"""Shared fixtures for Grafana dashboard JSON lint tests.
+
+Each test parametrizes over every dashboard JSON file found in:
+  - config/grafana/dashboards/
+  - postgres_ai_helm/config/grafana/dashboards/
+
+The two trees must be kept in sync — every fix that lands in one MUST land in
+the other. The lint tests therefore run against BOTH and fail if either
+diverges.
+"""
+from __future__ import annotations
+
+import glob
+import json
+import os
+from pathlib import Path
+from typing import Iterable
+
+import pytest
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+
+DASHBOARD_DIRS = (
+    REPO_ROOT / "config" / "grafana" / "dashboards",
+    REPO_ROOT / "postgres_ai_helm" / "config" / "grafana" / "dashboards",
+)
+
+
+def _dashboard_paths() -> list[Path]:
+    paths: list[Path] = []
+    for d in DASHBOARD_DIRS:
+        if not d.exists():
+            continue
+        paths.extend(sorted(Path(p) for p in glob.glob(str(d / "*.json"))))
+    return paths
+
+
+def dashboard_paths() -> list[Path]:
+    return _dashboard_paths()
+
+
+def iter_panels(dashboard: dict) -> Iterable[dict]:
+    """Yield every panel in a dashboard, descending into collapsed row panels."""
+    for p in dashboard.get("panels", []) or []:
+        yield p
+        # Row panels can contain nested panels (collapsed rows)
+        for nested in p.get("panels", []) or []:
+            yield nested
+
+
+@pytest.fixture(scope="session")
+def dashboards() -> list[tuple[Path, dict]]:
+    out: list[tuple[Path, dict]] = []
+    for path in _dashboard_paths():
+        with open(path) as f:
+            out.append((path, json.load(f)))
+    return out
+
+
+def pytest_generate_tests(metafunc):
+    """Parametrize tests that take a `dashboard_path` arg over every dashboard."""
+    if "dashboard_path" in metafunc.fixturenames:
+        paths = _dashboard_paths()
+        ids = [os.path.relpath(p, REPO_ROOT) for p in paths]
+        metafunc.parametrize("dashboard_path", paths, ids=ids)
diff --git a/tests/grafana_dashboards/test_d3_query_text_panel_compose.py b/tests/grafana_dashboards/test_d3_query_text_panel_compose.py
@@ -0,0 +1,75 @@
+"""Bug 1 + main(0916c30) compose — D3 first panel must keep BOTH fixes.
+
+Dashboard 3 panel id=18 (the "Query text" panel at the top) carries two
+independent fixes that must coexist:
+
+1. ``${db_name}`` template-variable pin in the predicate. Original Bug 1
+   fix on this MR's branch — without it Grafana renders a silent "No
+   data" with a pink/magenta warning triangle when the provisioned
+   Postgres data source has no default database (see
+   ``test_postgres_panels_pin_database.py``).
+
+2. Graceful "query text not yet collected" fallback that renders when
+   the requested queryid has not been scraped yet. Landed on ``main``
+   in commit 0916c30 (``fix(dashboards): emit clean var-db_name in
+   drill-down links``) via a UNION ALL against a CTE.
+
+The naive resolution of the rebase conflict between these two changes
+would have taken just one side and silently lost the other. This test
+asserts both invariants on D3 panel id=18 specifically so future
+maintainers cannot accidentally simplify the query and lose either
+behavior.
+"""
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+DASHBOARD = (
+    Path(__file__).parent.parent.parent
+    / "config"
+    / "grafana"
+    / "dashboards"
+    / "Dashboard_3_Single_query_analysis.json"
+)
+QUERY_TEXT_PANEL_ID = 18
+
+# Markers we expect in the composed rawSql.
+DB_NAME_PIN = "${db_name}"
+FALLBACK_MARKER = "query text not yet collected"
+
+
+def _query_text_panel():
+    with open(DASHBOARD) as f:
+        dashboard = json.load(f)
+    for panel in dashboard.get("panels", []) or []:
+        if panel.get("id") == QUERY_TEXT_PANEL_ID:
+            return panel
+    raise AssertionError(
+        f"D3 panel id={QUERY_TEXT_PANEL_ID} not found in {DASHBOARD.name}; "
+        "did the panel layout change?"
+    )
+
+
+def test_d3_query_text_panel_pins_db_name():
+    """The composed rawSql must reference ${db_name} so the partition
+    lookup uses the right index AND Grafana does not fall back to a
+    missing default database."""
+    panel = _query_text_panel()
+    raw_sql = (panel.get("targets") or [{}])[0].get("rawSql") or ""
+    assert DB_NAME_PIN in raw_sql, (
+        f"D3 panel id={QUERY_TEXT_PANEL_ID} rawSql lost the ${{db_name}} pin: "
+        f"{raw_sql!r}"
+    )
+
+
+def test_d3_query_text_panel_keeps_graceful_fallback():
+    """The composed rawSql must keep the 'query text not yet collected'
+    fallback so the panel never silently shows 'No data' when the
+    queryid is valid but hasn't been scraped yet."""
+    panel = _query_text_panel()
+    raw_sql = (panel.get("targets") or [{}])[0].get("rawSql") or ""
+    assert FALLBACK_MARKER in raw_sql, (
+        f"D3 panel id={QUERY_TEXT_PANEL_ID} rawSql lost the graceful "
+        f"'query text not yet collected' fallback: {raw_sql!r}"
+    )
diff --git a/tests/grafana_dashboards/test_default_time_range.py b/tests/grafana_dashboards/test_default_time_range.py
diff --git a/tests/grafana_dashboards/test_grafana_check_for_updates_disabled.py b/tests/grafana_dashboards/test_grafana_check_for_updates_disabled.py
diff --git a/tests/grafana_dashboards/test_no_duplicate_targets_in_panel.py b/tests/grafana_dashboards/test_no_duplicate_targets_in_panel.py
diff --git a/tests/grafana_dashboards/test_postgres_panels_pin_database.py b/tests/grafana_dashboards/test_postgres_panels_pin_database.py
diff --git a/tests/grafana_dashboards/test_titles_no_todo_markers.py b/tests/grafana_dashboards/test_titles_no_todo_markers.py

Original file line number	Diff line number	Diff line change
`@@ -1215,7 +1215,7 @@`
`1215`	`1215`	`]`
`1216`	`1216`	`},`
`1217`	`1217`	`"time": {`
`1218`		`- "from": "now-6h",`
	`1218`	`+ "from": "now-1h",`
`1219`	`1219`	`"to": "now"`
`1220`	`1220`	`},`
`1221`	`1221`	`"timepicker": {},`
Original file line number	Diff line number	Diff line change
`@@ -839,7 +839,7 @@`
`839`	`839`	`]`
`840`	`840`	`},`
`841`	`841`	`"time": {`
`842`		`- "from": "now-12h",`
	`842`	`+ "from": "now-1h",`
`843`	`843`	`"to": "now"`
`844`	`844`	`},`
`845`	`845`	`"timepicker": {},`
Original file line number	Diff line number	Diff line change
`@@ -3520,7 +3520,7 @@`
`3520`	`3520`	`]`
`3521`	`3521`	`},`
`3522`	`3522`	`"time": {`
`3523`		`- "from": "now-6h",`
	`3523`	`+ "from": "now-1h",`
`3524`	`3524`	`"to": "now"`
`3525`	`3525`	`},`
`3526`	`3526`	`"timepicker": {},`
Original file line number	Diff line number	Diff line change
`@@ -1767,7 +1767,7 @@`
`1767`	`1767`	`]`
`1768`	`1768`	`},`
`1769`	`1769`	`"time": {`
`1770`		`- "from": "now-6h",`
	`1770`	`+ "from": "now-1h",`
`1771`	`1771`	`"to": "now"`
`1772`	`1772`	`},`
`1773`	`1773`	`"timepicker": {},`
Original file line number	Diff line number	Diff line change
`@@ -3382,7 +3382,7 @@`
`3382`	`3382`	`]`
`3383`	`3383`	`},`
`3384`	`3384`	`"time": {`
`3385`		`- "from": "now-6h",`
	`3385`	`+ "from": "now-1h",`
`3386`	`3386`	`"to": "now"`
`3387`	`3387`	`},`
`3388`	`3388`	`"timepicker": {},`
Original file line number	Diff line number	Diff line change
`@@ -2058,7 +2058,7 @@`
`2058`	`2058`	`]`
`2059`	`2059`	`},`
`2060`	`2060`	`"time": {`
`2061`		`- "from": "now-6h",`
	`2061`	`+ "from": "now-1h",`
`2062`	`2062`	`"to": "now"`
`2063`	`2063`	`},`
`2064`	`2064`	`"timepicker": {},`