From 7bb2b90a138d965c502c377e6253d0a336e1d81b Mon Sep 17 00:00:00 2001
From: jenkins <jenkins@bstein.dev>
Date: Sat, 16 May 2026 12:31:19 -0300
Subject: [PATCH] monitoring: fix gate health status query

---
 scripts/dashboards_render_atlas.py              | 17 +++++++++++++----
 .../monitoring/dashboards/atlas-testing.json    |  2 +-
 .../monitoring/grafana-dashboard-testing.yaml   |  2 +-
 3 files changed, 15 insertions(+), 6 deletions(-)

diff --git a/scripts/dashboards_render_atlas.py b/scripts/dashboards_render_atlas.py
index bd31785b..a6b54b78 100644
--- a/scripts/dashboards_render_atlas.py
+++ b/scripts/dashboards_render_atlas.py
@@ -3601,11 +3601,20 @@ def build_jobs_dashboard():
         f'or on(suite) ((0 * ({runs_by_suite_24h})) - 1))'
     )
     non_failure = PLATFORM_TEST_NON_FAILURE_STATUS
+    current_gate_ok_vector = (
+        f'(({{{checks_selector},result=~"{non_failure}"}} > bool 0) '
+        f'or ({{{checks_selector},status=~"{non_failure}"}} > bool 0))'
+    )
+    current_gate_seen_vector = f'({{{checks_selector}}} > bool 0)'
+    current_gate_ok_checks = (
+        f"sum by (suite) (max by (suite, check) ({current_gate_ok_vector}))"
+    )
+    current_gate_seen_checks = (
+        f"sum by (suite) (max by (suite, check) ({current_gate_seen_vector}))"
+    )
     current_gate_health_by_suite = (
-        f'(100 * sum by (suite) (max by (suite, check) ((({{{checks_selector},result=~"{non_failure}"}} > bool 0) '
-        f'or ({{{checks_selector},status=~"{non_failure}"}} > bool 0))) '
-        f'/ clamp_min(sum by (suite) (max by (suite, check) (({{{checks_selector}}} > bool 0))), 1)) '
-        f'or on(suite) ({selected_suite_zero}))'
+        f"((100 * ({current_gate_ok_checks}) / clamp_min(({current_gate_seen_checks}), 1)) "
+        f"or on(suite) ({selected_suite_zero}))"
     )
     success_history_runs = f'sum by (suite) (increase(platform_quality_gate_runs_total{{{runs_selector}}}[7d]))'
     success_history_by_suite = (
diff --git a/services/monitoring/dashboards/atlas-testing.json b/services/monitoring/dashboards/atlas-testing.json
index 81cfb808..e4d79500 100644
--- a/services/monitoring/dashboards/atlas-testing.json
+++ b/services/monitoring/dashboards/atlas-testing.json
@@ -449,7 +449,7 @@
       },
       "targets": [
         {
-          "expr": "sort((100 * sum by (suite) (max by (suite, check) ((({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\",result=~\"ok|passed|success|not_applicable|skipped|na|n/a\"} > bool 0) or ({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\",status=~\"ok|passed|success|not_applicable|skipped|na|n/a\"} > bool 0))) / clamp_min(sum by (suite) (max by (suite, check) (({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\"} > bool 0))), 1)) or on(suite) ((0 * (count by (suite) (platform_quality_gate_build_info{suite=~\"${suite:regex}\",branch!=\"\",branch=~\"${branch:regex}\",exported_job=\"platform-quality-ci\"}) >= bool 0)))))",
+          "expr": "sort(((100 * (sum by (suite) (max by (suite, check) ((({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\",result=~\"ok|passed|success|not_applicable|skipped|na|n/a\"} > bool 0) or ({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\",status=~\"ok|passed|success|not_applicable|skipped|na|n/a\"} > bool 0))))) / clamp_min((sum by (suite) (max by (suite, check) (({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\"} > bool 0)))), 1)) or on(suite) ((0 * (count by (suite) (platform_quality_gate_build_info{suite=~\"${suite:regex}\",branch!=\"\",branch=~\"${branch:regex}\",exported_job=\"platform-quality-ci\"}) >= bool 0)))))",
           "refId": "A",
           "legendFormat": "{{suite}}",
           "instant": true
diff --git a/services/monitoring/grafana-dashboard-testing.yaml b/services/monitoring/grafana-dashboard-testing.yaml
index 7d6650cc..e01a86d1 100644
--- a/services/monitoring/grafana-dashboard-testing.yaml
+++ b/services/monitoring/grafana-dashboard-testing.yaml
@@ -458,7 +458,7 @@ data:
           },
           "targets": [
             {
-              "expr": "sort((100 * sum by (suite) (max by (suite, check) ((({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\",result=~\"ok|passed|success|not_applicable|skipped|na|n/a\"} > bool 0) or ({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\",status=~\"ok|passed|success|not_applicable|skipped|na|n/a\"} > bool 0))) / clamp_min(sum by (suite) (max by (suite, check) (({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\"} > bool 0))), 1)) or on(suite) ((0 * (count by (suite) (platform_quality_gate_build_info{suite=~\"${suite:regex}\",branch!=\"\",branch=~\"${branch:regex}\",exported_job=\"platform-quality-ci\"}) >= bool 0)))))",
+              "expr": "sort(((100 * (sum by (suite) (max by (suite, check) ((({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\",result=~\"ok|passed|success|not_applicable|skipped|na|n/a\"} > bool 0) or ({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\",status=~\"ok|passed|success|not_applicable|skipped|na|n/a\"} > bool 0))))) / clamp_min((sum by (suite) (max by (suite, check) (({__name__=~\".*_quality_gate_checks_total\",suite=~\"${suite:regex}\",exported_job=\"platform-quality-ci\"} > bool 0)))), 1)) or on(suite) ((0 * (count by (suite) (platform_quality_gate_build_info{suite=~\"${suite:regex}\",branch!=\"\",branch=~\"${branch:regex}\",exported_job=\"platform-quality-ci\"}) >= bool 0)))))",
               "refId": "A",
               "legendFormat": "{{suite}}",
               "instant": true