grafana-mixin: Fix GrafanaRequestsFailing alert (#43116)

Due to PromQL's label matching, both sides of the division will have same series; that means that, whenever there's a 5xx error, both sides will have the same value and the division will be `1`. I believe the idea was to get the ratio of 5xx compared will all status code, and to do that, we need to aggregate the `status_code` dimension away.

grafana-mixin/alerts/alerts.libsonnet

@@ -0,0 +1,31 @@
+{
+  _config+:: {
+    grafanaRequestsFailingThresholdPercent: 50,
+  },
+
+  prometheusAlerts+:: {
+    groups+: [
+      {
+        name: 'GrafanaAlerts',
+        rules: [
+          {
+            alert: 'GrafanaRequestsFailing',
+            expr: |||
+              100 * namespace_job_handler_statuscode:grafana_http_request_duration_seconds_count:rate5m{handler!~"/api/datasources/proxy/:id.*|/api/ds/query|/api/tsdb/query", status_code=~"5.."}
+              / ignoring (status_code)
+              sum without (status_code) (namespace_job_handler_statuscode:grafana_http_request_duration_seconds_count:rate5m{handler!~"/api/datasources/proxy/:id.*|/api/ds/query|/api/tsdb/query"})
+              > %(grafanaRequestsFailingThresholdPercent)s
+            ||| % $._config,
+            labels: {
+              severity: 'warning',
+            },
+            annotations: {
+              message: '{{ $labels.namespace }}/{{ $labels.job }}/{{ $labels.handler }} is experiencing {{ $value | humanize }}% errors',
+            },
+            'for': '5m',
+          },
+        ],
+      },
+    ],
+  },
+}

grafana-mixin/alerts/alerts.yaml

@@ -1,14 +0,0 @@
-groups:
-  - name: GrafanaAlerts
-    rules:
-      - alert: GrafanaRequestsFailing
-        for: 5m
-        expr: |
-          100 * namespace_job_handler_statuscode:grafana_http_request_duration_seconds_count:rate5m{handler!~"/api/datasources/proxy/:id.*|/api/ds/query|/api/tsdb/query", status_code=~"5.."}
-          /
-          namespace_job_handler_statuscode:grafana_http_request_duration_seconds_count:rate5m{handler!~"/api/datasources/proxy/:id.*|/api/ds/query|/api/tsdb/query"}
-          > 0.5
-        labels:
-          severity: 'warning'
-        annotations:
-          message: "'{{ $labels.namespace }}' / '{{ $labels.job }}' / '{{ $labels.handler }}' is experiencing {{ $value | humanize }}% errors"

grafana-mixin/dashboards/dashboards.libsonnet

@@ -0,0 +1,5 @@
+{
+  grafanaDashboards+:: {
+    'grafana-overview.json': (import 'grafana-overview.json'),
+  },
+}

grafana-mixin/mixin.libsonnet

@@ -1,15 +1,3 @@
-{
+(import 'alerts/alerts.libsonnet') +
-  grafanaDashboards: {
+(import 'dashboards/dashboards.libsonnet') +
-    'grafana-overview.json': (import 'dashboards/grafana-overview.json'),
+(import 'rules/rules.libsonnet')
-  },
-
-  // Helper function to ensure that we don't override other rules, by forcing
-  // the patching of the groups list, and not the overall rules object.
-  local importRules(rules) = {
-    groups+: std.native('parseYaml')(rules)[0].groups,
-  },
-
-  prometheusRules+: importRules(importstr 'rules/rules.yaml'),
-
-  prometheusAlerts+: importRules(importstr 'alerts/alerts.yaml'),
-}

grafana-mixin/rules/rules.libsonnet

@@ -0,0 +1,17 @@
+{
+  prometheusRules+:: {
+    groups+: [
+      {
+        name: 'grafana_rules',
+        rules: [
+          {
+            record: 'namespace_job_handler_statuscode:grafana_http_request_duration_seconds_count:rate5m',
+            expr: |||
+              sum by (namespace, job, handler, status_code) (rate(grafana_http_request_duration_seconds_count[5m]))
+            |||,
+          },
+        ],
+      },
+    ],
+  },
+}

grafana-mixin/rules/rules.yaml

@@ -1,7 +0,0 @@
-groups:
-  - name: grafana_rules
-    rules:
-    # Record error rate of http requests excluding dataproxy, /ds/query and /tsdb/query requests
-    - record: namespace_job_handler_statuscode:grafana_http_request_duration_seconds_count:rate5m
-      expr: |
-          sum by (namespace, job, handler, status_code) (rate(grafana_http_request_duration_seconds_count[5m]))