enhance: add missing cluster label to mixins

Signed-off-by: QuentinBisson <quentin@giantswarm.io>
grafana · May 3, 2024 · 07ac356 · 07ac356
1 parent a03846b
commit 07ac356
Show file tree

Hide file tree

Showing 3 changed files with 15 additions and 15 deletions.
diff --git a/production/loki-mixin-compiled-ssd/alerts.yaml b/production/loki-mixin-compiled-ssd/alerts.yaml
@@ -4,7 +4,7 @@ groups:
         - alert: LokiRequestErrors
           annotations:
             description: |
-                {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}% errors.
+                {{ $labels.cluster }} {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}% errors.
             summary: Loki request error rate is high.
           expr: |
             100 * sum(rate(loki_request_duration_seconds_count{status_code=~"5.."}[2m])) by (namespace, job, route)
@@ -17,7 +17,7 @@ groups:
         - alert: LokiRequestPanics
           annotations:
             description: |
-                {{ $labels.job }} is experiencing {{ printf "%.2f" $value }}% increase of panics.
+                {{ $labels.cluster }} {{ $labels.job }} is experiencing {{ printf "%.2f" $value }}% increase of panics.
             summary: Loki requests are causing code panics.
           expr: |
             sum(increase(loki_panic_total[10m])) by (namespace, job) > 0
@@ -26,7 +26,7 @@ groups:
         - alert: LokiRequestLatency
           annotations:
             description: |
-                {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}s 99th percentile latency.
+                {{ $labels.cluster }} {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}s 99th percentile latency.
             summary: Loki request error latency is high.
           expr: |
             cluster_namespace_job_route:loki_request_duration_seconds:99quantile{route!~"(?i).*tail.*|/schedulerpb.SchedulerForQuerier/QuerierLoop"} > 1

diff --git a/production/loki-mixin-compiled/alerts.yaml b/production/loki-mixin-compiled/alerts.yaml
@@ -4,7 +4,7 @@ groups:
         - alert: LokiRequestErrors
           annotations:
             description: |
-                {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}% errors.
+                {{ $labels.cluster }} {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}% errors.
             summary: Loki request error rate is high.
           expr: |
             100 * sum(rate(loki_request_duration_seconds_count{status_code=~"5.."}[2m])) by (namespace, job, route)
@@ -17,7 +17,7 @@ groups:
         - alert: LokiRequestPanics
           annotations:
             description: |
-                {{ $labels.job }} is experiencing {{ printf "%.2f" $value }}% increase of panics.
+                {{ $labels.cluster }} {{ $labels.job }} is experiencing {{ printf "%.2f" $value }}% increase of panics.
             summary: Loki requests are causing code panics.
           expr: |
             sum(increase(loki_panic_total[10m])) by (namespace, job) > 0
@@ -26,7 +26,7 @@ groups:
         - alert: LokiRequestLatency
           annotations:
             description: |
-                {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}s 99th percentile latency.
+                {{ $labels.cluster }} {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}s 99th percentile latency.
             summary: Loki request error latency is high.
           expr: |
             cluster_namespace_job_route:loki_request_duration_seconds:99quantile{route!~"(?i).*tail.*|/schedulerpb.SchedulerForQuerier/QuerierLoop"} > 1

diff --git a/production/loki-mixin/alerts.libsonnet b/production/loki-mixin/alerts.libsonnet
@@ -18,9 +18,9 @@
             },
             annotations: {
               summary: 'Loki request error rate is high.',
-              description: |||
-                {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}% errors.
-              |||,
+              description: std.strReplace(|||
+                {{ $labels.cluster }} {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}% errors.
+              |||, 'cluster', $._config.per_cluster_label),
             },
           },
           {
@@ -33,9 +33,9 @@
             },
             annotations: {
               summary: 'Loki requests are causing code panics.',
-              description: |||
-                {{ $labels.job }} is experiencing {{ printf "%.2f" $value }}% increase of panics.
-              |||,
+              description: std.strReplace(|||
+                {{ $labels.cluster }} {{ $labels.job }} is experiencing {{ printf "%.2f" $value }}% increase of panics.
+              |||, 'cluster', $._config.per_cluster_label),
             },
           },
           {
@@ -49,9 +49,9 @@
             },
             annotations: {
               summary: 'Loki request error latency is high.',
-              description: |||
-                {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}s 99th percentile latency.
-              |||,
+              description: std.strReplace(|||
+                {{ $labels.cluster }} {{ $labels.job }} {{ $labels.route }} is experiencing {{ printf "%.2f" $value }}s 99th percentile latency.
+              |||, 'cluster', $._config.per_cluster_label),
             },
           },
           {