infra/charts/victoria-metrics-k8s-stack/files/rules/generated/kube-apiserver-burnrate.rules.yaml
Konstantin Averkiev c45fd1a6ac added vm stack
2025-07-08 17:29:32 +03:00

325 lines
15 KiB
YAML

{{- $Values := (.helm).Values | default .Values }}
{{- $runbookUrl := ($Values.defaultRules).runbookUrl | default "https://runbooks.prometheus-operator.dev/runbooks" }}
{{- $clusterLabel := ($Values.global).clusterLabel | default "cluster" }}
{{- $additionalGroupByLabels := append $Values.defaultRules.additionalGroupByLabels $clusterLabel }}
{{- $groupLabels := join "," $additionalGroupByLabels }}
{{- $grafanaHost := ternary (index (($Values.grafana).ingress).hosts 0) (($Values.external).grafana).host ($Values.grafana).enabled }}
condition: '{{ ($Values.kubeApiServer).enabled }}'
name: kube-apiserver-burnrate.rules
rules:
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward"}[1d]))
-
(
(
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope=~"resource|",le=~"1(\\.0)?"}[1d]))
or
vector(0)
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="namespace",le=~"5(\\.0)?"}[1d]))
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="cluster",le=~"30(\\.0)?"}[1d]))
)
)
+
# errors
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET",code=~"5.."}[1d]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET"}[1d]))
labels:
verb: read
record: apiserver_request:burnrate1d
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward"}[1h]))
-
(
(
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope=~"resource|",le=~"1(\\.0)?"}[1h]))
or
vector(0)
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="namespace",le=~"5(\\.0)?"}[1h]))
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="cluster",le=~"30(\\.0)?"}[1h]))
)
)
+
# errors
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET",code=~"5.."}[1h]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET"}[1h]))
labels:
verb: read
record: apiserver_request:burnrate1h
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward"}[2h]))
-
(
(
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope=~"resource|",le=~"1(\\.0)?"}[2h]))
or
vector(0)
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="namespace",le=~"5(\\.0)?"}[2h]))
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="cluster",le=~"30(\\.0)?"}[2h]))
)
)
+
# errors
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET",code=~"5.."}[2h]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET"}[2h]))
labels:
verb: read
record: apiserver_request:burnrate2h
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward"}[30m]))
-
(
(
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope=~"resource|",le=~"1(\\.0)?"}[30m]))
or
vector(0)
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="namespace",le=~"5(\\.0)?"}[30m]))
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="cluster",le=~"30(\\.0)?"}[30m]))
)
)
+
# errors
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET",code=~"5.."}[30m]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET"}[30m]))
labels:
verb: read
record: apiserver_request:burnrate30m
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward"}[3d]))
-
(
(
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope=~"resource|",le=~"1(\\.0)?"}[3d]))
or
vector(0)
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="namespace",le=~"5(\\.0)?"}[3d]))
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="cluster",le=~"30(\\.0)?"}[3d]))
)
)
+
# errors
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET",code=~"5.."}[3d]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET"}[3d]))
labels:
verb: read
record: apiserver_request:burnrate3d
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward"}[5m]))
-
(
(
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope=~"resource|",le=~"1(\\.0)?"}[5m]))
or
vector(0)
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="namespace",le=~"5(\\.0)?"}[5m]))
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="cluster",le=~"30(\\.0)?"}[5m]))
)
)
+
# errors
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET",code=~"5.."}[5m]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET"}[5m]))
labels:
verb: read
record: apiserver_request:burnrate5m
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward"}[6h]))
-
(
(
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope=~"resource|",le=~"1(\\.0)?"}[6h]))
or
vector(0)
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="namespace",le=~"5(\\.0)?"}[6h]))
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"LIST|GET",subresource!~"proxy|attach|log|exec|portforward",scope="cluster",le=~"30(\\.0)?"}[6h]))
)
)
+
# errors
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET",code=~"5.."}[6h]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"LIST|GET"}[6h]))
labels:
verb: read
record: apiserver_request:burnrate6h
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward"}[1d]))
-
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward",le=~"1(\\.0)?"}[1d]))
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",code=~"5.."}[1d]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE"}[1d]))
labels:
verb: write
record: apiserver_request:burnrate1d
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward"}[1h]))
-
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward",le=~"1(\\.0)?"}[1h]))
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",code=~"5.."}[1h]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE"}[1h]))
labels:
verb: write
record: apiserver_request:burnrate1h
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward"}[2h]))
-
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward",le=~"1(\\.0)?"}[2h]))
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",code=~"5.."}[2h]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE"}[2h]))
labels:
verb: write
record: apiserver_request:burnrate2h
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward"}[30m]))
-
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward",le=~"1(\\.0)?"}[30m]))
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",code=~"5.."}[30m]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE"}[30m]))
labels:
verb: write
record: apiserver_request:burnrate30m
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward"}[3d]))
-
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward",le=~"1(\\.0)?"}[3d]))
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",code=~"5.."}[3d]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE"}[3d]))
labels:
verb: write
record: apiserver_request:burnrate3d
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward"}[5m]))
-
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward",le=~"1(\\.0)?"}[5m]))
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",code=~"5.."}[5m]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE"}[5m]))
labels:
verb: write
record: apiserver_request:burnrate5m
- condition: '{{ true }}'
expr: |-
(
(
# too slow
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_count{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward"}[6h]))
-
sum by ({{ $groupLabels }}) (rate(apiserver_request_sli_duration_seconds_bucket{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",subresource!~"proxy|attach|log|exec|portforward",le=~"1(\\.0)?"}[6h]))
)
+
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE",code=~"5.."}[6h]))
)
/
sum by ({{ $groupLabels }}) (rate(apiserver_request_total{job="apiserver",verb=~"POST|PUT|PATCH|DELETE"}[6h]))
labels:
verb: write
record: apiserver_request:burnrate6h