clusterloader2/pkg/prometheus/manifests/prometheus-rules.yaml (147 lines of code) (raw):

apiVersion: monitoring.coreos.com/v1 kind: PrometheusRule metadata: labels: prometheus: k8s name: prometheus-k8s-rules namespace: monitoring spec: groups: - name: apiserver.rules rules: - expr: | rate(apiserver_request_duration_seconds_bucket[5m]) and on (job) kubernetes_build_info{major="[0-1]",minor=~"^([0-9]|1[0-9]|2[0-2])$"} or rate(apiserver_request_slo_duration_seconds_bucket[5m]) and on (job) kubernetes_build_info{major="1",minor=~"^2[3-5]$"} or rate(apiserver_request_sli_duration_seconds_bucket[5m]) and on (job) kubernetes_build_info{major="1",minor=~"^(2[6-9]|[3-9][0-9])$"} record: apiserver:request_latency:rate5m - expr: | histogram_quantile(0.99, sum(apiserver:request_latency:rate5m) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_latency:histogram_quantile labels: quantile: "0.99" - expr: | histogram_quantile(0.9, sum(apiserver:request_latency:rate5m) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_latency:histogram_quantile labels: quantile: "0.90" - expr: | histogram_quantile(0.5, sum(apiserver:request_latency:rate5m) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_latency:histogram_quantile labels: quantile: "0.50" - name: probes.rules rules: - expr: | histogram_quantile(0.99, sum(rate(probes_in_cluster_network_latency_seconds_bucket[5m])) by (le)) record: probes:in_cluster_network_latency:histogram_quantile labels: quantile: "0.99" - expr: | histogram_quantile(0.90, sum(rate(probes_in_cluster_network_latency_seconds_bucket[5m])) by (le)) record: probes:in_cluster_network_latency:histogram_quantile labels: quantile: "0.90" - expr: | histogram_quantile(0.50, sum(rate(probes_in_cluster_network_latency_seconds_bucket[5m])) by (le)) record: probes:in_cluster_network_latency:histogram_quantile labels: quantile: "0.50" - expr: | histogram_quantile(0.99, sum(rate(probes_in_cluster_dns_latency_seconds_bucket[5m])) by (le)) record: probes:dns_lookup_latency:histogram_quantile labels: quantile: "0.99" - expr: | histogram_quantile(0.90, sum(rate(probes_in_cluster_dns_latency_seconds_bucket[5m])) by (le)) record: probes:dns_lookup_latency:histogram_quantile labels: quantile: "0.90" - expr: | histogram_quantile(0.50, sum(rate(probes_in_cluster_dns_latency_seconds_bucket[5m])) by (le)) record: probes:dns_lookup_latency:histogram_quantile labels: quantile: "0.50" - expr: | sum(max_over_time(probes_dns_propagation_seconds[1h]))/sum(max_over_time(probes_dns_propagation_count[1h])) record: probes:dns_propagation_average - name: kube-proxy.rules rules: - expr: | histogram_quantile(0.99, sum(rate(kubeproxy_network_programming_duration_seconds_bucket[5m])) by (le)) record: kubeproxy:kubeproxy_network_programming_duration:histogram_quantile labels: quantile: "0.99" - expr: | histogram_quantile(0.90, sum(rate(kubeproxy_network_programming_duration_seconds_bucket[5m])) by (le)) record: kubeproxy:kubeproxy_network_programming_duration:histogram_quantile labels: quantile: "0.90" - expr: | histogram_quantile(0.50, sum(rate(kubeproxy_network_programming_duration_seconds_bucket[5m])) by (le)) record: kubeproxy:kubeproxy_network_programming_duration:histogram_quantile labels: quantile: "0.50" - expr: | histogram_quantile(0.99, sum(rate(kubeproxy_network_programming_duration_seconds_bucket[5m])) by (le, pod)) record: kubeproxy:kubeproxy_network_programming_duration:histogram_quantile_by_pod labels: quantile: "0.99" - expr: | histogram_quantile(0.90, sum(rate(kubeproxy_network_programming_duration_seconds_bucket[5m])) by (le, pod)) record: kubeproxy:kubeproxy_network_programming_duration:histogram_quantile_by_pod labels: quantile: "0.90" - expr: | histogram_quantile(0.50, sum(rate(kubeproxy_network_programming_duration_seconds_bucket[5m])) by (le, pod)) record: kubeproxy:kubeproxy_network_programming_duration:histogram_quantile_by_pod labels: quantile: "0.50" - name: apiserver.1m.rules rules: - expr: | histogram_quantile(0.99, sum(rate(apiserver_request_duration_seconds_bucket[1m])) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_latency_1m:histogram_quantile labels: quantile: "0.99" - expr: | histogram_quantile(0.9, sum(rate(apiserver_request_duration_seconds_bucket[1m])) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_latency_1m:histogram_quantile labels: quantile: "0.90" - expr: | histogram_quantile(0.5, sum(rate(apiserver_request_duration_seconds_bucket[1m])) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_latency_1m:histogram_quantile labels: quantile: "0.50" - expr: | histogram_quantile(0.99, sum(rate(apiserver_request_slo_duration_seconds_bucket[1m])) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_slo_latency_1m:histogram_quantile labels: quantile: "0.99" - expr: | histogram_quantile(0.9, sum(rate(apiserver_request_slo_duration_seconds_bucket[1m])) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_slo_latency_1m:histogram_quantile labels: quantile: "0.90" - expr: | histogram_quantile(0.5, sum(rate(apiserver_request_slo_duration_seconds_bucket[1m])) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_slo_latency_1m:histogram_quantile labels: quantile: "0.50" - expr: | histogram_quantile(0.99, sum(rate(apiserver_request_sli_duration_seconds_bucket[1m])) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_sli_latency_1m:histogram_quantile labels: quantile: "0.99" - expr: | histogram_quantile(0.9, sum(rate(apiserver_request_sli_duration_seconds_bucket[1m])) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_sli_latency_1m:histogram_quantile labels: quantile: "0.90" - expr: | histogram_quantile(0.5, sum(rate(apiserver_request_sli_duration_seconds_bucket[1m])) by (resource, subresource, verb, scope, le)) record: apiserver:apiserver_request_sli_latency_1m:histogram_quantile labels: quantile: "0.50"