449 lines
18 KiB
Django/Jinja
449 lines
18 KiB
Django/Jinja
#jinja2:variable_start_string:'[[' , variable_end_string:']]', block_start_string:'[%' , block_end_string:'%]'
|
|
---
|
|
apiVersion: v1
|
|
data:
|
|
etcd2.rules: "### General cluster availability ###\n\n# alert if another failed
|
|
peer will result in an unavailable cluster\nALERT InsufficientPeers\n IF count(up{job=\"etcd-k8s\"}
|
|
== 0) > (count(up{job=\"etcd-k8s\"}) / 2 - 1)\n FOR 3m\n LABELS {\n severity
|
|
= \"critical\"\n }\n ANNOTATIONS {\n summary = \"Etcd cluster small\",\n
|
|
\ description = \"If one more etcd peer goes down the cluster will be unavailable\",\n
|
|
\ }\n\n### HTTP requests alerts ###\n\n# alert if more than 1% of requests to
|
|
an HTTP endpoint have failed with a non 4xx response\nALERT HighNumberOfFailedHTTPRequests\n
|
|
\ IF sum by(method) (rate(etcd_http_failed_total{job=\"etcd-k8s\", code!~\"4[0-9]{2}\"}[5m]))\n
|
|
\ / sum by(method) (rate(etcd_http_received_total{job=\"etcd-k8s\"}[5m])) >
|
|
0.01\n FOR 10m\n LABELS {\n severity = \"warning\"\n }\n ANNOTATIONS {\n
|
|
\ summary = \"a high number of HTTP requests are failing\",\n description
|
|
= \"{{ $value }}% of requests for {{ $labels.method }} failed on etcd instance
|
|
{{ $labels.instance }}\",\n }\n\n# alert if more than 5% of requests to an HTTP
|
|
endpoint have failed with a non 4xx response\nALERT HighNumberOfFailedHTTPRequests\n
|
|
\ IF sum by(method) (rate(etcd_http_failed_total{job=\"etcd-k8s\", code!~\"4[0-9]{2}\"}[5m]))
|
|
\n / sum by(method) (rate(etcd_http_received_total{job=\"etcd-k8s\"}[5m]))
|
|
> 0.05\n FOR 5m\n LABELS {\n severity = \"critical\"\n }\n ANNOTATIONS
|
|
{\n summary = \"a high number of HTTP requests are failing\",\n description
|
|
= \"{{ $value }}% of requests for {{ $labels.method }} failed on etcd instance
|
|
{{ $labels.instance }}\",\n }\n\n# alert if 50% of requests get a 4xx response\nALERT
|
|
HighNumberOfFailedHTTPRequests\n IF sum by(method) (rate(etcd_http_failed_total{job=\"etcd-k8s\",
|
|
code=~\"4[0-9]{2}\"}[5m]))\n / sum by(method) (rate(etcd_http_received_total{job=\"etcd-k8s\"}[5m]))
|
|
> 0.5\n FOR 10m\n LABELS {\n severity = \"critical\"\n }\n ANNOTATIONS
|
|
{\n summary = \"a high number of HTTP requests are failing\",\n description
|
|
= \"{{ $value }}% of requests for {{ $labels.method }} failed with 4xx responses
|
|
on etcd instance {{ $labels.instance }}\",\n }\n\n# alert if the 99th percentile
|
|
of HTTP requests take more than 150ms\nALERT HTTPRequestsSlow\n IF histogram_quantile(0.99,
|
|
rate(etcd_http_successful_duration_second_bucket[5m])) > 0.15\n FOR 10m\n LABELS
|
|
{\n severity = \"warning\"\n }\n ANNOTATIONS {\n summary = \"slow HTTP
|
|
requests\",\n description = \"on ectd instance {{ $labels.instance }} HTTP
|
|
requests to {{ $label.method }} are slow\",\n }\n\n### File descriptor alerts
|
|
###\n\ninstance:fd_utilization = process_open_fds / process_max_fds\n\n# alert
|
|
if file descriptors are likely to exhaust within the next 4 hours\nALERT FdExhaustionClose\n
|
|
\ IF predict_linear(instance:fd_utilization[1h], 3600 * 4) > 1\n FOR 10m\n LABELS
|
|
{\n severity = \"warning\"\n }\n ANNOTATIONS {\n summary = \"file descriptors
|
|
soon exhausted\",\n description = \"{{ $labels.job }} instance {{ $labels.instance
|
|
}} will exhaust in file descriptors soon\",\n }\n\n# alert if file descriptors
|
|
are likely to exhaust within the next hour\nALERT FdExhaustionClose\n IF predict_linear(instance:fd_utilization[10m],
|
|
3600) > 1\n FOR 10m\n LABELS {\n severity = \"critical\"\n }\n ANNOTATIONS
|
|
{\n summary = \"file descriptors soon exhausted\",\n description = \"{{
|
|
$labels.job }} instance {{ $labels.instance }} will exhaust in file descriptors
|
|
soon\",\n }\n\n### etcd proposal alerts ###\n\n# alert if there are several failed
|
|
proposals within an hour\nALERT HighNumberOfFailedProposals\n IF increase(etcd_server_proposal_failed_total{job=\"etcd\"}[1h])
|
|
> 5\n LABELS {\n severity = \"warning\"\n }\n ANNOTATIONS {\n summary
|
|
= \"a high number of failed proposals within the etcd cluster are happening\",\n
|
|
\ description = \"etcd instance {{ $labels.instance }} has seen {{ $value }}
|
|
proposal failures within the last hour\",\n }\n\n### etcd disk io latency alerts
|
|
###\n\n# alert if 99th percentile of fsync durations is higher than 500ms\nALERT
|
|
HighFsyncDurations\n IF histogram_quantile(0.99, rate(etcd_wal_fsync_durations_seconds_bucket[5m]))
|
|
> 0.5\n FOR 10m\n LABELS {\n severity = \"warning\"\n }\n ANNOTATIONS {\n
|
|
\ summary = \"high fsync durations\",\n description = \"ectd instance {{
|
|
$labels.instance }} fync durations are high\",\n }\n"
|
|
kubernetes.rules: |+
|
|
# NOTE: These rules were kindly contributed by the SoundCloud engineering team.
|
|
|
|
### Container resources ###
|
|
|
|
cluster_namespace_controller_pod_container:spec_memory_limit_bytes =
|
|
sum by (cluster,namespace,controller,pod_name,container_name) (
|
|
label_replace(
|
|
container_spec_memory_limit_bytes{container_name!=""},
|
|
"controller", "$1",
|
|
"pod_name", "^(.*)-[a-z0-9]+"
|
|
)
|
|
)
|
|
|
|
cluster_namespace_controller_pod_container:spec_cpu_shares =
|
|
sum by (cluster,namespace,controller,pod_name,container_name) (
|
|
label_replace(
|
|
container_spec_cpu_shares{container_name!=""},
|
|
"controller", "$1",
|
|
"pod_name", "^(.*)-[a-z0-9]+"
|
|
)
|
|
)
|
|
|
|
cluster_namespace_controller_pod_container:cpu_usage:rate =
|
|
sum by (cluster,namespace,controller,pod_name,container_name) (
|
|
label_replace(
|
|
irate(
|
|
container_cpu_usage_seconds_total{container_name!=""}[5m]
|
|
),
|
|
"controller", "$1",
|
|
"pod_name", "^(.*)-[a-z0-9]+"
|
|
)
|
|
)
|
|
|
|
cluster_namespace_controller_pod_container:memory_usage:bytes =
|
|
sum by (cluster,namespace,controller,pod_name,container_name) (
|
|
label_replace(
|
|
container_memory_usage_bytes{container_name!=""},
|
|
"controller", "$1",
|
|
"pod_name", "^(.*)-[a-z0-9]+"
|
|
)
|
|
)
|
|
|
|
cluster_namespace_controller_pod_container:memory_working_set:bytes =
|
|
sum by (cluster,namespace,controller,pod_name,container_name) (
|
|
label_replace(
|
|
container_memory_working_set_bytes{container_name!=""},
|
|
"controller", "$1",
|
|
"pod_name", "^(.*)-[a-z0-9]+"
|
|
)
|
|
)
|
|
|
|
cluster_namespace_controller_pod_container:memory_rss:bytes =
|
|
sum by (cluster,namespace,controller,pod_name,container_name) (
|
|
label_replace(
|
|
container_memory_rss{container_name!=""},
|
|
"controller", "$1",
|
|
"pod_name", "^(.*)-[a-z0-9]+"
|
|
)
|
|
)
|
|
|
|
cluster_namespace_controller_pod_container:memory_cache:bytes =
|
|
sum by (cluster,namespace,controller,pod_name,container_name) (
|
|
label_replace(
|
|
container_memory_cache{container_name!=""},
|
|
"controller", "$1",
|
|
"pod_name", "^(.*)-[a-z0-9]+"
|
|
)
|
|
)
|
|
|
|
cluster_namespace_controller_pod_container:disk_usage:bytes =
|
|
sum by (cluster,namespace,controller,pod_name,container_name) (
|
|
label_replace(
|
|
container_disk_usage_bytes{container_name!=""},
|
|
"controller", "$1",
|
|
"pod_name", "^(.*)-[a-z0-9]+"
|
|
)
|
|
)
|
|
|
|
cluster_namespace_controller_pod_container:memory_pagefaults:rate =
|
|
sum by (cluster,namespace,controller,pod_name,container_name,scope,type) (
|
|
label_replace(
|
|
irate(
|
|
container_memory_failures_total{container_name!=""}[5m]
|
|
),
|
|
"controller", "$1",
|
|
"pod_name", "^(.*)-[a-z0-9]+"
|
|
)
|
|
)
|
|
|
|
cluster_namespace_controller_pod_container:memory_oom:rate =
|
|
sum by (cluster,namespace,controller,pod_name,container_name,scope,type) (
|
|
label_replace(
|
|
irate(
|
|
container_memory_failcnt{container_name!=""}[5m]
|
|
),
|
|
"controller", "$1",
|
|
"pod_name", "^(.*)-[a-z0-9]+"
|
|
)
|
|
)
|
|
|
|
### Cluster resources ###
|
|
|
|
cluster:memory_allocation:percent =
|
|
100 * sum by (cluster) (
|
|
container_spec_memory_limit_bytes{pod_name!=""}
|
|
) / sum by (cluster) (
|
|
machine_memory_bytes
|
|
)
|
|
|
|
cluster:memory_used:percent =
|
|
100 * sum by (cluster) (
|
|
container_memory_usage_bytes{pod_name!=""}
|
|
) / sum by (cluster) (
|
|
machine_memory_bytes
|
|
)
|
|
|
|
cluster:cpu_allocation:percent =
|
|
100 * sum by (cluster) (
|
|
container_spec_cpu_shares{pod_name!=""}
|
|
) / sum by (cluster) (
|
|
container_spec_cpu_shares{id="/"} * on(cluster,instance) machine_cpu_cores
|
|
)
|
|
|
|
cluster:node_cpu_use:percent =
|
|
100 * sum by (cluster) (
|
|
rate(node_cpu{mode!="idle"}[5m])
|
|
) / sum by (cluster) (
|
|
machine_cpu_cores
|
|
)
|
|
|
|
### API latency ###
|
|
|
|
# Raw metrics are in microseconds. Convert to seconds.
|
|
cluster_resource_verb:apiserver_latency:quantile_seconds{quantile="0.99"} =
|
|
histogram_quantile(
|
|
0.99,
|
|
sum by(le,cluster,job,resource,verb) (apiserver_request_latencies_bucket)
|
|
) / 1e6
|
|
cluster_resource_verb:apiserver_latency:quantile_seconds{quantile="0.9"} =
|
|
histogram_quantile(
|
|
0.9,
|
|
sum by(le,cluster,job,resource,verb) (apiserver_request_latencies_bucket)
|
|
) / 1e6
|
|
cluster_resource_verb:apiserver_latency:quantile_seconds{quantile="0.5"} =
|
|
histogram_quantile(
|
|
0.5,
|
|
sum by(le,cluster,job,resource,verb) (apiserver_request_latencies_bucket)
|
|
) / 1e6
|
|
|
|
### Scheduling latency ###
|
|
|
|
cluster:scheduler_e2e_scheduling_latency:quantile_seconds{quantile="0.99"} =
|
|
histogram_quantile(0.99,sum by (le,cluster) (scheduler_e2e_scheduling_latency_microseconds_bucket)) / 1e6
|
|
cluster:scheduler_e2e_scheduling_latency:quantile_seconds{quantile="0.9"} =
|
|
histogram_quantile(0.9,sum by (le,cluster) (scheduler_e2e_scheduling_latency_microseconds_bucket)) / 1e6
|
|
cluster:scheduler_e2e_scheduling_latency:quantile_seconds{quantile="0.5"} =
|
|
histogram_quantile(0.5,sum by (le,cluster) (scheduler_e2e_scheduling_latency_microseconds_bucket)) / 1e6
|
|
|
|
cluster:scheduler_scheduling_algorithm_latency:quantile_seconds{quantile="0.99"} =
|
|
histogram_quantile(0.99,sum by (le,cluster) (scheduler_scheduling_algorithm_latency_microseconds_bucket)) / 1e6
|
|
cluster:scheduler_scheduling_algorithm_latency:quantile_seconds{quantile="0.9"} =
|
|
histogram_quantile(0.9,sum by (le,cluster) (scheduler_scheduling_algorithm_latency_microseconds_bucket)) / 1e6
|
|
cluster:scheduler_scheduling_algorithm_latency:quantile_seconds{quantile="0.5"} =
|
|
histogram_quantile(0.5,sum by (le,cluster) (scheduler_scheduling_algorithm_latency_microseconds_bucket)) / 1e6
|
|
|
|
cluster:scheduler_binding_latency:quantile_seconds{quantile="0.99"} =
|
|
histogram_quantile(0.99,sum by (le,cluster) (scheduler_binding_latency_microseconds_bucket)) / 1e6
|
|
cluster:scheduler_binding_latency:quantile_seconds{quantile="0.9"} =
|
|
histogram_quantile(0.9,sum by (le,cluster) (scheduler_binding_latency_microseconds_bucket)) / 1e6
|
|
cluster:scheduler_binding_latency:quantile_seconds{quantile="0.5"} =
|
|
histogram_quantile(0.5,sum by (le,cluster) (scheduler_binding_latency_microseconds_bucket)) / 1e6
|
|
|
|
ALERT K8SNodeDown
|
|
IF up{job="kubelet"} == 0
|
|
FOR 1h
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "warning"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Kubelet cannot be scraped",
|
|
description = "Prometheus could not scrape a {{ $labels.job }} for more than one hour",
|
|
}
|
|
|
|
ALERT K8SNodeNotReady
|
|
IF kube_node_status_ready{condition="true"} == 0
|
|
FOR 1h
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "warning",
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Node status is NotReady",
|
|
description = "The Kubelet on {{ $labels.node }} has not checked in with the API, or has set itself to NotReady, for more than an hour",
|
|
}
|
|
|
|
ALERT K8SManyNodesNotReady
|
|
IF
|
|
count by (cluster) (kube_node_status_ready{condition="true"} == 0) > 1
|
|
AND
|
|
(
|
|
count by (cluster) (kube_node_status_ready{condition="true"} == 0)
|
|
/
|
|
count by (cluster) (kube_node_status_ready{condition="true"})
|
|
) > 0.2
|
|
FOR 1m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "critical",
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Many K8s nodes are Not Ready",
|
|
description = "{{ $value }} K8s nodes (more than 10% of cluster {{ $labels.cluster }}) are in the NotReady state.",
|
|
}
|
|
|
|
ALERT K8SKubeletNodeExporterDown
|
|
IF up{job="node-exporter"} == 0
|
|
FOR 15m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "warning"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Kubelet node_exporter cannot be scraped",
|
|
description = "Prometheus could not scrape a {{ $labels.job }} for more than one hour.",
|
|
}
|
|
|
|
ALERT K8SKubeletDown
|
|
IF absent(up{job="kubelet"}) or count by (cluster) (up{job="kubelet"} == 0) / count by (cluster) (up{job="kubelet"}) > 0.1
|
|
FOR 1h
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "critical"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Many Kubelets cannot be scraped",
|
|
description = "Prometheus failed to scrape more than 10% of kubelets, or all Kubelets have disappeared from service discovery.",
|
|
}
|
|
|
|
ALERT K8SApiserverDown
|
|
IF up{job="kubernetes"} == 0
|
|
FOR 15m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "warning"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "API server unreachable",
|
|
description = "An API server could not be scraped.",
|
|
}
|
|
|
|
# Disable for non HA kubernetes setups.
|
|
ALERT K8SApiserverDown
|
|
IF absent({job="kubernetes"}) or (count by(cluster) (up{job="kubernetes"} == 1) < count by(cluster) (up{job="kubernetes"}))
|
|
FOR 5m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "critical"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "API server unreachable",
|
|
description = "Prometheus failed to scrape multiple API servers, or all API servers have disappeared from service discovery.",
|
|
}
|
|
|
|
ALERT K8SSchedulerDown
|
|
IF absent(up{job="kube-scheduler"}) or (count by(cluster) (up{job="kube-scheduler"} == 1) == 0)
|
|
FOR 5m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "critical",
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Scheduler is down",
|
|
description = "There is no running K8S scheduler. New pods are not being assigned to nodes.",
|
|
}
|
|
|
|
ALERT K8SControllerManagerDown
|
|
IF absent(up{job="kube-controller-manager"}) or (count by(cluster) (up{job="kube-controller-manager"} == 1) == 0)
|
|
FOR 5m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "critical",
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Controller manager is down",
|
|
description = "There is no running K8S controller manager. Deployments and replication controllers are not making progress.",
|
|
}
|
|
|
|
ALERT K8SConntrackTableFull
|
|
IF 100*node_nf_conntrack_entries / node_nf_conntrack_entries_limit > 50
|
|
FOR 10m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "warning"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Number of tracked connections is near the limit",
|
|
description = "The nf_conntrack table is {{ $value }}% full.",
|
|
}
|
|
|
|
ALERT K8SConntrackTableFull
|
|
IF 100*node_nf_conntrack_entries / node_nf_conntrack_entries_limit > 90
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "critical"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Number of tracked connections is near the limit",
|
|
description = "The nf_conntrack table is {{ $value }}% full.",
|
|
}
|
|
|
|
# To catch the conntrack sysctl de-tuning when it happens
|
|
ALERT K8SConntrackTuningMissing
|
|
IF node_nf_conntrack_udp_timeout > 10
|
|
FOR 10m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "warning",
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Node does not have the correct conntrack tunings",
|
|
description = "Nodes keep un-setting the correct tunings, investigate when it happens.",
|
|
}
|
|
|
|
ALERT K8STooManyOpenFiles
|
|
IF 100*process_open_fds{job=~"kubelet|kubernetes"} / process_max_fds > 50
|
|
FOR 10m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "warning"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "{{ $labels.job }} has too many open file descriptors",
|
|
description = "{{ $labels.node }} is using {{ $value }}% of the available file/socket descriptors.",
|
|
}
|
|
|
|
ALERT K8STooManyOpenFiles
|
|
IF 100*process_open_fds{job=~"kubelet|kubernetes"} / process_max_fds > 80
|
|
FOR 10m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "critical"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "{{ $labels.job }} has too many open file descriptors",
|
|
description = "{{ $labels.node }} is using {{ $value }}% of the available file/socket descriptors.",
|
|
}
|
|
|
|
# Some verbs excluded because they are expected to be long-lasting:
|
|
# WATCHLIST is long-poll, CONNECT is `kubectl exec`.
|
|
ALERT K8SApiServerLatency
|
|
IF histogram_quantile(
|
|
0.99,
|
|
sum without (instance,node,resource) (apiserver_request_latencies_bucket{verb!~"CONNECT|WATCHLIST|WATCH"})
|
|
) / 1e6 > 1.0
|
|
FOR 10m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "warning"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Kubernetes apiserver latency is high",
|
|
description = "99th percentile Latency for {{ $labels.verb }} requests to the kube-apiserver is higher than 1s.",
|
|
}
|
|
|
|
ALERT K8SApiServerEtcdAccessLatency
|
|
IF etcd_request_latencies_summary{quantile="0.99"} / 1e6 > 1.0
|
|
FOR 15m
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "warning"
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Access to etcd is slow",
|
|
description = "99th percentile latency for apiserver to access etcd is higher than 1s.",
|
|
}
|
|
|
|
ALERT K8SKubeletTooManyPods
|
|
IF kubelet_running_pod_count > 100
|
|
LABELS {
|
|
service = "k8s",
|
|
severity = "warning",
|
|
}
|
|
ANNOTATIONS {
|
|
summary = "Kubelet is close to pod limit",
|
|
description = "Kubelet {{$labels.instance}} is running {{$value}} pods, close to the limit of 110",
|
|
}
|
|
|
|
kind: ConfigMap
|
|
metadata:
|
|
creationTimestamp: null
|
|
name: prometheus-k8s-rules
|