Aleksandr Bogomiakov
9b0530fb70
Update Kubernetes Services Wiki / Generate and Update K8s Wiki (push) Successful in 9s
72 lines
3.8 KiB
YAML
72 lines
3.8 KiB
YAML
apiVersion: monitoring.coreos.com/v1alpha1
|
|
kind: AlertmanagerConfig
|
|
metadata:
|
|
name: telegram
|
|
labels:
|
|
alertmanager: main
|
|
annotations:
|
|
argocd.argoproj.io/sync-wave: "1"
|
|
spec:
|
|
route:
|
|
groupBy: [cluster, namespace, alertname]
|
|
# A two-minute grace period absorbs alerts which resolve immediately and
|
|
# never need to interrupt a human.
|
|
groupWait: 2m
|
|
groupInterval: 10m
|
|
repeatInterval: 12h
|
|
receiver: telegram
|
|
routes:
|
|
- receiver: "null"
|
|
matchers:
|
|
- name: alertname
|
|
value: Watchdog|InfoInhibitor
|
|
matchType: =~
|
|
# Informational alerts remain queryable in Prometheus but do not page the
|
|
# Telegram chat. Warning and critical alerts still use the parent route.
|
|
- receiver: "null"
|
|
matchers:
|
|
- name: severity
|
|
value: info
|
|
matchType: =
|
|
receivers:
|
|
- name: telegram
|
|
telegramConfigs:
|
|
- botToken:
|
|
name: alertmanager-telegram
|
|
key: bot-token
|
|
# Telegram chat IDs are routing identifiers rather than credentials;
|
|
# AlertmanagerConfig only supports an integer here, not a Secret ref.
|
|
chatID: 124317807
|
|
sendResolved: true
|
|
parseMode: HTML
|
|
# Bound the rendered HTML, including escaped text, before Telegram's
|
|
# 4096-character limit. Keep the full group accessible in Prometheus.
|
|
message: |-
|
|
{{- define "telegram.alert.text" -}}
|
|
{{- $escaped := html . -}}
|
|
{{- $short := printf "%.500s" $escaped -}}
|
|
{{- $short | reReplaceAll "&[^;]*$" "" | safeHtml -}}{{- if ne $short $escaped }}…{{ end -}}
|
|
{{- end -}}
|
|
{{- $alerts := .Alerts.Resolved -}}
|
|
{{- if eq .Status "firing" }}{{ $alerts = .Alerts.Firing }}{{ end -}}
|
|
{{ if eq .Status "firing" }}🔥 <b>FIRING · {{ len .Alerts.Firing }}</b>{{ else }}✅ <b>RESOLVED · {{ len .Alerts.Resolved }}</b>{{ end }}
|
|
{{ with .CommonLabels.alertname }}<b>{{ printf "%.64s" . }}</b>{{ end }}{{ with .CommonLabels.severity }} · <code>{{ printf "%.16s" . }}</code>{{ end }}
|
|
{{ range $index, $alert := $alerts }}{{ if lt $index 3 }}
|
|
{{- $context := "" -}}
|
|
{{- with .Labels.namespace }}{{ $context = printf "Namespace: %s\n" . }}{{ end -}}
|
|
{{- if .Labels.daemonset }}{{ $context = printf "%sDaemonSet: %s\n" $context .Labels.daemonset -}}
|
|
{{- else if .Labels.deployment }}{{ $context = printf "%sDeployment: %s\n" $context .Labels.deployment -}}
|
|
{{- else if .Labels.statefulset }}{{ $context = printf "%sStatefulSet: %s\n" $context .Labels.statefulset -}}
|
|
{{- else if .Labels.job_name }}{{ $context = printf "%sJob: %s\n" $context .Labels.job_name -}}
|
|
{{- else if .Labels.pod }}{{ $context = printf "%sPod: %s\n" $context .Labels.pod -}}
|
|
{{- end -}}
|
|
{{- with .Labels.node }}{{ $context = printf "%sNode: %s\n" $context . -}}
|
|
{{- else }}{{ if and (not .Labels.daemonset) (not .Labels.deployment) (not .Labels.statefulset) .Labels.instance }}{{ $context = printf "%sInstance: %s\n" $context .Labels.instance }}{{ end }}{{ end -}}
|
|
|
|
{{ if eq .Status "firing" }}🔴{{ else }}🟢{{ end }} {{ template "telegram.alert.text" (printf "%s%s\n%s" $context .Annotations.summary .Annotations.description) }}
|
|
Started: {{ .StartsAt | tz "Europe/London" | date "2006-01-02 15:04 MST" }}
|
|
{{ end }}{{ end }}
|
|
{{ if gt (len $alerts) 3 }}Showing the first 3 of {{ len $alerts }} alerts; open Prometheus for the full group.{{ end }}
|
|
<a href="https://prom.hexor.cy/alerts">📈 Open alerts</a>{{ with .CommonAnnotations.runbook_url }}{{ if le (len .) 256 }} · <a href="{{ . }}">📖 Runbook</a>{{ end }}{{ end }}
|
|
- name: "null"
|