# uptime-kuma ![Version: 1.23.13](https://img.shields.io/badge/Version-1.23.13-informational?style=flat-square) ![Type: application](https://img.shields.io/badge/Type-application-informational?style=flat-square) ![AppVersion: 1.23.13](https://img.shields.io/badge/AppVersion-1.23.13-informational?style=flat-square) A self-hosted Monitoring tool like "Uptime-Robot". **Homepage:** ## Maintainers | Name | Email | Url | | -------- | -------------------------------- | --- | | PulseDev | | | ## Source Code - ## Values | Key | Type | Default | Description | | -------------------------------------------------------------------- | ------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------ | | additionalVolumeMounts | list | `[]` | A list of additional volumeMounts to be added to the pod | | additionalVolumes | list | `[]` | A list of additional volumes to be added to the pod | | affinity | object | `{}` | | | dnsConfig | object | `{}` | Use this option to set custom DNS configurations to the created deployment | | dnsPolicy | string | `""` | Use this option to set a custom DNS policy to the created deployment | | fullnameOverride | string | `""` | | | image.pullPolicy | string | `"IfNotPresent"` | | | image.repository | string | `"louislam/uptime-kuma"` | | | image.tag | string | `"1.23.13-debian"` | | | imagePullSecrets | list | `[]` | | | ingress.annotations."nginx.ingress.kubernetes.io/proxy-read-timeout" | string | `"3600"` | | | ingress.annotations."nginx.ingress.kubernetes.io/proxy-send-timeout" | string | `"3600"` | | | ingress.annotations."nginx.ingress.kubernetes.io/server-snippets" | string | `"location / {\n proxy_set_header Upgrade $http_upgrade;\n proxy_http_version 1.1;\n proxy_set_header X-Forwarded-Host $http_host;\n proxy_set_header X-Forwarded-Proto $scheme;\n proxy_set_header X-Forwarded-For $remote_addr;\n proxy_set_header Host $host;\n proxy_set_header Connection \"upgrade\";\n proxy_set_header X-Real-IP $remote_addr;\n proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;\n proxy_set_header Upgrade $http_upgrade;\n proxy_cache_bypass $http_upgrade;\n}\n"` | | | ingress.enabled | bool | `false` | | | ingress.extraLabels | object | `{}` | | | ingress.hosts[0].host | string | `"chart-example.local"` | | | ingress.hosts[0].paths[0].path | string | `"/"` | | | ingress.hosts[0].paths[0].pathType | string | `"ImplementationSpecific"` | | | ingress.tls | list | `[]` | | | livenessProbe.enabled | bool | `true` | | | livenessProbe.initialDelaySeconds | int | `15` | | | livenessProbe.timeoutSeconds | int | `2` | | | nameOverride | string | `""` | | | nodeSelector | object | `{}` | | | podAnnotations | object | `{}` | | | podEnv[0].name | string | `"UPTIME_KUMA_PORT"` | | | podEnv[0].value | string | `"3001"` | | | podLabels | object | `{}` | | | podSecurityContext | object | `{}` | | | readinessProbe.enabled | bool | `true` | | | readinessProbe.initialDelaySeconds | int | `5` | | | resources | object | `{}` | | | securityContext | object | `{}` | | | service.annotations | object | `{}` | | | service.nodePort | string | `nil` | | | service.port | int | `3001` | | | service.type | string | `"ClusterIP"` | | | serviceAccount.annotations | object | `{}` | | | serviceAccount.create | bool | `false` | | | serviceAccount.name | string | `""` | | | serviceMonitor.additionalLabels | object | `{}` | Additional labels to add to the ServiceMonitor | | serviceMonitor.annotations | object | `{}` | Additional annotations to add to the ServiceMonitor | | serviceMonitor.enabled | bool | `false` | | | serviceMonitor.interval | string | `"60s"` | Scrape interval. If not set, the Prometheus default scrape interval is used. | | serviceMonitor.metricRelabelings | list | `[]` | Prometheus [MetricRelabelConfigs] to apply to samples before ingestion | | serviceMonitor.namespace | string | `nil` | Namespace where the ServiceMonitor resource should be created, default is the same as the release namespace | | serviceMonitor.relabelings | list | `[]` | Prometheus [RelabelConfigs] to apply to samples before scraping | | serviceMonitor.scheme | string | `nil` | Scheme to use when scraping, e.g. http (default) or https. | | serviceMonitor.scrapeTimeout | string | `"10s"` | Timeout if metrics can't be retrieved in given time interval | | serviceMonitor.selector | object | `{}` | Prometheus ServiceMonitor selector, only select Prometheus's with these labels (if not set, select any Prometheus) | | serviceMonitor.tlsConfig | object | `{}` | TLS configuration to use when scraping, only applicable for scheme https. | | strategy.type | string | `"Recreate"` | | | tolerations | list | `[]` | | | useDeploy | bool | `true` | | | volume.accessMode | string | `"ReadWriteOnce"` | | | volume.enabled | bool | `true` | | | volume.existingClaim | string | `""` | | | volume.size | string | `"4Gi"` | | --- Autogenerated from chart metadata using [helm-docs v1.13.1](https://github.com/norwoodj/helm-docs/releases/v1.13.1)