netflux-kubernetes/deploy/dev/charts/prometheus/README.md

227 lines
9.1 KiB
Markdown
Raw Normal View History

2022-05-02 19:05:01 +00:00
# Prometheus
[Prometheus](https://prometheus.io/), a [Cloud Native Computing Foundation](https://cncf.io/) project, is a systems and service monitoring system. It collects metrics from configured targets at given intervals, evaluates rule expressions, displays the results, and can trigger alerts if some condition is observed to be true.
This chart bootstraps a [Prometheus](https://prometheus.io/) deployment on a [Kubernetes](http://kubernetes.io) cluster using the [Helm](https://helm.sh) package manager.
## Prerequisites
- Kubernetes 1.16+
- Helm 3+
## Get Repo Info
```console
helm repo add prometheus-community https://prometheus-community.github.io/helm-charts
helm repo update
```
_See [helm repo](https://helm.sh/docs/helm/helm_repo/) for command documentation._
## Install Chart
```console
helm install [RELEASE_NAME] prometheus-community/prometheus
```
_See [configuration](#configuration) below._
_See [helm install](https://helm.sh/docs/helm/helm_install/) for command documentation._
## Dependencies
By default this chart installs additional, dependent charts:
- [kube-state-metrics](https://github.com/prometheus-community/helm-charts/tree/main/charts/kube-state-metrics)
To disable the dependency during installation, set `kubeStateMetrics.enabled` to `false`.
_See [helm dependency](https://helm.sh/docs/helm/helm_dependency/) for command documentation._
## Uninstall Chart
```console
helm uninstall [RELEASE_NAME]
```
This removes all the Kubernetes components associated with the chart and deletes the release.
_See [helm uninstall](https://helm.sh/docs/helm/helm_uninstall/) for command documentation._
## Upgrading Chart
```console
helm upgrade [RELEASE_NAME] [CHART] --install
```
_See [helm upgrade](https://helm.sh/docs/helm/helm_upgrade/) for command documentation._
### To 15.0
Version 15.0.0 changes the relabeling config, aligning it with the [Prometheus community conventions](https://github.com/prometheus/prometheus/pull/9832). If you've made manual changes to the relabeling config, you have to adapt your changes.
Before you update please execute the following command, to be able to update kube-state-metrics:
```bash
kubectl delete deployments.apps -l app.kubernetes.io/instance=prometheus,app.kubernetes.io/name=kube-state-metrics --cascade=orphan
```
### To 9.0
Version 9.0 adds a new option to enable or disable the Prometheus Server. This supports the use case of running a Prometheus server in one k8s cluster and scraping exporters in another cluster while using the same chart for each deployment. To install the server `server.enabled` must be set to `true`.
### To 5.0
As of version 5.0, this chart uses Prometheus 2.x. This version of prometheus introduces a new data format and is not compatible with prometheus 1.x. It is recommended to install this as a new release, as updating existing releases will not work. See the [prometheus docs](https://prometheus.io/docs/prometheus/latest/migration/#storage) for instructions on retaining your old data.
Prometheus version 2.x has made changes to alertmanager, storage and recording rules. Check out the migration guide [here](https://prometheus.io/docs/prometheus/2.0/migration/).
Users of this chart will need to update their alerting rules to the new format before they can upgrade.
### Example Migration
Assuming you have an existing release of the prometheus chart, named `prometheus-old`. In order to update to prometheus 2.x while keeping your old data do the following:
1. Update the `prometheus-old` release. Disable scraping on every component besides the prometheus server, similar to the configuration below:
```yaml
alertmanager:
enabled: false
alertmanagerFiles:
alertmanager.yml: ""
kubeStateMetrics:
enabled: false
nodeExporter:
enabled: false
pushgateway:
enabled: false
server:
extraArgs:
storage.local.retention: 720h
serverFiles:
alerts: ""
prometheus.yml: ""
rules: ""
```
1. Deploy a new release of the chart with version 5.0+ using prometheus 2.x. In the values.yaml set the scrape config as usual, and also add the `prometheus-old` instance as a remote-read target.
```yaml
prometheus.yml:
...
remote_read:
- url: http://prometheus-old/api/v1/read
...
```
Old data will be available when you query the new prometheus instance.
## Configuration
See [Customizing the Chart Before Installing](https://helm.sh/docs/intro/using_helm/#customizing-the-chart-before-installing). To see all configurable options with detailed comments, visit the chart's [values.yaml](./values.yaml), or run these configuration commands:
```console
helm show values prometheus-community/prometheus
```
You may similarly use the above configuration commands on each chart [dependency](#dependencies) to see it's configurations.
### Scraping Pod Metrics via Annotations
This chart uses a default configuration that causes prometheus to scrape a variety of kubernetes resource types, provided they have the correct annotations. In this section we describe how to configure pods to be scraped; for information on how other resource types can be scraped you can do a `helm template` to get the kubernetes resource definitions, and then reference the prometheus configuration in the ConfigMap against the prometheus documentation for [relabel_config](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#relabel_config) and [kubernetes_sd_config](https://prometheus.io/docs/prometheus/latest/configuration/configuration/#kubernetes_sd_config).
In order to get prometheus to scrape pods, you must add annotations to the the pods as below:
```yaml
metadata:
annotations:
prometheus.io/scrape: "true"
prometheus.io/path: /metrics
prometheus.io/port: "8080"
```
You should adjust `prometheus.io/path` based on the URL that your pod serves metrics from. `prometheus.io/port` should be set to the port that your pod serves metrics from. Note that the values for `prometheus.io/scrape` and `prometheus.io/port` must be enclosed in double quotes.
### Sharing Alerts Between Services
Note that when [installing](#install-chart) or [upgrading](#upgrading-chart) you may use multiple values override files. This is particularly useful when you have alerts belonging to multiple services in the cluster. For example,
```yaml
# values.yaml
# ...
# service1-alert.yaml
serverFiles:
alerts:
service1:
- alert: anAlert
# ...
# service2-alert.yaml
serverFiles:
alerts:
service2:
- alert: anAlert
# ...
```
```console
helm install [RELEASE_NAME] prometheus-community/prometheus -f values.yaml -f service1-alert.yaml -f service2-alert.yaml
```
### RBAC Configuration
Roles and RoleBindings resources will be created automatically for `server` service.
To manually setup RBAC you need to set the parameter `rbac.create=false` and specify the service account to be used for each service by setting the parameters: `serviceAccounts.{{ component }}.create` to `false` and `serviceAccounts.{{ component }}.name` to the name of a pre-existing service account.
> **Tip**: You can refer to the default `*-clusterrole.yaml` and `*-clusterrolebinding.yaml` files in [templates](templates/) to customize your own.
### ConfigMap Files
AlertManager is configured through [alertmanager.yml](https://prometheus.io/docs/alerting/configuration/). This file (and any others listed in `alertmanagerFiles`) will be mounted into the `alertmanager` pod.
Prometheus is configured through [prometheus.yml](https://prometheus.io/docs/operating/configuration/). This file (and any others listed in `serverFiles`) will be mounted into the `server` pod.
### Ingress TLS
If your cluster allows automatic creation/retrieval of TLS certificates (e.g. [cert-manager](https://github.com/jetstack/cert-manager)), please refer to the documentation for that mechanism.
To manually configure TLS, first create/retrieve a key & certificate pair for the address(es) you wish to protect. Then create a TLS secret in the namespace:
```console
kubectl create secret tls prometheus-server-tls --cert=path/to/tls.cert --key=path/to/tls.key
```
Include the secret's name, along with the desired hostnames, in the alertmanager/server Ingress TLS section of your custom `values.yaml` file:
```yaml
server:
ingress:
## If true, Prometheus server Ingress will be created
##
enabled: true
## Prometheus server Ingress hostnames
## Must be provided if Ingress is enabled
##
hosts:
- prometheus.domain.com
## Prometheus server Ingress TLS configuration
## Secrets must be manually created in the namespace
##
tls:
- secretName: prometheus-server-tls
hosts:
- prometheus.domain.com
```
### NetworkPolicy
Enabling Network Policy for Prometheus will secure connections to Alert Manager and Kube State Metrics by only accepting connections from Prometheus Server. All inbound connections to Prometheus Server are still allowed.
To enable network policy for Prometheus, install a networking plugin that implements the Kubernetes NetworkPolicy spec, and set `networkPolicy.enabled` to true.
If NetworkPolicy is enabled for Prometheus' scrape targets, you may also need to manually create a networkpolicy which allows it.