minio/docs/metrics/prometheus/README.md

154 lines
5.7 KiB
Markdown

# How to monitor MinIO server with Prometheus? [![Slack](https://slack.min.io/slack?type=svg)](https://slack.min.io)
[Prometheus](https://prometheus.io) is a cloud-native monitoring platform. Prometheus offers a multi-dimensional data model with time series data identified by metric name and key/value pairs. The data collection happens via a pull model over HTTP/HTTPS. Users looking to monitor their MinIO instances can point Prometheus configuration to scrape data from following endpoints.
- MinIO exports Prometheus compatible data by default as an authorized endpoint at `/minio/v2/metrics/cluster`.
- MinIO exports Prometheus compatible data by default which is bucket centric as an authorized endpoint at `/minio/v2/metrics/bucket`.
This document explains how to setup Prometheus and configure it to scrape data from MinIO servers.
## Prerequisites
To get started with MinIO, refer [MinIO QuickStart Document](https://min.io/docs/minio/linux/index.html#quickstart-for-linux).
Follow below steps to get started with MinIO monitoring using Prometheus.
### 1. Download Prometheus
[Download the latest release](https://prometheus.io/download) of Prometheus for your platform, then extract it
```sh
tar xvfz prometheus-*.tar.gz
cd prometheus-*
```
Prometheus server is a single binary called `prometheus` (or `prometheus.exe` on Microsoft Windows). Run the binary and pass `--help` flag to see available options
```sh
./prometheus --help
usage: prometheus [<flags>]
The Prometheus monitoring server
. . .
```
Refer [Prometheus documentation](https://prometheus.io/docs/introduction/first_steps/) for more details.
### 2. Configure authentication type for Prometheus metrics
MinIO supports two authentication modes for Prometheus either `jwt` or `public`, by default MinIO runs in `jwt` mode. To allow public access without authentication for prometheus metrics set environment as follows.
```
export MINIO_PROMETHEUS_AUTH_TYPE="public"
minio server ~/test
```
### 3. Configuring Prometheus
#### 3.1 Authenticated Prometheus config
> If MinIO is configured to expose metrics without authentication, you don't need to use `mc` to generate prometheus config. You can skip reading further and move to 3.2 section.
The Prometheus endpoint in MinIO requires authentication by default. Prometheus supports a bearer token approach to authenticate prometheus scrape requests, override the default Prometheus config with the one generated using mc. To generate a Prometheus config for an alias, use [mc](https://min.io/docs/minio/linux/reference/minio-mc.html#quickstart) as follows `mc admin prometheus generate <alias>`.
The command will generate the `scrape_configs` section of the prometheus.yml as follows:
##### Cluster
```yaml
scrape_configs:
- job_name: minio-job
bearer_token: <secret>
metrics_path: /minio/v2/metrics/cluster
scheme: http
static_configs:
- targets: ['localhost:9000']
```
##### Bucket centric
```
- job_name: minio-job-bucket
bearer_token: <secret>
metrics_path: /minio/v2/metrics/bucket
scheme: http
static_configs:
- targets: ['localhost:9000']
```
#### 3.2 Public Prometheus config
If Prometheus endpoint authentication type is set to `public`. Following prometheus config is sufficient to start scraping metrics data from MinIO.
This can be collected from any server once per collection.
##### Cluster
```yaml
scrape_configs:
- job_name: minio-job
metrics_path: /minio/v2/metrics/cluster
scheme: http
static_configs:
- targets: ['localhost:9000']
```
#### Bucket centric
```yaml
scrape_configs:
- job_name: minio-job-bucket
metrics_path: /minio/v2/metrics/bucket
scheme: http
static_configs:
- targets: ['localhost:9000']
```
##### Node (optional)
Optionally you can also collect per node metrics. This needs to be done on a per server instance.
```yaml
scrape_configs:
- job_name: minio-job
metrics_path: /minio/v2/metrics/node
scheme: http
static_configs:
- targets: ['localhost:9000']
```
### 4. Update `scrape_configs` section in prometheus.yml
To authorize every scrape request, copy and paste the generated `scrape_configs` section in the prometheus.yml and restart the Prometheus service.
### 5. Start Prometheus
Start (or) Restart Prometheus service by running
```sh
./prometheus --config.file=prometheus.yml
```
Here `prometheus.yml` is the name of configuration file. You can now see MinIO metrics in Prometheus dashboard. By default Prometheus dashboard is accessible at `http://localhost:9090`.
Prometheus sets the `Host` header to `domain:port` as part of HTTP operations against the MinIO metrics endpoint. For MinIO deployments behind a load balancer, reverse proxy, or other control plane (HAProxy, nginx, pfsense, opnsense, etc.), ensure the network service supports routing these requests to the deployment.
### 6. Configure Grafana
After Prometheus is configured, you can use Grafana to visualize MinIO metrics. Refer the [document here to setup Grafana with MinIO prometheus metrics](https://github.com/minio/minio/blob/master/docs/metrics/prometheus/grafana/README.md).
## List of metrics exposed by MinIO
- MinIO exports Prometheus compatible data by default as an authorized endpoint at `/minio/v2/metrics/cluster`.
- MinIO exports Prometheus compatible data by default which is bucket centric as an authorized endpoint at `/minio/v2/metrics/bucket`.
All of these can be accessed via Prometheus dashboard. A sample list of exposed metrics along with their definition is available on our public demo server at
```sh
curl https://play.min.io/minio/v2/metrics/cluster
```
### List of metrics reported Cluster and Bucket level
[The list of metrics reported can be here](https://github.com/minio/minio/blob/master/docs/metrics/prometheus/list.md)