etcd monitoring

YMatrix 6 introduces etcd to store the configuration and status information of the cluster. etcd is very critical. If an exception occurs, it will cause the database to be instable or even crash.

This article introduces the methods of installing and deploying etcd monitoring. It is recommended that clusters in production environments install this monitoring because the normal etcd status is the prerequisite for the healthy operation of the database.

1 Install Prometheus (If you already have this software, please skip this step)

Please go to Prometheus official website to download the following content:

Upload the downloaded tar package to the Linux server. If conditions permit, it is recommended to use a separate server to run Prometheus. If the server is limited, it can also be temporarily placed on the master standby node or the master node host.

$ tar xvfz prometheus-*.tar.gz

You can move the unzipped prometheus-* directory to /usr/local/.

2 Set Prometheus pull etcd status

Edit the configuration file.

$ cd prometheus-*
sudo vi prometheus.yml

At the end of the file, add the following content:

  - job_name: "etcd"
    static_configs:
      - targets: ["172.31.33.128:4679", "172.31.45.253:4679", "172.31.35.134:4679"]

Among them, targets is an array that needs to be replaced with the address of each etcd node in the cluster. This information can be found in the /etc/matrixdb6/physical_cluster.toml file on the Master host.

$ cat physical_cluster.toml
cluster_id = '79LhQxjuwmXgSWZCjcdigF'
supervisord_grpc_port = 4617
deployer_port = 4627
etcd_endpoints = ['http://10.0.159.1:4679', 'http://10.0.172.185:4679', 'http://10.0.170.90:4679', 'http://10.0.146.2:4679', 'http://10.0.146.195:4679', 'http://10.0.150.110:4679', 'http://10.0.169.149:4679']

Notes!
If you do not have the /etc/matrixdb6/physical_cluster.toml file on your host and the etcd process cannot be found, it means that your cluster is not deployed with the 6.x architecture and does not involve monitoring etcd.

3 Start Prometheus

./prometheus --config.file=prometheus.yml

Notes!
Usually we want Prometheus to run in the background as a system service, which requires it configured as a systemd service.

4 Install Grafana (If you already have this software, please skip this step)

Please refer to the official document Installing Grafana Software.

Notes!
Grafana 8.2.5 or higher is required.

5 Configure Grafana's etcd board

First log in to the Grafana page, the default address is as follows:

http://<安装节点的IP或者域名>:3000

5.1 Configuring DataSource

172.31.23.21:9090 is the service port of Prometheus deployed the previous step.

5.2 Import Dashboard

Visit the Grafana community: https://grafana.com/grafana/dashboards/?search=Etcd+Cluster+Overview

As shown in the above link, you can search for the Dashboard provided by the community:

Click to enter to get the ID of the Dashboard:

Notes!
The Dashboard corresponding to this ID does not rule out the possibility of changes. Currently, it is 15308. Users please refer to their own search results.

As shown in the figure below, here we import Dashboard No. 15308.

As shown in the figure below, select the data source name we just configured and finally load the Dashboard.

If the server is in the intranet, it may not be possible to import with number 15308. You need to import the JSON file Download Dashboard on other computers that can access the Internet.

6 etcd Dashboard Key Project Introduction

etcd service status and number of members
- Has Leader must be Yes, otherwise the etcd service is not available.
- The number of Members is automatically determined by the number of hosts in the system, with a maximum of 7.
Quorum member list and their respective service process status
- DB Capacity represents the remaining available space. The default database size limit for etcd is 2G. If the remaining space is too small, it may be full in a short time, resulting in the etcd service being unavailable. Compact or Defragment (similar to Vacuum, Vacuum Full) should be considered at this time.
- Mem Used represents the actual memory usage of each etcd instance, usually in several hundred MB.
DB Total Size indicates that the amount of etcd data increases over time. If the curve increases abnormally, it may cause problems such as the database space reaching the upper limit.

Notes!
For more indicators and detailed introductions, please refer to [etcd official documentation] (https://etcd.io/docs/v3.5/metrics/).