Synchronize MySQL master and slave data from GitHub Orchestrator's API into ProxySQL.
When nodes fail, GitHub Orchestrator will mark these nodes as down. The synchronization script will remove failed nodes from its rotation.
This PHP script will poll the Orchestrator API at regular intervals and compare the state of the nodes to what is known in ProxySQL.
If the state differs, the ProxySQL servers will be flushed and re-populated with Orchestrator data.
The /api/cluster/alias/yourClusterName endpoint is used as the "source of truth". This API call returns a collection of JSON objects, where each object represents a server.
The following object properties are used in the decision making process:
IsDowntimed: is the server deliberately put in downtime?IsLastCheckValid: did the server pass its last health check?
And specifically for slave servers, the following properties are used:
Slave_SQL_Running: is the SQL replication mechanism running on a slave?Slave_IO_Running: is the I/O replication mechanism running on a slave?
If a server doesn't meet the assertions above, it is not taken into rotation and will not be inserted in ProxySQL upon the next synchronisation run.
The pollInterval configuration item defines the frequency of the runs in seconds. By default this is 10 seconds.
However, this script also offers a webhook that explicitly forces a re-population of the ProxySQL servers. By default the scripts binds to port 8080 on all network interfaces, but the binding can be configured in the binding configuration item.
The webhook was specifically built to be called by the Orchestrator hooks.
In your orchestrator.conf.json file, you could for example add the following call:
...
{
"PostFailoverProcesses": [
"curl -s http://proxy-sync-host:9527/"
],
}
...
After a failover, you can force the synchronization script to run, but calling the webhook. You can also perform this call in the PreFailoverProcesses hook.