WARNING: This is for proof-of-concept demonstration only.
This can be scheduled via the ProxySQL Scheduler like this:
INSERT INTO scheduler(id, active, interval_ms, filename) values (1, 1, 10000, '/tmp/proxysql-tidb');
LOAD SCHEDULER TO RUNTIME;
SAVE SCHEDULER TO DISK;
You then should check journalctl -u proxysql -f or elswhere for output.
What this does is to do a soft offline of TiDB servers that are in a graceful shutdown.
This require the TiDB server to have a non-zero graceful-wait-before-shutdown
Possible future improvements:
- Logging via slog
- More defensive coding and error checking
- Also onlineing servers when they are healty again
- Adding hostgroup filtering
Other possible solutions:
- pingcap/tidb#58008: Let TiDB no longer respond with OK on
COM_PINGduring shudown. - Add a global variable for the server health/shutdown status. See pingcap/tidb#58007.