MDEV-29252: Avoid masking specific IO thread errors with generic relay log write failure

[abhishek593]

This PR ensures that the slave IO thread reports specific error codes instead of masking them with the generic ER_SLAVE_RELAY_LOG_WRITE_FAILURE.

Changes:

• Updated logic in sql/slave.cc to preserve specific error codes from queue_event.
• Updated existing tests.

[gkodinov]

Please squash all of your commits into a single one.
Please also target 10.11: this is a bug fix and it needs to be in all affected versions.

[gkodinov]

I'd suggest not to send anything here and fix queue_event to send it in all cases.
Note that this also requires checking all the error paths in queue_event() and making sure no tricks are played to work around the below check.

[abhishek593]

I have updated the implementation. I found only 1 instance where we could return without going through err: label in queue_event(), which is queue_old_event(). There, previously, it just sent error 1, instead of standard error codes, and in handle_slave_io(), it printed errors for ER_SLAVE_RELAY_LOG_WRITE_FAILURE.

Now, I change this return error 1 in queue_old_event() directly to ER_SLAVE_RELAY_LOG_WRITE_FAILURE, which goes through the err: label in queue_event(), and all error reporting happens there.

[gkodinov]

Any specific reason to do this? How is it connected to the problem?

[abhishek593]

I did this because previously, when ER_SLAVE_RELAY_LOG_WRITE_FAILURE was handled in handle_slave_io(), it used ER_THD, so to keep the same behaviour. With this change, all errors use ER_THD.

[gkodinov]

OK. I hope there's always a THD when you are here. Let's have the final reviewer check on that.

[CLAassistant]

All committers have signed the CLA.

[gkodinov]

LGTM. Please stand by for the final review.