MDEV-41014 START SLAVE zeroes Rpl_semi_sync_slave_send_ack

[FarihaIS]

Description

START SLAVE zeroed the global Rpl_semi_sync_slave_send_ack status variable even when another multi-source connection was already active and had accumulated ACKs.

Rpl_semi_sync_slave_send_ack is a single global counter, but slave_start() runs once per master connection and resets it to 0 unconditionally. With multi-source replication, starting one connection's I/O thread wiped the ACKs already counted by other still-running connections.

Remove the reset from slave_start(). Like every other semi-sync status counter, this variable is now cumulative and no longer cleared when an individual connection's I/O thread starts.

Add a test to suite/rpl.

Release Notes

N/A

How can this PR be tested?

Execute the rpl suite in mysql-test-run.

This commit adds rpl.rpl_semi_sync_slave_send_ack, which sets up a slave with two master connections ('' and other), lets the '' connection send semi-sync ACKs, then starts the other connection's I/O thread and checks that the counter is preserved.

Before the fix

Running rpl.rpl_semi_sync_slave_send_ack fails as below:

rpl.rpl_semi_sync_slave_send_ack         [ fail ]
        Test ended at 2026-09-23 23:14:18

CURRENT_TEST: rpl.rpl_semi_sync_slave_send_ack
--- /quick-rebuilds/mariadb-server/mysql-test/suite/rpl/r/rpl_semi_sync_slave_send_ack.result       2026-09-23 23:13:52.965758407 +0000
+++ /quick-rebuilds/mariadb-server/mysql-test/suite/rpl/r/rpl_semi_sync_slave_send_ack.reject       2026-09-23 23:14:18.766451533 +0000
@@ -30,7 +30,7 @@
 include/wait_for_slave_io_to_start.inc
 SHOW STATUS LIKE 'Rpl_semi_sync_slave_send_ack';
 Variable_name  Value
-Rpl_semi_sync_slave_send_ack   2
+Rpl_semi_sync_slave_send_ack   0
 include/stop_slave_io.inc
 RESET SLAVE ALL;
 SET @@SESSION.default_master_connection= '';

Result content mismatch

After the fix

Running rpl.rpl_semi_sync_slave_send_ack succeeds:

rpl.rpl_semi_sync_slave_send_ack         [ pass ]    464
--------------------------------------------------------------------------
The servers were restarted 0 times
Spent 0.464 of 2 seconds executing testcases

Completed: All 1 tests were successful.

Basing the PR against the correct MariaDB version

• This is a bug fix, and the PR is based against the earliest maintained branch on which the bug can be reproduced: 11.4.

Copyright

All new code of the whole pull request, including one or several files that are either new files or modified ones, are contributed under the BSD-new license. I am contributing on behalf of my employer Amazon Web Services, Inc.

[gkodinov]

Thank you for your contribution! This is a preliminary review.

LGTM. Please stand by for the final review.

[bnestere]

Hi @FarihaIS !

Thank you for submitting this patch, though I'm not so sure we want to do this. Where the original report is a bug, this change makes it so the counter will never reset on a slave restart. I imagine the overall fix for this will be more akin to MDEV-40941, which will likely move many of these semi-sync counters/variables to be per-channel rather than global.

But with that said, semi-sync replication has much deeper & more important problems to fix (which are unfortunately not beginner friendly).

@ParadoxV5 do you have any thoughts here? I lean towards leaving this as-is in GA (I don't imagine many people encounter it, as no other users have reported this).

[ParadoxV5]

Thank you for submitting this patch, though I'm not so sure we want to do this. Where the original report is a bug, this change makes it so the counter will never reset on a slave restart.

Yeah, if Rpl_semi_sync_slave_send_ack is defined to be a lifetime counter, then never resetting it makes sense; but it clearly was de facto defined to reset each START SLAVE.

I imagine the overall fix for this will be more akin to MDEV-40941, which will likely move many of these semi-sync counters/variables to be per-channel rather than global.

Yeah, this has the same situation as in #5615 (comment).

I wonder if there’s a creative compromise for maintenance versions.
What’s our clarification on “semi-sync status variables do not support multi-source”?