MDEV-27317 wsrep_checkpoint order violation due to certification failure

With binlogs enabled, debug assertion ut_ad(xid_seqno > wsrep_seqno)
fired in trx_rseg_update_wsrep_checkpoint() when an applier thread
synced the seqno out of order for write set which had failed
certification. This was caused by releasing commit
order too early when binlogs were on, allowing group
commit to run in parallel and commit following transactions
too early.

Fixed by extending the commit order critical section to cover
call to wsrep_set_SE_checkpoint() also when binlogs are on.

Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
This commit is contained in:
Teemu Ollakka 2023-02-13 18:14:50 +02:00 committed by Julius Goryavsky
parent f70de1451b
commit cadc3efcdd

View File

@ -472,7 +472,13 @@ int Wsrep_high_priority_service::log_dummy_write_set(const wsrep::ws_handle& ws_
if (!WSREP_EMULATE_BINLOG(m_thd))
{
wsrep_register_for_group_commit(m_thd);
ret = ret || cs.provider().commit_order_leave(ws_handle, ws_meta, err);
/* wait_for_prior_commit() ensures that all preceding transactions
have been committed and seqno has been synced into
storage engine. We don't release commit order here yet to
avoid following transactions to sync seqno before
wsrep_set_SE_checkpoint() below returns. This effectively pauses
group commit for the checkpoint operation, but is the only way to
ensure proper ordering. */
m_thd->wait_for_prior_commit();
}
@ -482,10 +488,7 @@ int Wsrep_high_priority_service::log_dummy_write_set(const wsrep::ws_handle& ws_
{
wsrep_unregister_from_group_commit(m_thd);
}
else
{
ret= ret || cs.provider().commit_order_leave(ws_handle, ws_meta, err);
}
ret= ret || cs.provider().commit_order_leave(ws_handle, ws_meta, err);
cs.after_applying();
}
DBUG_RETURN(ret);