commit af11f212173e6d92629aaface66982c5a776d78e Author: Sachin <sachinsetia1001@gmail.com> Date: Thu Jul 16 14:24:30 2020 +0530 MDEV-23089 rpl_parallel2 fails in 10.5 Problem:- rpl_parallel2 was failing non-deterministically Analysis:- So when we issue FTWRL. Worker thread will be waiting in do_ftwrl_wait(). Then from other connection we issue STOP SLAVE , which turn force_abort = 1 So we exit out of loop in do_fterl_wait(). Then we get deadlock further down the execution path. Solution:- We will assign 1 to skip_event_group when we are aborted in do_ftwrl_wait. And rpl_parallel_entry->pause_sub-id is only reset when force_abort is off in rpl_pause_after_ftwrl diff --git a/mysql-test/suite/rpl/disabled.def b/mysql-test/suite/rpl/disabled.def index 640c4b56cd0..9e52c277726 100644 --- a/mysql-test/suite/rpl/disabled.def +++ b/mysql-test/suite/rpl/disabled.def @@ -19,4 +19,3 @@ rpl_semi_sync_after_sync : fails after MDEV-16172 rpl_slave_grp_exec: MDEV-10514 rpl_auto_increment_update_failure : disabled for now rpl_current_user : waits for MDEV-22374 fix -rpl_parallel2 : waits for MDEV-23089 diff --git a/sql/rpl_parallel.cc b/sql/rpl_parallel.cc index 94882230682..63e0c6bfc20 100644 --- a/sql/rpl_parallel.cc +++ b/sql/rpl_parallel.cc @@ -396,13 +396,14 @@ do_gco_wait(rpl_group_info *rgi, group_commit_orderer *gco, } -static void +static bool do_ftwrl_wait(rpl_group_info *rgi, bool *did_enter_cond, PSI_stage_info *old_stage) { THD *thd= rgi->thd; rpl_parallel_entry *entry= rgi->parallel_entry; uint64 sub_id= rgi->gtid_sub_id; + bool aborted= false; DBUG_ENTER("do_ftwrl_wait"); mysql_mutex_assert_owner(&entry->LOCK_parallel_entry); @@ -425,7 +426,10 @@ do_ftwrl_wait(rpl_group_info *rgi, do { if (entry->force_abort || rgi->worker_error) + { + aborted= true; break; + } if (unlikely(thd->check_killed())) { slave_output_error_info(rgi, thd); @@ -444,7 +448,7 @@ do_ftwrl_wait(rpl_group_info *rgi, if (sub_id > entry->largest_started_sub_id) entry->largest_started_sub_id= sub_id; - DBUG_VOID_RETURN; + DBUG_RETURN(aborted); } @@ -530,7 +534,18 @@ rpl_unpause_after_ftwrl(THD *thd) mysql_mutex_lock(&e->LOCK_parallel_entry); rpt->pause_for_ftwrl = false; mysql_mutex_unlock(&rpt->LOCK_rpl_thread); - e->pause_sub_id= (uint64)ULONGLONG_MAX; + /* + * Dont change pause_sub_id if force_abort is on + * Reason:- This will make skip_event_group one while calling + * do_ftwrl_wait which will make worker to execute transaction in the + * queue. And will create gaps, Transaction between setting of pause_sub_id + * and unsetting wont be applied on slave. So if there is force_abort + * we wont change the pause_sub_id. + * We are not changing pause_for_ftwrl because we want worker to exit in + * event of STOP SLAVE otherwise it will hang forever + */ + if (!e->force_abort) + e->pause_sub_id= (uint64)ULONGLONG_MAX; mysql_cond_broadcast(&e->COND_parallel_entry); mysql_mutex_unlock(&e->LOCK_parallel_entry); } @@ -1224,7 +1239,7 @@ handle_rpl_parallel_thread(void *arg) rgi->worker_error= 1; } if (likely(!skip_event_group)) - do_ftwrl_wait(rgi, &did_enter_cond, &old_stage); + skip_event_group= do_ftwrl_wait(rgi, &did_enter_cond, &old_stage); /* Register ourself to wait for the previous commit, if we need to do -- Regards Sachin Setiya Software Engineer at MariaDB