mariadb/sql/wsrep_thd.cc
sjaakola ef2dbb8dbc MDEV-23328 Server hang due to Galera lock conflict resolution
Mutex order violation when wsrep bf thread kills a conflicting trx,
the stack is

          wsrep_thd_LOCK()
          wsrep_kill_victim()
          lock_rec_other_has_conflicting()
          lock_clust_rec_read_check_and_lock()
          row_search_mvcc()
          ha_innobase::index_read()
          ha_innobase::rnd_pos()
          handler::ha_rnd_pos()
          handler::rnd_pos_by_record()
          handler::ha_rnd_pos_by_record()
          Rows_log_event::find_row()
          Update_rows_log_event::do_exec_row()
          Rows_log_event::do_apply_event()
          Log_event::apply_event()
          wsrep_apply_events()

and mutexes are taken in the order

          lock_sys->mutex -> victim_trx->mutex -> victim_thread->LOCK_thd_data

When a normal KILL statement is executed, the stack is

          innobase_kill_query()
          kill_handlerton()
          plugin_foreach_with_mask()
          ha_kill_query()
          THD::awake()
          kill_one_thread()

        and mutexes are

          victim_thread->LOCK_thd_data -> lock_sys->mutex -> victim_trx->mutex

This patch is the plan D variant for fixing potetial mutex locking
order exercised by BF aborting and KILL command execution.

In this approach, KILL command is replicated as TOI operation.
This guarantees total isolation for the KILL command execution
in the first node: there is no concurrent replication applying
and no concurrent DDL executing. Therefore there is no risk of
BF aborting to happen in parallel with KILL command execution
either. Potential mutex deadlocks between the different mutex
access paths with KILL command execution and BF aborting cannot
therefore happen.

TOI replication is used, in this approach,  purely as means
to provide isolated KILL command execution in the first node.
KILL command should not (and must not) be applied in secondary
nodes. In this patch, we make this sure by skipping KILL
execution in secondary nodes, in applying phase, where we
bail out if applier thread is trying to execute KILL command.
This is effective, but skipping the applying of KILL command
could happen much earlier as well.

This also fixed unprotected calls to wsrep_thd_abort
that will use wsrep_abort_transaction. This is fixed
by holding THD::LOCK_thd_data while we abort transaction.

Reviewed-by: Jan Lindström <jan.lindstrom@mariadb.com>
2021-10-29 20:40:35 +02:00

477 lines
15 KiB
C++

/* Copyright (C) 2013 Codership Oy <info@codership.com>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1335 USA. */
#include "mariadb.h"
#include "wsrep_thd.h"
#include "wsrep_trans_observer.h"
#include "wsrep_high_priority_service.h"
#include "wsrep_storage_service.h"
#include "transaction.h"
#include "rpl_rli.h"
#include "log_event.h"
#include "sql_parse.h"
#include "mysqld.h" // start_wsrep_THD();
#include "wsrep_applier.h" // start_wsrep_THD();
#include "mysql/service_wsrep.h"
#include "debug_sync.h"
#include "slave.h"
#include "rpl_rli.h"
#include "rpl_mi.h"
extern "C" pthread_key(struct st_my_thread_var*, THR_KEY_mysys);
static Wsrep_thd_queue* wsrep_rollback_queue= 0;
static Atomic_counter<uint64_t> wsrep_bf_aborts_counter;
int wsrep_show_bf_aborts (THD *thd, SHOW_VAR *var, char *buff,
enum enum_var_type scope)
{
wsrep_local_bf_aborts= wsrep_bf_aborts_counter;
var->type= SHOW_LONGLONG;
var->value= (char*)&wsrep_local_bf_aborts;
return 0;
}
static void wsrep_replication_process(THD *thd,
void* arg __attribute__((unused)))
{
DBUG_ENTER("wsrep_replication_process");
Wsrep_applier_service applier_service(thd);
WSREP_INFO("Starting applier thread %llu", thd->thread_id);
enum wsrep::provider::status
ret= Wsrep_server_state::get_provider().run_applier(&applier_service);
WSREP_INFO("Applier thread exiting ret: %d thd: %llu", ret, thd->thread_id);
mysql_mutex_lock(&LOCK_wsrep_slave_threads);
wsrep_close_applier(thd);
mysql_cond_broadcast(&COND_wsrep_slave_threads);
mysql_mutex_unlock(&LOCK_wsrep_slave_threads);
delete thd->wsrep_rgi->rli->mi;
delete thd->wsrep_rgi->rli;
thd->wsrep_rgi->cleanup_after_session();
delete thd->wsrep_rgi;
thd->wsrep_rgi= NULL;
if(thd->has_thd_temporary_tables())
{
WSREP_WARN("Applier %lld has temporary tables at exit.",
thd->thread_id);
}
DBUG_VOID_RETURN;
}
static bool create_wsrep_THD(Wsrep_thd_args* args, bool mutex_protected)
{
if (!mutex_protected)
mysql_mutex_lock(&LOCK_wsrep_slave_threads);
ulong old_wsrep_running_threads= wsrep_running_threads;
DBUG_ASSERT(args->thread_type() == WSREP_APPLIER_THREAD ||
args->thread_type() == WSREP_ROLLBACKER_THREAD);
bool res= mysql_thread_create(args->thread_type() == WSREP_APPLIER_THREAD
? key_wsrep_applier : key_wsrep_rollbacker,
args->thread_id(), &connection_attrib,
start_wsrep_THD, (void*)args);
if (res)
WSREP_ERROR("Can't create wsrep thread");
/*
if starting a thread on server startup, wait until the this thread's THD
is fully initialized (otherwise a THD initialization code might
try to access a partially initialized server data structure - MDEV-8208).
*/
if (!mysqld_server_initialized)
{
while (old_wsrep_running_threads == wsrep_running_threads)
{
mysql_cond_wait(&COND_wsrep_slave_threads, &LOCK_wsrep_slave_threads);
}
}
if (!mutex_protected)
mysql_mutex_unlock(&LOCK_wsrep_slave_threads);
return res;
}
bool wsrep_create_appliers(long threads, bool mutex_protected)
{
/* Dont' start slave threads if wsrep-provider or wsrep-cluster-address
is not set.
*/
if (!WSREP_PROVIDER_EXISTS)
{
return false;
}
DBUG_ASSERT(wsrep_cluster_address[0]);
long wsrep_threads=0;
while (wsrep_threads++ < threads)
{
Wsrep_thd_args* args(new Wsrep_thd_args(wsrep_replication_process,
WSREP_APPLIER_THREAD,
pthread_self()));
if (create_wsrep_THD(args, mutex_protected))
{
WSREP_WARN("Can't create thread to manage wsrep replication");
return true;
}
}
return false;
}
static void wsrep_remove_streaming_fragments(THD* thd, const char* ctx)
{
wsrep::transaction_id transaction_id(thd->wsrep_trx().id());
Wsrep_storage_service* storage_service= wsrep_create_storage_service(thd, ctx);
storage_service->store_globals();
storage_service->adopt_transaction(thd->wsrep_trx());
storage_service->remove_fragments();
storage_service->commit(wsrep::ws_handle(transaction_id, 0),
wsrep::ws_meta());
Wsrep_server_state::instance().server_service()
.release_storage_service(storage_service);
wsrep_store_threadvars(thd);
}
static void wsrep_rollback_high_priority(THD *thd, THD *rollbacker)
{
WSREP_DEBUG("Rollbacker aborting SR applier thd (%llu %lu)",
thd->thread_id, thd->real_id);
char* orig_thread_stack= thd->thread_stack;
thd->thread_stack= rollbacker->thread_stack;
DBUG_ASSERT(thd->wsrep_cs().mode() == Wsrep_client_state::m_high_priority);
/* Must be streaming and must have been removed from the
server state streaming appliers map. */
DBUG_ASSERT(thd->wsrep_trx().is_streaming());
DBUG_ASSERT(!Wsrep_server_state::instance().find_streaming_applier(
thd->wsrep_trx().server_id(),
thd->wsrep_trx().id()));
DBUG_ASSERT(thd->wsrep_applier_service);
/* Fragment removal should happen before rollback to make
the transaction non-observable in SR table after the rollback
completes. For correctness the order does not matter here,
but currently it is mandated by checks in some MTR tests. */
wsrep_remove_streaming_fragments(thd, "high priority");
thd->wsrep_applier_service->rollback(wsrep::ws_handle(),
wsrep::ws_meta());
thd->wsrep_applier_service->after_apply();
thd->thread_stack= orig_thread_stack;
WSREP_DEBUG("rollbacker aborted thd: (%llu %lu)",
thd->thread_id, thd->real_id);
/* Will free THD */
Wsrep_server_state::instance().server_service()
.release_high_priority_service(thd->wsrep_applier_service);
}
static void wsrep_rollback_local(THD *thd, THD *rollbacker)
{
WSREP_DEBUG("Rollbacker aborting local thd (%llu %lu)",
thd->thread_id, thd->real_id);
char* orig_thread_stack= thd->thread_stack;
thd->thread_stack= rollbacker->thread_stack;
if (thd->wsrep_trx().is_streaming())
{
wsrep_remove_streaming_fragments(thd, "local");
}
/* Set thd->event_scheduler.data temporarily to NULL to avoid
callbacks to threadpool wait_begin() during rollback. */
auto saved_esd= thd->event_scheduler.data;
thd->event_scheduler.data= 0;
mysql_mutex_lock(&thd->LOCK_thd_data);
/* prepare THD for rollback processing */
thd->reset_for_next_command();
thd->lex->sql_command= SQLCOM_ROLLBACK;
mysql_mutex_unlock(&thd->LOCK_thd_data);
/* Perform a client rollback, restore globals and signal
the victim only when all the resources have been
released */
thd->wsrep_cs().client_service().bf_rollback();
wsrep_reset_threadvars(thd);
/* Assign saved event_scheduler.data back before letting
client to continue. */
thd->event_scheduler.data= saved_esd;
thd->thread_stack= orig_thread_stack;
thd->wsrep_cs().sync_rollback_complete();
WSREP_DEBUG("rollbacker aborted thd: (%llu %lu)",
thd->thread_id, thd->real_id);
}
static void wsrep_rollback_process(THD *rollbacker,
void *arg __attribute__((unused)))
{
DBUG_ENTER("wsrep_rollback_process");
THD* thd= NULL;
DBUG_ASSERT(!wsrep_rollback_queue);
wsrep_rollback_queue= new Wsrep_thd_queue(rollbacker);
WSREP_INFO("Starting rollbacker thread %llu", rollbacker->thread_id);
thd_proc_info(rollbacker, "wsrep aborter idle");
while ((thd= wsrep_rollback_queue->pop_front()) != NULL)
{
mysql_mutex_lock(&thd->LOCK_thd_data);
wsrep::client_state& cs(thd->wsrep_cs());
const wsrep::transaction& tx(cs.transaction());
if (tx.state() == wsrep::transaction::s_aborted)
{
WSREP_DEBUG("rollbacker thd already aborted: %llu state: %d",
(long long)thd->real_id,
tx.state());
mysql_mutex_unlock(&thd->LOCK_thd_data);
continue;
}
mysql_mutex_unlock(&thd->LOCK_thd_data);
wsrep_reset_threadvars(rollbacker);
wsrep_store_threadvars(thd);
thd->wsrep_cs().acquire_ownership();
thd_proc_info(rollbacker, "wsrep aborter active");
/* Rollback methods below may free thd pointer. Do not try
to access it after method returns. */
if (wsrep_thd_is_applying(thd))
{
wsrep_rollback_high_priority(thd, rollbacker);
}
else
{
wsrep_rollback_local(thd, rollbacker);
}
wsrep_store_threadvars(rollbacker);
thd_proc_info(rollbacker, "wsrep aborter idle");
}
delete wsrep_rollback_queue;
wsrep_rollback_queue= NULL;
WSREP_INFO("rollbacker thread exiting %llu", rollbacker->thread_id);
DBUG_ASSERT(rollbacker->killed != NOT_KILLED);
DBUG_PRINT("wsrep",("wsrep rollbacker thread exiting"));
DBUG_VOID_RETURN;
}
void wsrep_create_rollbacker()
{
DBUG_ASSERT(wsrep_cluster_address[0]);
Wsrep_thd_args* args(new Wsrep_thd_args(wsrep_rollback_process,
WSREP_ROLLBACKER_THREAD,
pthread_self()));
/* create rollbacker */
if (create_wsrep_THD(args, false))
WSREP_WARN("Can't create thread to manage wsrep rollback");
}
/*
Start async rollback process
Asserts thd->LOCK_thd_data ownership
*/
void wsrep_fire_rollbacker(THD *thd)
{
DBUG_ASSERT(thd->wsrep_trx().state() == wsrep::transaction::s_aborting);
DBUG_PRINT("wsrep",("enqueuing trx abort for %llu", thd->thread_id));
WSREP_DEBUG("enqueuing trx abort for (%llu)", thd->thread_id);
if (wsrep_rollback_queue->push_back(thd))
{
WSREP_WARN("duplicate thd %llu for rollbacker",
thd->thread_id);
}
}
int wsrep_abort_thd(THD *bf_thd_ptr, THD *victim_thd_ptr, my_bool signal)
{
DBUG_ENTER("wsrep_abort_thd");
THD *victim_thd= (THD *) victim_thd_ptr;
THD *bf_thd= (THD *) bf_thd_ptr;
mysql_mutex_lock(&victim_thd->LOCK_thd_data);
/* Note that when you use RSU node is desynced from cluster, thus WSREP(thd)
might not be true.
*/
if ((WSREP(bf_thd) ||
((WSREP_ON || bf_thd->variables.wsrep_OSU_method == WSREP_OSU_RSU) &&
wsrep_thd_is_toi(bf_thd))) &&
victim_thd &&
!wsrep_thd_is_aborting(victim_thd))
{
WSREP_DEBUG("wsrep_abort_thd, by: %llu, victim: %llu", (bf_thd) ?
(long long)bf_thd->real_id : 0, (long long)victim_thd->real_id);
mysql_mutex_unlock(&victim_thd->LOCK_thd_data);
ha_abort_transaction(bf_thd, victim_thd, signal);
mysql_mutex_lock(&victim_thd->LOCK_thd_data);
}
else
{
WSREP_DEBUG("wsrep_abort_thd not effective: %p %p", bf_thd, victim_thd);
}
mysql_mutex_unlock(&victim_thd->LOCK_thd_data);
DBUG_RETURN(1);
}
bool wsrep_bf_abort(THD* bf_thd, THD* victim_thd)
{
WSREP_LOG_THD(bf_thd, "BF aborter before");
WSREP_LOG_THD(victim_thd, "victim before");
DBUG_EXECUTE_IF("sync.wsrep_bf_abort",
{
const char act[]=
"now "
"SIGNAL sync.wsrep_bf_abort_reached "
"WAIT_FOR signal.wsrep_bf_abort";
DBUG_ASSERT(!debug_sync_set_action(bf_thd,
STRING_WITH_LEN(act)));
};);
if (WSREP(victim_thd) && !victim_thd->wsrep_trx().active())
{
WSREP_DEBUG("wsrep_bf_abort, BF abort for non active transaction."
" Victim state %s bf state %s",
wsrep::to_c_string(victim_thd->wsrep_trx().state()),
wsrep::to_c_string(bf_thd->wsrep_trx().state()));
switch (victim_thd->wsrep_trx().state()) {
case wsrep::transaction::s_aborting: /* fall through */
case wsrep::transaction::s_aborted:
WSREP_DEBUG("victim is aborting or has aborted");
break;
default: break;
}
/* victim may not have started transaction yet in wsrep context, but it may
have acquired MDL locks (due to DDL execution), and this has caused BF conflict.
such case does not require aborting in wsrep or replication provider state.
*/
return false;
}
bool ret;
wsrep::seqno bf_seqno(bf_thd->wsrep_trx().ws_meta().seqno());
if (wsrep_thd_is_toi(bf_thd))
{
ret= victim_thd->wsrep_cs().total_order_bf_abort(bf_seqno);
}
else
{
DBUG_ASSERT(WSREP(victim_thd) ? victim_thd->wsrep_trx().active() : 1);
ret= victim_thd->wsrep_cs().bf_abort(bf_seqno);
}
if (ret)
{
wsrep_bf_aborts_counter++;
}
return ret;
}
int wsrep_create_threadvars()
{
int ret= 0;
if (thread_handling == SCHEDULER_TYPES_COUNT)
{
/* Caller should have called wsrep_reset_threadvars() before this
method. */
DBUG_ASSERT(!pthread_getspecific(THR_KEY_mysys));
pthread_setspecific(THR_KEY_mysys, 0);
ret= my_thread_init();
}
return ret;
}
void wsrep_delete_threadvars()
{
if (thread_handling == SCHEDULER_TYPES_COUNT)
{
/* The caller should have called wsrep_store_threadvars() before
this method. */
DBUG_ASSERT(pthread_getspecific(THR_KEY_mysys));
/* Reset psi state to avoid deallocating applier thread
psi_thread. */
#ifdef HAVE_PSI_INTERFACE
PSI_thread *psi_thread= PSI_CALL_get_thread();
if (PSI_server)
{
PSI_server->set_thread(0);
}
#endif /* HAVE_PSI_INTERFACE */
my_thread_end();
PSI_CALL_set_thread(psi_thread);
pthread_setspecific(THR_KEY_mysys, 0);
}
}
void wsrep_assign_from_threadvars(THD *thd)
{
if (thread_handling == SCHEDULER_TYPES_COUNT)
{
st_my_thread_var *mysys_var= (st_my_thread_var *)pthread_getspecific(THR_KEY_mysys);
DBUG_ASSERT(mysys_var);
thd->set_mysys_var(mysys_var);
}
}
Wsrep_threadvars wsrep_save_threadvars()
{
return Wsrep_threadvars{
current_thd,
(st_my_thread_var*) pthread_getspecific(THR_KEY_mysys)
};
}
void wsrep_restore_threadvars(const Wsrep_threadvars& globals)
{
set_current_thd(globals.cur_thd);
pthread_setspecific(THR_KEY_mysys, globals.mysys_var);
}
void wsrep_store_threadvars(THD *thd)
{
if (thread_handling == SCHEDULER_TYPES_COUNT)
{
pthread_setspecific(THR_KEY_mysys, thd->mysys_var);
}
thd->store_globals();
}
void wsrep_reset_threadvars(THD *thd)
{
if (thread_handling == SCHEDULER_TYPES_COUNT)
{
pthread_setspecific(THR_KEY_mysys, 0);
}
else
{
thd->reset_globals();
}
}