wsrep-slave-threads is counted towards max_connections

[Raghavendra D Prabhu]

Originally reported here, https://mariadb.atlassian.net/browse/MDEV-6206

Easy to reproduce.

 /pxc/bin/mysqld --defaults-file=/pxc/etc/my.cnf.local --basedir=/pxc --user=mysql --wsrep-debug --wsrep-new-cluster --wsrep-start-position='2d2c14c1-cdcd-11e3-9eff-3eb6075d3f68:0' --max-connections=50 --wsrep-slave-threads=200

starting new mysql client gives "ERROR 1040 (HY000): Too many connections"

[Raghavendra D Prabhu (raghavendra-prabhu)]

This is because of:

  mysql_mutex_lock(&LOCK_connection_count);
  ++connection_count;
  mysql_mutex_unlock(&LOCK_connection_count);

in start_wsrep_THD.

and subsequent check:

  if (*thd->scheduler->connection_count >=
      *thd->scheduler->max_connections + 1 || abort_loop)

in create_new_thread.

[Raghavendra D Prabhu (raghavendra-prabhu)]

Since start_wsrep_THD is also used for creating appliers (the slave threads), the fix here would be to not increment that for appliers: wsrep_replication_process(THD*) and also wsrep_rollback_process.

[Philip Stoev (philip-stoev-f)]

MTR test case. Please push when fixing the bug

#
# LP:1315588 wsrep-slave-threads is counted towards max_connections
#

--source include/galera_cluster.inc
--source include/have_innodb.inc

--let $max_connections_orig = `SELECT @@max_connections`
--let $wsrep_slave_threads_orig = `SELECT @@wsrep_slave_threads`

SET GLOBAL max_connections = 5;
SET GLOBAL wsrep_slave_threads = 10;

SELECT COUNT(*) > 5 FROM INFORMATION_SCHEMA.PROCESSLIST;

--let $galera_connection_name = node_1a
--let $galera_server_number = 1
--source include/galera_connect.inc

--connection node_1a
SELECT 1 FROM DUAL;

--eval SET GLOBAL max_connections = $max_connections_orig
--eval SET GLOBAL wsrep_slave_threads = $wsrep_slave_threads_orig

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PXC-1103