wsrep aborter thread does not show in processlist

[Seppo Jaakola]

current 0.7 branch head version gives strange process list:

[galera@haageman ~]$ processlist $cats
Processlist for node: abyssinian
+------+-------------+------------------+------+---------+------+--------------------+------------------+
| Id | User | Host | db | Command | Time | State | Info |
+------+-------------+------------------+------+---------+------+--------------------+------------------+
| 1 | system user | | NULL | Sleep | | NULL | NULL |
| 2 | system user | | NULL | Sleep | 405 | wsrep aborter idle | NULL |
| 1120 | root | 10.0.0.101:60102 | NULL | Query | 0 | NULL | show processlist |
+------+-------------+------------------+------+---------+------+--------------------+------------------+

Processlist for node: bengal
+------+-------------+------------------+------+---------+------+--------------------+------------------+
| Id | User | Host | db | Command | Time | State | Info |
+------+-------------+------------------+------+---------+------+--------------------+------------------+
| 1 | system user | | test | Sleep | | NULL | NULL |
| 2 | system user | | NULL | Sleep | 401 | wsrep aborter idle | NULL |
| 1186 | root | 10.0.0.101:60304 | NULL | Query | 0 | NULL | show processlist |
+------+-------------+------------------+------+---------+------+--------------------+------------------+

Processlist for node: carthusian
+-----+------+------------------+------+---------+------+-------+------------------+
| Id | User | Host | db | Command | Time | State | Info |
+-----+------+------------------+------+---------+------+-------+------------------+
| 573 | root | 10.0.0.101:33689 | NULL | Query | 0 | NULL | show processlist |
+-----+------+------------------+------+---------+------+-------+------------------+

wsrep aborter thread is missing in the process list in node carthusian. Checking by gdb, shows that aborter thread is alive and working well.
Also whole cluster is healthy and works good under sqlgen test load-

This situation is right after starting the cluster

[Seppo Jaakola (seppo-jaakola)]

This seems to happen only when gratstate file is missing

[Seppo Jaakola (seppo-jaakola)]

Reason for this appeared to be a bug in mysqld.cc:wsrep_close_client_connections()
Scanning through 'threads' list also emptied the list, which caused wsrep system threads to
disappear from the threads list.

This caused issues also for shutdown phase, where threads list is used to close all connections gracefully.