Laurynas Biveinis added a comment - 2017-01-30 06:37 This is caused by incorrect testcase sync: the testcase disconnects, waits for the threads disappear from the processlist, then checks the status variable. But a disconnecting thread first removes itself from the processlist, and becomes cached only later, creating a race condition. A proper fix appears to drop processlist checks and SHOW STATUS, and wait for Threads_cached to become desired value by wait_for_status_var.inc, see https://github.com/percona/percona-server/pull/1337

Elena Stepanova added a comment - 2017-05-24 15:46 Recent occurrence in 10.1: http://buildbot.askmonty.org/buildbot/builders/kvm-deb-wheezy-x86/builds/4633/steps/mtr/logs/stdio sys_vars.thread_cache_size_func w1 [ fail ] Test ended at 2017-05-23 05:28:55   CURRENT_TEST: sys_vars.thread_cache_size_func --- /usr/share/mysql/mysql-test/suite/sys_vars/r/thread_cache_size_func.result 2017-05-23 03:48:45.000000000 -0400 +++ /run/shm/var/1/log/thread_cache_size_func.reject 2017-05-23 05:28:55.080905256 -0400 @@ -18,7 +18,7 @@ ** Disconnecting conn1,conn2,conn3,conn4 ** SHOW STATUS LIKE 'Threads_cached'; Variable_name Value -Threads_cached 3 +Threads_cached 0 3 Expected SET @@GLOBAL.thread_cache_size= 1; ** Connecting conn1 using username 'root' **   mysqltest: Result content mismatch