cent os 6.6 , mariadb 10.0.22

jplindst, could you please take a look?

Hi, Jan Lindström
At January 4th， slave of use fusion-io device automatic restart.
At 20:46 today , the slave again automatic restart.
Before the restart , we add "innodb_use_stacktrace = 1" in my.cnf file.

attach error log file: log.err.160117

after we set innodb_adaptive_hash_index = 0 ,innodb_change_buffering = none,
master and slave's mysql service not occur automatic restart.
But, at March 13th , Last day, master server's mysql service was stacked, not response .
after execute pstack command at below , mysql resume normal work.
pstack `pidof mysqld` > /tmp/pstack.log

attach pstack log file: pstack0313.log

It contains __tz_convert function in log file,
0x00000038fb29da67 in __tz_convert () from /lib64/libc.so.6

someone said , tz_function has performance problem , must modify time_zone ,
example , we live in east 8 zone , set global time_zone = '+8:00' replace default time_zone='SYSTEM', This is Correctly ? please help me , Thanks.

I do not think this is related to time zones. Both error logs contain long semaphore wait error messages, not sure why. Did you do some long running transactions ?

I sure not long running transaction . we main run two type statement:
(1) singe value statement :INSERT INTO table01 VALUES ON DUPLICATE KEY UPDATE ...
(2) single value call procedure: call update_log(v1,v2,v3);

besides, fusion-io device must format 4k block ?
This link said at below:* Low level formatting device with 4k page size will reduce memory usage of driver and improve performance*
http://www.voleg.info/fusion_io-mezzonine-card-UCS-redhat6.html

Hi, yes it could be that fusion-io device provides better performance and reduced memory usage formatted as 4K blocks, but then your problem is not about that. Do you use xtradb (default) or innodb_plugin ? I recommend using innodb_plugin when fusion-io is used, gives you better performance.

Thank you .we using default xtradb engine.
If swithch to innodb_plugin, add two lines in [mysqld] of my.cnf ?
ignore_builtin_innodb
plugin_load=innodb=ha_innodb.so

last month ,after we set innodb_adaptive_hash_index = 0 ,innodb_change_buffering = none,
master and slave's mysql service not occur automatic restart. And error logs not contain long semaphore wait too.

now , Please you help analyze attach file pstack0313.log ,thanks you again

Hi, yes correct. pstack0313.log seems to contain a lot of thread waiting either new connections or doing something on audit_plugin. I do not know much about audit_plugin so not sure what it is waiting there.

Thank you .
Recently ,We test MairaDB10.1,Percona xtradb cluster 5.6, found write stop question, google groups link:
https://groups.google.com/forum/#!topic/codership-team/Ne6WsTWixH8

Last ,We contact other company's DBA , found it is CentOS 6.6 bug.
receference link:
https://www.infoq.com/news/2015/05/redhat-futex
https://groups.google.com/forum/#!msg/mechanical-sympathy/QbmpZxp6C64/0M4_EbzSLj4J

after we replace to CentOS 6.5, mariadb 10.0, mariadb 10.1 (galera cluster) run normal

Both the bug and Mdev-8098 can closed , thanks you again