Details
Blocker
Description
Our MariaDB server is crashing on long semaphor wait. It is happening usually at night when database backups are running (but sometimes also in the middle of the day). I'm 100% sure it's related to one of our users because when i moved him to different server, the problem moves with him. There were enough of memory and we had the same problem with MySQL 5.5. Everytime it happend, the mentioned user is running simultanious SELECT and TRUNCATE on the same table.
Gliffy Diagrams
Attachments
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
Jan,
there is attached file 'global_variables.txt' with all global variables from that server - you are probably asking for this:
performance_schema OFF
Maybe that user is enabling it locally (if it's possible)?
It seems that some of the table statistics are calculated even in the case when performance_schema is off. Especially in case when table becomes empty (truncate table or delete from <table>).
Could you please try MariaDB 5.5.34 with startup variable --innodb_use_stacktrace and if there is again long semaphore wait errors, I would need full error log. There should be additional information that would greatly help to identify the actual issue.
Fixed some incorrect and not logical mutex operations and added a new configuration variable innobase_use_stacktrace (default off) to produce a stack trace from waiting and holding thread if long semaphore wait is detected. These, should help to resolve possible future issues more easily.
Hi,
There is no truncate in the innodb status output in the attached log, so I assume you see it somewhere else. Do you have the general log enabled?
Which table is it? Can you provide the structure of the table (SHOW CREATE TABLE) and either the data dump or, if you cannot do it on security reasons, at least estimates on the data contents (number of rows, SHOW INDEX, SHOW TABLE STATUS and such)?
Thanks.