While looking into this issue, I managed to hit several crashes on both PS 5.1 and 5.5 trunks (5.1.73-14.12 revno 595 and 5.5.35-33.1 revno 607). The test was quite simple, fire up a server instance with the following options:
innodb_file_format=barracuda
innodb_file_per_table=1
innodb_dict_size_limit=0
Start RQG with the percona grammar in a loop against the server instance. In each case, this can run for hours without problem.
Then set innodb_dict_size_limit to just about any value, I used 1000, 10000, 100000 and 500000. The actual value really didn't seem to have any impact on the failure as long as it was > 0. Within minutes, both the 5.1 and 5.5 servers would crash.
A very cursory analysis of all of the core files revealed that there seems to be something 'unprotected' as all the issues seemed to related to using an already freed record, releasing a lock too many times, etc...
The next few comments will contain info on the various crashes.
While looking into this issue, I managed to hit several crashes on both PS 5.1 and 5.5 trunks (5.1.73-14.12 revno 595 and 5.5.35-33.1 revno 607). The test was quite simple, fire up a server instance with the following options: file_format= barracuda file_per_ table=1 dict_size_ limit=0
innodb_
innodb_
innodb_
Start RQG with the percona grammar in a loop against the server instance. In each case, this can run for hours without problem. dict_size_ limit to just about any value, I used 1000, 10000, 100000 and 500000. The actual value really didn't seem to have any impact on the failure as long as it was > 0. Within minutes, both the 5.1 and 5.5 servers would crash.
Then set innodb_
A very cursory analysis of all of the core files revealed that there seems to be something 'unprotected' as all the issues seemed to related to using an already freed record, releasing a lock too many times, etc...
The next few comments will contain info on the various crashes.