27 Apr
2013
27 Apr
'13
2:09 p.m.
In the absence of any other solution I have set innodb_flush_log_at_trx_commit=0. This stops mysql issuing an fsync() after each write and instead issues an fsync() after 1 second (give or take a few ticks). It means I could lose up to slightly more than 1 second of transactions in the event of a crash or hardware failure, but the nature of the application in question (backup software) makes that a non-issue as the data is record of another process (the backup) that has now failed and will need to be restarted anyway. Having just restarted the backup jobs, performance appears to be back to the expected level based on 5 minutes of running. I still don't know what changed to make this problem appear though. James