Thanks, the patches in that bug looks like it can speed up the gathering of statistics. But our problem is during normal operation, when there are a large amount of concurrent users.
Do any of you have experience with using flush_log_at_trx_commit 2 on the master mysql server? Our sysadmin is reluctant, because he says it will compromise the reliability of the replication (we use master-slave replication). But I would gladly trade the risk of 1 second of transaction loss for a usable site.
I notice moodle.org uses flush_log_at_trx_commit=2. Do they use replication for high availability?