Description of problem: Two of our customers have high-traffic websites backed by MySQL 4.1 (RHN packages) on RHEL 4 (i386). One is using PHP, the other Perl. Their database schemas are completely different. Both were using MyISAM tables exclusively. In the past couple of months they both started having their sites go down under heavy load. Every time it happened, mysqld was still running, but not responsive. Restarting mysqld restored access, but sometimes only minutes later it would hang again. Then sometimes it would go a few days without hanging. We never saw anything in the logs. In both cases, we had them upgrade to MySQL 5.0 from MySQL AB's RPM packages, and the problems immediately stopped. We've used MySQL 4.1 from MySQL AB's own RPM packages on RHEL 3, and have never had any problem with it. Version-Release number of selected component (if applicable): mysql-server-4.1.20-1.RHEL4.1, and I think also mysql-server-4.1.12-3.RHEL4.1 How reproducible: Not easily. Kind of a Heisenbug. Sorry I don't have anything more concrete than this, but I thought it'd be best to get a bug in here in case anyone else runs into the same problem, so we can try to narrow it down.
Are you sure you've seen this since updating to 4.1.20? It sounds a bit similar to some other reports against 4.1.12, eg bug #183277.
You may be right. We solved the problem the first time several weeks before 4.1.20 was built (according to package date 2006-06-02). The second time was only a few days ago, but it's quite possible they weren't using 4.1.20 yet. Our problems didn't manifest themselves quite the same way as in bug #183277, but it's close enough that this should probably be marked as a duplicate. Bugzilla's not letting me do that, though ... Thanks.
It's premature to mark as dup if we're not sure. I'll leave this open for the moment. Please run 4.1.20 for awhile and update this bug once you know if the problem's still there.
Since I've not heard anything, I'm going to close this as a duplicate of bug #183277. Please re-open if you think it's not. *** This bug has been marked as a duplicate of 183277 ***