>>>>> "David" == David Simmons <simmons@stripped> writes:
David> I'm running MySQL on a Linux 2.0.36 server (RedHat 5.2) with an
David> extremely high load (and the MySQL server has a rather high
David> load as well), and occasionally MySQL will freeze and its
David> UNIX-domain socket will be removed, preventing clients from
David> Sometimes the stuck mysqld can be forcefully terminated ungracefully
David> with a "kill -9", other times a "kill -9" will not even dislodge the
David> process and the server must be rebooted. I'm using a RPM'ed packaged
David> version of MySQL that was statically linked with glibc-2.0.7.
David> Has anyone experienced this condition before? Could this
David> possibly be a problem related to MySQL's use of the Linux
When the above happens, do you get anything in the 'hostname.err' file
Which MySQL version are you using?
Any way you can simulate this with some test script (like fork_test.pl)
Could this problem be related to some kernel resources like available
memory or available file descriptors?
The Unix domain socket should only be removed by mysqld on a
shutdown. Are you sure you don't have any process that removes the
socket file (like a cron job)?
If kill -9 doesn't work, then you have found a kernel bug; Maybe it's
time to upgrade your kernel...)
Note that you should normally NOT use kill -9; You should first try a
simple kill and only when this doesn't work try kill -9. After a kill
-9, you MUST check all tables with isamchk before restarting mysqld!