Hello everyone!
we recently upgraded from 2.23.27 to 2.23.39, and in the same time changed machine
from a p3-666, 512MB, ide to a dual p3 550, 896MB, scsi. The platform is Linux. As
expected
we got a enourmous increase in database speed. Then we encountered a strange behaivour:
Suddenly MySQL starts locking a lot of threads, with only 3-4 running, causing all new
querries
to also get locked, and when sooner or later it's done with the non-locked querries it
starts all locked
ones, and this slows the server down enourmously. Normal load average is around
0.5->3.0 and
when this happens it bumps up to 50 or more. Then it takes a minute or two for the
server to
get back to normal load levels.. This happens a few times per hour. The problem can
be reduced
in recovery time by setting the "max_connections" lower, around 30 is the lowest still
reasonably
functional, as some website users will get an error message once in a while instead.
Is there anyone with this problem, or similar, with an idea of how to stop this from
happening?
Why is this happening now and not before?
Eearlier we thought that it would be one special querry that took a lot more time than
others that
was causing this, but when we get rid of that querry we got the "non-locked" processes
more randomly
distributed over our connections...
regards, Johan Wahlström
