> One of our servers which is configured primarily for InnoDB has crashed
> in the same fashion three times in the past 48 hours. I've been going
> over the error log, and I would like to know if anyone else has seen
> similar errors, or can give me some pointers to work out this problem.
> (truncated message)
After googling and searching the mysql list archive and forums, and
finding other folks who ran into the "semaphore wait has lasted > 600
seconds" error, I saw no consistency between what caused that error for
the other people and for my server. I started checking the error log
very closely, and came upon something that seems very odd to me; I am
hoping someone else can explain this to me (maybe I'm making a silly
I cut out the very last dump of InnoDB Monitor Output in the server.err
file (just to make it easier to read and grep through). I pulled out all
the transactions that have a "lock struct", and examined each. They are
ALL inserts (large ones), but there are two things which seem wrong to
me: there are 9 transactions with lock structs listed, even though there
are only "8 queries inside InnoDB"; two of the locks are on the same table.
If the same table was locked by two threads at the same time, that would
explain the crash (wouldn't it?), but how is that possible? I thought
the express purpose of semaphores was to prevent that.
I'm hoping someone else can shed some light on this :) Thanks!