>>>>> "Jocelyn" == Jocelyn Fournier <joc@stripped> writes:
Jocelyn> It seems something else in the code "reset" the value of
Jocelyn> delayed_insert_threads to 0, which could explain the jump from 42 to 0, but
Jocelyn> I not yet found what :(
Just a note about this thread. I am just now making a merge of 4.0
and the 4.1 trees so that we can make the 4.1 tree publicly available.
As soon as this is done I will write extend some test cases regarding insert
delayed to try to find this problem.
Jocelyn> Anyway, I'm happy to hear I'm not the only one experiencing problem with
Jocelyn> delayed_threads under high QPS :)
Have you taken a look at the tests/test_delayed_insert.pl test we use
to test insert delayed ?
If you could someway modify this to also look up, it would be much
easier for us to debug this and find what is going on...
My guess is that somehow the insert thread handler gets stuck in some
of the functions handle_inserts() calls waiting for a mutex or table
lock to get freed.
(Maybe thr_upgrade_write_delay_lock() ?)
If you can find out which function, it would help us a lot in tracking this...
For technical support contracts, goto https://order.mysql.com/
__ ___ ___ ____ __
/ |/ /_ __/ __/ __ \/ / Mr. Michael Widenius <monty@stripped>
/ /|_/ / // /\ \/ /_/ / /__ MySQL AB, CTO
/_/ /_/\_, /___/\___\_\___/ Helsinki, Finland