>>>>> "Mauricio" == Mauricio Breternitz <mbjsql@stripped> writes:
Mauricio> Looks like we have been chasing similar issues. (I am trying to
Mauricio> get several MySQL servers in a a system that has some similarities with
Mauricio> yours, albeit in a different environment. The system has several processors
Mauricio> that can see a shared disk, but also shared memory (in addition to local
Mauricio> The symptoms that you describe appear to stem from caching.
Mauricio> I am studying the source at this moment, and would like to point you to
Mauricio> functions like flush_io_cache and flush_key_blocks. Perhaps you
Mauricio> could change the code to invoke those functions at the end of each
Mauricio> query before you release the file locks to make sure all changes are
Mauricio> reflected globally.
This is done by default.
Mauricio> By the way, did you only change the fcntl calls in sql/locks.cc ?
Mauricio> I am under the impression that one would have to change the locks in
Mauricio> thr_lock.cc also.
If run with --enable-locking (or not --skip-locking) then MySQL will
first do an external lock (flock) and then an internal lock (thr_lock)
for each table.
In Denis setup, he don't need to worry about thr_lock at all.
Mauricio> best regards