On sab, 2003-02-08 at 17:25, Ian Gilfillan wrote:
Thanks for the slow query log - I don't have the
db in front of me now, but
I'll take a look later.
Great, thanks for your help!
It may be worthwhile changing user_newtalk to InnoDb
to
see what happens. It's currently MyISAM, and if the talk pages are being
updated frequently table locking may be a problem here, so row level locking
may be more efficient.
It's updated very infrequently, and the slow query log lists the time
spent as being in the query, not in waiting for locks. That's why I'm so
stumped...
My recollection of looking at the SHOW PROCESS list during peak times is
that these queries were usually sitting in "Statistics" or "Opening
tables" state. (What _is_ "Statistics"? The online manual doesn't seem
to explain it.)
It occurs to me that our table cache is only 256; we've got a
max_connections of 250, and have a fair number of 2-table joins
(searching, checking status of linked pages on view, recentchanges and
watchlist) and, taking all wikis into account, some 400+ tables.
Lesse...
| Open_tables | 256 |
| Opened_tables | 6671 |
Yikes! I'm going to up it to about 1000 and see what that does...
-- brion vibber (brion @
pobox.com)