I have converted the following tables on the English wiki to InnoDB:
cur
links
brokenlinks
recentchanges
user
watchlist
archive
The new 'searchindex' table needs to stay as MyISAM for the fulltext
index support.
I have not yet converted:
old
...images, ipblocks, math yadda yadda...
Most of them because I'm lazy and they're low-profile. I didn't do
'old' yet because it's, well, FRIGGING HUGE and will take some time to
convert. And, it's read relatively infrequently compared to cur & co, so
again it's lower priority. I'll convert it later if things are moving
well.
My patented 'load a big special page and then tail -f the logs to see if
other people are still able to access stuff' preliminary benchmark seems
positive so far.
Eloquence pointed out a problem with the page move code where I had
forgot to merge in the changes relating to search index fields. This is
fixed, and I don't think there are any other such remaining, though
there might still be some rare cases where an operation that should
update a search index doesn't.
At some point we'll want to think about explicit support for
transactions, so we don't have to worry about broken operations, say,
renaming an article's current revision but not it's old revisions or
updating link tables.
I'll merge the code into CVS in the morning after I'm awake and my brain
works again, but there's nothing terribly new to it; just some merging
in from the older innodb branch that's sitting in there somewhere
already.
-- brion vibber (brion @
pobox.com)