"Erik Moeller" erik_moeller@gmx.de schrieb:
The MediaWiki FT search has locking issues. It's very fast for a normal wiki, but for a huge one like Wikipedia it tends to get in a deadlock state. That's why we have to disable it on Wikipedia. Some other queries could also be made faster. Many of our queries have been optimized now and every single one should eventually return results with a response time of milliseconds.
Does this mean that we will NOT get full text search back soon? That would be a shame. If it is indeed the case that with another DB server we will still have locking problems with searches, we might want another architecture, with a dedicated search (plus several other functions, such as special pages and SQL queries, perhaps even Whatlinkshere) machine, which gets its database updated only once per second/minute/hour/whatever rather than in real time.
Andre Engels
Andre Engels wrote:
Does this mean that we will NOT get full text search back soon? That would be a shame. If it is indeed the case that with another DB server we will still have locking problems with searches, we might want another architecture, with a dedicated search (plus several other functions, such as special pages and SQL queries, perhaps even Whatlinkshere) machine, which gets its database updated only once per second/minute/hour/whatever rather than in real time.
I'm very interested in this, too.
If for whatever reason it seems mysql's built-in fulltext search is not going to work for us, we could pretty easily come up with some kind of alternative, I think. We could parse the database daily and generate a very fast btree dbm file, for example.
--Jimbo
On Feb 27, 2004, at 13:51, Andre Engels wrote:
Does this mean that we will NOT get full text search back soon?
No, it means we'll get it back when we have Geoffrin back online, on which fulltext search ran just fine. Geoffrin, running on 64-bit Linux with 4GB of RAM available to MySQL, was able to deal with the large search indexes in memory. Our current database server is on a 32-bit kernel and can only assign 2GB of RAM to the database. Less is available for the (large!) search indexes, and it gets very slow dealing with them.
Penguin's fixing it, and if they actually do it right this time it should be back online within a couple of weeks, at which time we can transition over a copy of the database that's being separately synched up now and will be ready with minimal downtime.
-- brion vibber (brion @ pobox.com)
wikitech-l@lists.wikimedia.org