User:Brooke Vibber/MySQL UC roundup

From Meta, a Wikimedia project coordination wiki

Monday[edit]

The MySQL support crew are a bunch of crazies; Domas fits right in. ;)

Tuesday[edit]

MySQL 5.1[edit]

MySQL 5.1 features that caught my eye at the presentation:

  • processlist available through information_schema, can SELECT on it (YAY!!!!!)
  • "faster alter table" making at least some operations such as dropping indexes and changing enums go much much faster
  • changes to fulltext search: plugins etc. keep an eye on this for effect on out-of-the-box MediaWiki search
  • partitioning, federated table engine: such things may make it easier to separate out 'archival' data from the main hot data sets while keeping it conveniently online when needed

PHP[edit]

Rasmus again advocates using require() instead of require_once() if you're using APC to save a few syscalls. Would be worth doing some tests of cleaning up all the includes and such to see how much we can trim.

Apparently you can shave a few more syscalls off by using full paths in all includes and turning off the option to check if files have changed. ;)

He also pushed Yahoo!'s open-source javascript widget libraries. AJAX-y people should take a look at such things to see if there's anything useful, or at least ideas on packaging and internal API.

MW users[edit]

Spoke to a nice fellow from a three-letter CPU manufacturer we all love, using MediaWiki for some internal documentation. The company is apparently way into Microsoft Sharepoint for internal document sharing, and they'd love better integration with it (UI, logins, etc).

Search[edit]

Sphinx in action

Spoke to some nice Russian fellows pushing the Sphinx search engine:

  • GPL
  • C++
  • Had a demo up comparing search on a gig's worth of Wikipedia text between Sphinx, Lucene, and MyISAM fulltext search; claiming 10x faster than Lucene and better relevancy ordering of results
  • They would like help adding support for eg Chinese and Japanese


Backup tools[edit]

Zmanda interested in making some sort of convenient automated 'back up your wiki' tool (database plus files). Might be nice to point people at when they ask "how do I back up my wiki???"

Sessions in Cluster[edit]

Was a talk on using MySQL Cluster as storage for PHP sessions. The talk was 90% interminable detail about the concept of session state in web apps and details of PHP's implementation, but the idea was interesting.

Might be worth trying out: compare performance and loss rate between our current memcached session store and MySQL Cluster.

Wednesday[edit]

Domas and I (if he leaves me room) are presenting, then we have a BoF in the evening.

La Tribune will run an article on Wikipedia in a couple days, got cornered for an interview.

Had a long chat with Rich Morrin:

  • some notes about Doxygen features
  • ideas for inline-generated interactive SVG/JS graph magic
  • chatted a bit about potential of getlets

BoF:

  • BoF sessions were hidden, hard to find, but we had a decent crowd shuffle in
  • Bullied Monty into promising to add 4-byte UTF-8 support as an option for MySQL charsets (whee)

Talked with Robla:

  • ManyOne's customizations to MediaWiki will be open sourced Real Soon Now; includes an approval/publishing interface which may be worth looking at as we prepare review features for Wikipedia
  • Thinking about commercial support options we might do for MediaWiki. Is this something Wikimedia wants to get involved directly in?

Thursday[edit]

  • Mitch Kapor did a keynote where he talked about Wikipedia and such, but I missed it as it was earlier in the morning than I planned for. Domas says he dissed MediaWiki though. ;)
  • ...

Friday[edit]

Visiting Google.