Panasonic Youth rob sanheim writes about software, business, ruby, music, stuff and things

RailsConf 2007 - Memcaching Rails - Chris Wansrath

• Chris Wanstrath helped scaling gamespot
• had to turn off gzipping to avoid burning the cpus
• 3000 req/seconds
• 50M pages in the day, no downtime
• Chris is a sloppy programmer (jk)

• a distributed hash with no each or keys – just get and set
• written in c – fast

• at first – you dont need memcached – YAGNI
• unless you really do need it - you have a millions of views, you have a huge DB, you have really ugly sql

• the standard pattern—try the cache first, if its not there, grab it from the DB and set it in the cache
• use memcached-client, not ruby-memcache
• cache-fu = acts_as-cache 2.0
• don’t use DNS names for the servers – always use ip’s
• expire_cache – issues a delete to the cache
• get_cache – duh, grab from the cache
• so after_save :expire_cache

• you dont ever cache nil – since that screws ruby. cache_fu will cache false behind the scenes
• get the DB down to 0.0%
• Model.cached(:finder)—wraps the idiom of caching a custom finder

• can do Model.get_cache(1,2,3) to do a get_multi call to memcached ( ie get many objects in parallel)
• can do a before_filter to force all cache misses when a url gets a magic query string
• but be careful – don’t crash your site
• use reset_cache!

• need :version to invalidate old version of AR models—otherwise you get screwed when you do a migration
• libketama – developed by lastfm guys so you can throw memcache servers in and out w/o invalidating all your keys
• the perl client already handles this under the covers, the ruby client doesn’t
• someone should implement libketama in ruby, to make big memcached installations safer