Problem with Memcache / forking

[jerrett]

http://www.modrails.com/documentati...xample_1_memcached_connection_sharing_harmful


I'm running into a similar problem with Litespeed... Memcache is returning incorrect objects, sometimes.

Any ideas of a solution? How would one go about reseting the connection to memcache each time it's forked, if this is the problem?

 

[jerrett]

This also just started happening with the new Litespeed and Rails 2.3 (Rack)..

 

[mistwang]

Yes, need to reset the connection to memcache. Just close all connections after the initialization.

 

[jerrett]

Would it be a good place to do this in environment.rb, or is that not going to get executed when it's forked?

 

[mistwang]

It should be fine. environment.rb only get loaded once I think.

 

[jerrett]

Doesn't seem to help, if it's only loaded once i'm assuming that means it won't reset when the process gets forked.. where would we hook into that?

 

[mistwang]

maybe it is too early as the connection to memcache was established after loading environment.rb.

Please try to patch RailsRunner.rb, at where "ActiveRecord::Base.clear_active_connections!" is called.

 

[jerrett]

It worked, but we had to manually create a cache object for the session store so we could call reset on it (and then just SESSION_CACHE.reset right before the clear_active_connections!)

In the edge of rails it looks like there is a way to access the Session memcache instance, which would mean it could in the future be fixed in RailsRunner.rb by default.. something to look forward to perhaps

 

[mistwang]

Yes, please keep us posted when it is available, so we can add som generic code like "clear_active_connections!" to address it in RailsRunner.rb .