#495 Racing Inside ActiveSupport::Dependencies - Ruby on Rails

Type	To find
responsible:me	tickets assigned to you
tagged:"@high"	tickets tagged @high
milestone:next	tickets in the upcoming milestone
state:invalid	tickets with the state invalid
created:"last week"	tickets created last week
sort:number, importance, updated	tickets sorted by #, importance or updated
Combine keywords for powerful searching.
Use advanced searching »

This project is archived and is in readonly mode.

#495 ✓stale

Racing Inside ActiveSupport::Dependencies

Reported by skibumsworld | June 27th, 2008 @ 02:11 AM | in 2.1.3

I'm working on a project that uses mongrel handlers to service requests. We use ActiveRecord but not ActionController. We've been seeing weird errors which look like:

#<LoadError: Expected ./my_model.rb to define MyModel>

the file does define the class but an exception is thrown. After some digging we found that the problem is in dependencies.rb in the require_or_load method.

If a thread is interrupted right after:

loaded << expanded

then any other threads that attempt to load the same file will get the error above, because the file hasn't been loaded yet even though it's name already appears in the loaded list.

I've included a test to reproduce this. The test only works against the development version of rails. If you want you can run it against other version but you will need to change the way Dependencies module is declared since it didn't used to be inside the ActiveSupport module.

I'm working on a patch right now, but any suggestions on how to fix this issue are more then welcome.

One last thing. When running the test please don't forget to change the first require to point at the correct location for activerecord.

Comments and changes to this ticket

You flagged this item as spam.
skibumsworld June 27th, 2008 @ 03:43 AM
Same test, modified to run with activesupport 2.0.2
- dependencies_test_rails2.0.2.zip 1.2 KB
Jeremy Kemper June 27th, 2008 @ 08:53 AM
- State changed from “new” to “open”
- Milestone set to “2.1.1”
- Assigned user set to “Jeremy Kemper”
You flagged this item as spam.
DrMark June 27th, 2008 @ 09:07 AM
Jeremy,
While you are working in there, would you fix the typo in will_unload?. It is currently:
def will_unload?(const_desc)
autoloaded?(desc) ||
explicitly_unloadable_constants.include?(to_constant_name(const_desc))
end
but should be:
def will_unload?(const_desc)
autoloaded?(const_desc) ||
explicitly_unloadable_constants.include?(to_constant_name(const_desc))
end
Note the autoloaded line. I will make a separate ticket if necessary.
Thanks!
DrMark
You flagged this item as spam.
skibumsworld June 27th, 2008 @ 07:29 PM
Here's a patch that fixes the problem.
- dependencies.patch 2.5 KB
Jeremy Kemper June 28th, 2008 @ 01:15 AM
- Tag changed from “activesupport, dependencies” to “activesupport, dependencies, patch”
skibumsworld, could you use a Mutex instead? Thread.critical has issues on jruby so other synchronization primitives are preferred.
Jeremy Kemper June 28th, 2008 @ 01:16 AM
DrMark, will_unload? appears to be completely unused. Do you use it?
You flagged this item as spam.
DrMark June 28th, 2008 @ 03:52 AM
Hi Jeremy,

Yes. The benchmarkforrails plugin uses it (among others). I thought that it was odd that a bug that obvious could get through :)

If you could fix it or remove it that would be great. That would save me having to monkey patch it all the time :)

Thanks!
Jeremy Kemper June 28th, 2008 @ 05:09 AM
Quite so, DrMark. I fixed it in master, 2-1-stable, and 2-0-stable. How odd.
Jeremy Kemper July 30th, 2008 @ 10:06 AM
- Assigned user changed from “Jeremy Kemper” to “josh”
- State changed from “open” to “stale”
- Milestone cleared.
Jeremy Kemper October 24th, 2008 @ 03:19 AM
- Milestone set to “2.1.3”
- State changed from “stale” to “open”
- Assigned user changed from “josh” to “Jeremy Kemper”
You flagged this item as spam.
Stephen Bannasch December 27th, 2008 @ 04:23 PM
Here are links to two related bugs in ruby's implementation of require when used with threads:

require is not thread-safe http://redmine.ruby-lang.org/iss...

autoload is not thread-safe http://redmine.ruby-lang.org/iss...

Basically there does not appear to be a safe way for require to work in a multi-threaded application unless the first use of require locks until the require is complete.
Pratik March 8th, 2009 @ 02:07 PM
- State changed from “open” to “stale”
Putting the ticket in stale. Please reopen if this is still an issue.

Thanks.