I’ve got a REST webservice I’ve built with JRuby 1.7 p2, Sinatra and
Mizuno.
Intermittently, and only under some load, I see the following error from
it:
[2012-08-19 21:13:50] ERROR RuntimeError: can’t add a new key into hash
during iteration
org/jruby/RubyHash.java:905:in []=' /opt/jruby/active/lib/ruby/1.9/webrick/utils.rb:205:in register’
/opt/jruby/active/lib/ruby/1.9/webrick/utils.rb:161:in register' org/jruby/ext/thread/Mutex.java:149:in synchronize’
/opt/jruby/active/lib/ruby/1.9/webrick/utils.rb:160:in register' /opt/jruby/active/lib/ruby/1.9/webrick/utils.rb:232:in timeout’
/opt/jruby/active/lib/ruby/1.9/webrick/httprequest.rb:398:in _read_data' /opt/jruby/active/lib/ruby/1.9/webrick/httprequest.rb:409:in read_line’
/opt/jruby/active/lib/ruby/1.9/webrick/httprequest.rb:314:in read_header' /opt/jruby/active/lib/ruby/1.9/webrick/httprequest.rb:92:in parse’
/opt/jruby/active/lib/ruby/1.9/webrick/httpserver.rb:81:in run' /opt/jruby/active/lib/ruby/1.9/webrick/server.rb:191:in start_thread’
NoMethodError: undefined method `<’ for nil:NilClass
initialize at /opt/jruby/active/lib/ruby/1.9/webrick/utils.rb:181
each at org/jruby/RubyArray.java:1612
initialize at /opt/jruby/active/lib/ruby/1.9/webrick/utils.rb:179
each at org/jruby/RubyHash.java:1192
initialize at /opt/jruby/active/lib/ruby/1.9/webrick/utils.rb:178
It seems to be coming from inside of webrick, so I’m at a loss as to
what I
could be doing that would be causing this?
The upstart command for Mizuno is the following:
exec su -c ‘cd /home/foo/current; ./bin/mizuno >> output.log 2>&1’ -l
deploy
Any help would be appreciated in working out what is causing the error.
I’ve had issues like this when modifying a hash that I am already
iterating over. So instead of:
my_hash.each_pair { … do something that potentially modifies the hash
… }
keys = my_hash.keys
keys.each { … do something that potentially modifies the hash … }
(or my_hash.keys.each {}, I think would work too.)
Don’t know if this is your issue, but I hope it helps.
Are you deleting a key inside the loop? If so, I don’t think the
enumerable on which the loop depends would know. It would try to
perform an iteration on that key, that would now be missing.
If you’re just adding keys, and you’re iterating over the hash to which
you’re adding them, you might instead want to create a temporary hash
for the additions within the loop, and then merge the two hashes after
the loop.
Seems like a bug in the webrick implementation, no?
@timeout_info.each {} is iterating, while @timeout_info[thread] ||=
Array.new
is adding new values to the hash.
The error stops happening once the thread pool has filled, and the hash
doesn’t get iterated over anymore. And it happens intermittently, just
because the while loop could be sleeping during that time.
I think you are right though, I could monkey patch this to use @timeout_info.keys.each, and see if that resolves the issue, and if so,
push it back.
the additions within the loop, and then merge the two hashes after the loop.
/opt/jruby/active/lib/ruby/1.9/webrick/utils.rb:232:in `timeout’
initialize at /opt/jruby/active/lib/ruby/1.9/webrick/utils.rb:179
Any help would be appreciated in working out what is causing the error.
E: [email protected]
T: http://www.twitter.com/neurotic
W: www.compoundtheory.com
My opinion is that even if the error only manifests itself intermittently,
the issue is a conceptual one (not to modify the hash you’re iterating
over), and it’s best to avoid it.
My opinion is that even if the error only manifests itself
intermittently, the issue is a conceptual one (not to modify the hash
you’re iterating over), and it’s best to avoid it.
I think were we are getting our wires crossed is over one specific
point:
This is not my code. This is code that is implement in JRuby’s
implementation of Webrick.
I have no control over that code whatsoever, unless I monkey patch it or
similar.
(if you look at the source links, you can see it all goes back to the
official JRuby GitHub repository).
Either way, what you were saying is making a lot of sense, and I may be
able to patch this code and push it back to JRuby (here’s hoping I don’t
break anything).
After reading through this thread, I’m not clear whether there’s an
issue to be fixed here. Seems like the original reporter’s issue was
due to a bad config.ru. And if there’s actually anything to be fixed,
it sounds like it’s something in WEBrick.
At this point, if there’s an issue, please file it and we’ll try to fix
it.
Don’t use webrick in anything in but development (which is well
documented). It tends to fall over under load.
I had a bad config.ru, and also very little knowledge of J/Ruby, and
din’t
realise this, and thought I was running Mizuno (which is fine under
load),
when in fact all I was running was Webrick.
