Konstigt beteende i köhantering sänker hela maskinen

[jesjos]

Detta sker när jag gör commit-requests från webbklienten.
Vissa meddelanden verkar fastna i beanstalk.
Den första commit-requesten fungerar oftast inte, ingen aning om varför.
Allvarligare är att poller fastnar i ngnslags loop där den klagar på DEADLINE_SOON.
Det visar sig som hundratals meddelanden som detta:

15:20:33 poller.1      | ActiveMessaging: thread[default]: Exception from connection.receive: DEADLINE_SOON
15:20:33 poller.1      | /Users/jesper/.rvm/gems/ruby-1.9.2-p180/gems/beanstalk-client-1.1.1/lib/beanstalk-client/connection.rb:212:in `check_resp'
15:20:33 poller.1      |    /Users/jesper/.rvm/gems/ruby-1.9.2-p180/gems/beanstalk-client-1.1.1/lib/beanstalk-client/connection.rb:224:in `read_job'
15:20:33 poller.1      |    /Users/jesper/.rvm/gems/ruby-1.9.2-p180/gems/beanstalk-client-1.1.1/lib/beanstalk-client/connection.rb:109:in `reserve'
15:20:33 poller.1      |    /Users/jesper/.rvm/gems/ruby-1.9.2-p180/gems/beanstalk-client-1.1.1/lib/beanstalk-client/connection.rb:394:in `call_wrap'
15:20:33 poller.1      |    /Users/jesper/.rvm/gems/ruby-1.9.2-p180/gems/beanstalk-client-1.1.1/lib/beanstalk-client/connection.rb:419:in `block in send_to_rand_conn'
15:20:33 poller.1      |    /Users/jesper/.rvm/gems/ruby-1.9.2-p180/gems/beanstalk-client-1.1.1/lib/beanstalk-client/connection.rb:403:in `retry_wrap'
15:20:33 poller.1      |    /Users/jesper/.rvm/gems/ruby-1.9.2-p180/gems/beanstalk-client-1.1.1/lib/beanstalk-client/connection.rb:419:in `send_to_rand_conn'
15:20:33 poller.1      |    /Users/jesper/.rvm/gems/ruby-1.9.2-p180/gems/beanstalk-client-1.1.1/lib/beanstalk-client/connection.rb:311:in `reserve'
15:20:33 poller.1      |    /Users/jesper/.rvm/gems/ruby-1.9.2-p180/bundler/gems/activemessaging-1c9e1ce5e7cc/lib/activemessaging/adapters/beanstalk.rb:50:in `receive'
15:20:33 poller.1      |    /Users/jesper/.rvm/gems/ruby-1.9.2-p180/bundler/gems/activemessaging-1c9e1ce5e7cc/lib/activemessaging/gateway.rb:36:in `block (2 levels) in start'

Processor usage går upp taket och alla operationer tar jättelänge.

Vet @oleander vad detta kan bero på?