interrupted system call when process is signaled

[erickt]

I'm using curb inside of a unicorn app, and I found that whenever I use unicorn's signal-based restarting mechanism, I inevitably get a lot of these errors:

RuntimeError - select(): Interrupted system call:
 /shared/bundle/ruby/1.9.1/gems/curb-0.7.17/lib/curl/easy.rb:39:in `perform'
 /shared/bundle/ruby/1.9.1/gems/curb-0.7.17/lib/curl/easy.rb:39:in `perform'
 /shared/bundle/ruby/1.9.1/gems/tire-0.4.2/lib/tire/http/clients/curb.rb:23:in `http_post'
 /shared/bundle/ruby/1.9.1/gems/tire-0.4.2/lib/tire/http/clients/curb.rb:23:in `get'
 /shared/bundle/ruby/1.9.1/gems/tire-0.4.2/lib/tire/search/multiget.rb:45:in `perform'
 /shared/bundle/ruby/1.9.1/gems/tire-0.4.2/lib/tire/search/multiget.rb:23:in `results'
...

I've reduced this bug to somewhere between v0.7.16 and v0.7.17. I'm guessing it's commit 1c8ec56, which added a curb_select call that uses select. I think the proper way to handle this that when curb_select is interrupted, it should adjust the tv and retry the select. See Ruby's thread.c for one way of handling this.

[stayhero]

Having the same probs with running curb from a delayed job.

[dcerri]

i get the same in

vendor/bundle/ruby/1.9.1/gems/rpm_contrib-2.1.6/lib/rpm_contrib/instrumentation/curb.rb:44:in `perform'
vendor/bundle/ruby/1.9.1/gems/rpm_contrib-2.1.6/lib/rpm_contrib/instrumentation/curb.rb:44:in `block in perform_with_newrelic_trace'
vendor/bundle/ruby/1.9.1/gems/newrelic_rpm-3.5.2.17/lib/new_relic/agent/method_tracer.rb:242:in `trace_execution_scoped'
vendor/bundle/ruby/1.9.1/gems/rpm_contrib-2.1.6/lib/rpm_contrib/instrumentation/curb.rb:43:in `perform_with_newrelic_trace'
vendor/bundle/ruby/1.9.1/gems/curb-0.8.1/lib/curl/easy.rb:56:in `perform'

[taf2]

c.nosignal = true # maybe fixes this issue?