Sending USR2 to mongrel_rails sometimes results in an “Address already in use” on the restart
- by Ben
We have a rolling-restart mode for our mongrel cluster that sends a USR2 signal to each running process.
This works great, most of the time. But very occasionally the mongrel process will shutdown, and then fail to restart, with the following error:
/usr/local/lib/ruby/gems/1.8/gems/mongrel-1.1.5/bin/../lib/mongrel/tcphack.rb:12:in `initialize_without_backlog': Address already in use - bind(2) (Errno::EADDRINUSE)
from /usr/local/lib/ruby/gems/1.8/gems/mongrel-1.1.5/bin/../lib/mongrel/tcphack.rb:12:in `initialize'
from /usr/local/lib/ruby/gems/1.8/gems/mongrel-1.1.5/bin/../lib/mongrel.rb:93:in `new'
from /usr/local/lib/ruby/gems/1.8/gems/mongrel-1.1.5/bin/../lib/mongrel.rb:93:in `initialize'
from /usr/local/lib/ruby/gems/1.8/gems/mongrel-1.1.5/bin/../lib/mongrel/configurator.rb:139:in `new'
from /usr/local/lib/ruby/gems/1.8/gems/mongrel-1.1.5/bin/../lib/mongrel/configurator.rb:139:in `listener'
Looking though the mongrel source, the USR2 handler calls a synchronous stop on the running server, so it ought to block until the socket has been released.
Has anyone seen this error?
Does anyone have any ideas what might cause it?
(I asked this question over on StackOverflow initially, but thought it might be more appropriate here)
