Failing deferred tasks caught by EM.error_handler are run repeatedly

[leekirchhoff]

If an EM.error_handler is introduced that does not call EM.stop (e.g., it just logs an error and allows EM.run loop to continue), a task that fails and is caught by this error_handler is run repeatedly and no other deferred tasks get run.

This is due to code in EventMachine#run_deferred_callbacks that does not remove callbacks from @next_tick_queue if there is a failure when executing the callback. This code failure is not caught in test_error_handler because the test error_handler calls EM.stop.

The run_deferred_callbacks method was patched as follows to fix this problem:

def self.run_deferred_callbacks
  until (@resultqueue ||= []).empty?
    result,cback = @resultqueue.pop
    cback.call result if cback
  end

  @next_tick_queue ||= []
  if (l = @next_tick_queue.length) > 0
    l.times { cback=@next_tick_queue.shift and cback.call }
  end
end

[paulhammond]

Testing this bug in EventMachine 1.0.0.beta.3 shows that this is no longer the case. EventMachine#run_deferred_callbacks now removes callbacks from the @next_tick_queue before calling them.

Unfortunately you get a different problem (as described in #185) but I think this bug as described is no longer an issue.

[dj2]

@tmm1, anything left to do on this one?

[tmm1]

Closing.