Signal handling for sleep

[azrle]

@kazuho

We observed that KILL_OLD_DELAY may be interrupted by signal while waiting for new workers, using Perl 5.16, Linux.

sleep $kill_old_delay;

As far as I know, sleep can be interrupted and Perl will not call the system call (in our environment, implemented by nanosleep) again for rest of seconds.

As a result, in some use cases, old workers are unexpectedly killed before new workers become capable to accept requests.
For example,

• try to reload env while waiting for new workers
• reload codes while waiting for new workers

Is it by design?
Or is it okay to fix it by sleeping again for rest of seconds when it was interrupted?

[kazuho]

Thank you for opening the issue.

Or is it okay to fix it by sleeping again for rest of seconds when it was interrupted?

I think this is the right thing to do.

[azrle]

Thanks.
I would send a PR recently. :)

[azrle]

@kazuho I have sent a pull request #45. Please check it.

Thanks.

[azrle]

See it in 0.33. :)