Cluster round-robin scheduler loses request

[alexeyten]

Example code is here: https://gist.github.com/alexeyten/45fb4319e4158feb793a

I've tested this code with Node.js 4.2.1

This server just response with OK and restart workers after 200-300 completed requests. With default cluster scheduler (SCHED_RR) server eventually loses some requests and ab stops with timeout error. When I uncomment line 5 and use SCHED_NONE scheduler, then server works fine.

/cc @indutny

[indutny]

cc @bnoordhuis

[evanlucas]

I am able to reproduce this consistently

[indutny]

@alexeyten I think it should be fixed by da21dba in master, could you please verify?

[bnoordhuis]

I can still reproduce with master. I'll try to look into it next week.

[bnoordhuis]

Interestingly enough, today's master works fine. @alexeyten Can you confirm?

[bnoordhuis]

Okay, I take that back. It still happens, just infrequently.

[bnoordhuis]

@alexeyten Can you confirm that #3677 fixes the issue for you?

[alexeyten]

@bnoordhuis looks so.

At least it handled 1M requests without problems.