Implement --max-keepalive-requests

[eserte]

I would like to see something like Apache's MaxKeepAliveRequests implemented for Starman.

Rationale: on a heavily used server with few clients it is possible that a worker process lives for very long time, even with the default setting of --max-requests=1000. Reason is that --max-requests does not count requests within a keepalive connection.

The implementation is probably quite simple, something like the following in process_request in Starman::Server (untested):

     if (exists $self->{options}->{max_keepalive_requests}) {
         last if ++$keepalive_requests >= $self->{options}->{max_keepalive_requests};
     }
     DEBUG && warn "[$$] Waiting on previous connection for keep-alive request...\n";

If this feature request is accepted, then I would do a formal pull request.

[miyagawa]

Sounds reasonable to me!

[Logioniz]

Oh.... i have already make pull request to fix this situation with max_requests and keepalive connections without max-keepalive-requests param. #142

@eserte i look at your code eserte@06cee30 and detect race between closing connection and sending new request to starman because your code not changed "Connection" header to value "close" which led to http 502 from the balancer under heavy load.

I can to help to fix your problem but first need to understand what kind of logic we will implement (mine or with flag --max-keepalive-requests).

[pepl]

@miyagawa any update on this?