It seems that a stream of [fileevent] events can starve [after] events,
which presents a problem since Httpd uses [after] events to communicate
between coroutines.
Unsure of the root cause of this starvation, but it may be that we need to
limit the number of requests on a pipeline, or to introduce a break on the
pipeline's readable [fileevent] stream by occasionally setting the handler
to {}, ie: turning off the fileevent stream, then turning it on sometime later.
Probably need to measure the number of calculated but undelivered responses
to see if this is necessary. That may be an analog of lag.
Best solution would be to make event handling starvation-free.
Original issue reported on code.google.com by mcc...@gmail.com on 5 Mar 2009 at 11:41
Original issue reported on code.google.com by
mcc...@gmail.com
on 5 Mar 2009 at 11:41