VSL record.  Still not quite sure how they happen, but we shouldn't hang
no matter what.

the same address again, even with hinting.

pretend nothing else happened.

this should solve issues where the c2-v1 connection times out before
it ever got used.

fd's early in the case of a timeout.

Fixes	#1023

Apache (and nginx) uses 8k, so use the same default size to avoid
people being surprised by more 413 responses than necessary.

Fixes: #1016

Thanks to cleberjsantos for initial patch which this is based on.

Fixes: #976

are no lingering bits.

Convert an assert to a failure.

Fixes	#1022

steps the clock...

Fixes	#1022

Make pipe mode use the acct_tmp like everybody else.

Make one single thread, which creates thread pools, and since that
is not particularly ardous work, also have it maintain the stats
gauge of number of sessions queued.

(Removing pools should now be possible, but the code is not there yet.)

Each pool has its own herder thread, which creates and destroys
threads as required, and updates stats for dropped and queued
sessions.

Sanitize VSC fields associated with this area.

Schedule sessions from busy-lists and from the waiter via
SES_Schedule(), to Pool_Schedule() in the right pool.

Store the data from accept(2) on the worker->ws, we only need it for
a brief moment until the worker configures a session with it.

If we fail to allocate a session after accepting the connection,
we silently close the connection again:  Presumably it is a DoS
situation.  (false positive:  Extremely popular busy objects)

Make three clearly defined counters for sessions:  Accepted, Dropped
and failed (accept(2) failure).

Lots of spit&polish.

Also check if the http_max_hdr param changes.

Describe http_max_hdr as per pool.

Add SES_DeletePool() function.

sockaddr's directly into struct sess.

and it's all a mess.

Try to get it right, and document in the head-comment what I mean with
"right".

Goal:  Cleaned up for 3.1

that they should be mined and dumped into the documentation.

Feel free to add (or send patches with descriptions, preferably
only one line or paragraph for each.

Instead of a single thread which accepts all sockets with a poll/accept
sequence, each thread-pool will have a thread accepting on each socket.

If no threads are available, the sockets will not be accepted on.

CAVEATS:

This commit undoubtedly leaves a number of minor issues dangling,
they will be cleaned up as we find them.

Please notice that there are changes to stats counters (some don't
work right now a nd more changes will be coming)

Changing certain acceptor-related params on the fly may not work.

Testing would be very welcome.