GitLab

And additional debug code for tracking errant close()s

Bryan Duncan authored Nov 14, 2011 and Quanah Gibson-Mount committed Oct 19, 2017

Patch obtained from:
        http://public.me.com/bryan.duncan/bryan-duncan.kqueue.090922.patch

 # This patch file is derived from OpenLDAP Software. All of the
 # modifications to OpenLDAP Software represented in the following
 # patch(es) were developed by Apple.  I, Bryan Duncan, am authorized by
 # Apple, my employer, to release this work under the following terms.
 #
 # Copyright 2009 Apple Inc. All rights reserved.
 #
 # Redistribution and use in source and binary forms, with or without
 # modification, are permitted only as authorized by the OpenLDAP
 # Public License.
 #
 # A copy of this license is available in file LICENSE in the
 # top-level directory of the distribution or, alternatively, at
 # http://www.OpenLDAP.org/license.html.
 #

for /dev/poll is neither enabled nor tested, so other issues may exist.

Howard Chu authored Feb 09, 2017 and Howard Chu committed Sep 08, 2017

Typically only shows up under valgrind, not in regular runs

Remove poolq_hash, it wasn't distributing work evenly to the queues.
Just walk through all queues and use the one with smallest
active+pending count. Since pool_retract also relied on the hash,
a different means of locating the thread to retract was needed.
Add pool_submit2 which returns the threadpool task structure,
and record which poolq this task lives on.

Writer threads do their own wait using select/poll instead of
asking the listener thread. Eliminates one mutex+one condvar
per conn plus multiple wakeups of the listener thread. Also
fixes writetimeout to wait an exact time, instead of the
approximation used in the listener thread.

Also allow the max to be redefined at compile time. Of course, there is no
legitimate reason to do so.

This broke if i was incremented before the new thread read i:
    for(...; i++) { ldap_pvt_thread_create(,,, &i); ... }