This change allows timing implementation data to be stored directly
on the timer itself thus removing the requirement for many
implementations to do a container lookup for the same information.
This means that API calls into timing implementations can directly
access the information they need instead of having to find it.
Review: https://reviewboard.asterisk.org/r/3175/
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@407749 65c4cc65-6c06-0410-ace0-fbb531ad65f3
There were several reports of deadlock when using
res_timing_pthread. Backtraces indicated that one thread was blocked
waiting for the write to the pipe to complete and this thread held
the container lock for the timers. Therefore any thread that wanted
to create a new timer or read an existing timer would block waiting
for either the timer lock or the container lock and deadlock ensued.
This patch changes the way the pipe is used to eliminate this source
of deadlocks:
1) The pipe is placed in non-blocking mode so that it would never
block even if the following changes someone fail...
2) Instead of writing bytes into the pipe for each "tick" that's
fired the pipe now has two states--signaled and unsignaled. If
signaled, the pipe is hot and any pollers of the read side
filedescriptor will be woken up. If unsigned the pipe is idle. This
eliminates even the chance of filling up the pipe and reduces the
potential overhead of calling unnecessary writes.
3) Since we're tracking the signaled / unsignaled state, we can
eliminate the exta poll system call for every firing because we know
that there is data to be read.
(closes issue ASTERISK-21389)
Reported by: Matt Jordan
Tested by: Shaun Ruffell, Matt Jordan, Tony Lewis
patches:
0001-res_timing_pthread-Reduce-probability-of-deadlocking.patch uploaded by sruffell (License 5417)
(closes issue ASTERISK-19754)
Reported by: Nikola Ciprich
(closes issue ASTERISK-20577)
Reported by: Kien Kennedy
(closes issue ASTERISK-17436)
Reported by: Henry Fernandes
(closes issue ASTERISK-17467)
Reported by: isrl
(closes issue ASTERISK-17458)
Reported by: isrl
Review: https://reviewboard.asterisk.org/r/2441/
........
Merged revisions 386109 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 386159 from http://svn.asterisk.org/svn/asterisk/branches/11
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@386160 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Currently, if an acknowledgement of a timer fails Asterisk will not realize
that a serious error occurred and will continue attempting to use the timer's
file descriptor. This can lead to situations where errors stream to the
CLI/log file. This consumes significant resources, masks the actual problem
that occurred (whatever caused the timer to fail in the first place), and
can leave channels in odd states.
This patch propagates the errors in the timing resource modules up through
the timer core, and makes users of these timers handle acknowledgement
failures. It also adds some defensive coding around the use of timers
to prevent using bad file descriptors in off nominal code paths.
Note that the patch created by the issue reporter was modified slightly for
this commit and backported to 1.8, as it was originally written for
Asterisk 10.
Review: https://reviewboard.asterisk.org/r/2178/
(issue ASTERISK-20032)
Reported by: Jeremiah Gowdy
patches:
jgowdy-timerfd-6-22-2012.diff uploaded by Jeremiah Gowdy (license 6358)
........
Merged revisions 375893 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 375894 from http://svn.asterisk.org/svn/asterisk/branches/10
........
Merged revisions 375895 from http://svn.asterisk.org/svn/asterisk/branches/11
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@375896 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Prior to this patch, res_musiconhold existed at the same module priority level
as the timing sources that it depends on. This would cause a problem when
music on hold was reloaded, as the timing source could be changed after
res_musiconhold was processed. This patch adds a new module priority level,
AST_MODPRI_TIMING, that the various timing modules are now loaded at. This
now occurs before loading other resource modules, such that the timing source
is guaranteed to be set prior to resolving the timing source dependencies.
(closes issue ASTERISK-17474)
Reporter: Luke H
Tested by: Luke H, Vladimir Mikhelson, zzsurf, Wes Van Tlghem, elguero, Thomas Arimont
Patches:
asterisk-17474-dahdi_timing-infinite-wait-fix_v3_branch-1.8.diff uploaded by elguero (License #5026)
asterisk-17474-dahdi_timing-infinite-wait-fix_v3_branch-10.diff uploaded by elguero (License #5026)
asterisk-17474-dahdi_timing-infinite-wait-fix_v3.diff uploaded by elguero (License #5026)
Review: https://reviewboard.asterisk.org/r/1578/
........
Merged revisions 349194 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 349195 from http://svn.asterisk.org/svn/asterisk/branches/10
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@349196 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This patch adds the option to give a module a load priority. The value represents the order in which a module's load() function is initialized. The lower the value, the higher the priority. The value is only checked if the AST_MODFLAG_LOAD_ORDER flag is set. If the AST_MODFLAG_LOAD_ORDER flag is not set, the value will never be read and the module will be given the lowest possible priority
on load. Since some modules are reliant on a timing interface, the timing modules have been given a high load priorty.
(closes issue #15191)
Reported by: alecdavis
Tested by: dvossel
Review: https://reviewboard.asterisk.org/r/262/
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@199743 65c4cc65-6c06-0410-ace0-fbb531ad65f3
The situation that caused this problem was when continuous mode was being
turned on and off while a rate was set for a timing interface. A very easy
way to replicate this bug was to do a Playback() from behind a Local channel.
In this scenario, a rate gets set on the channel for doing file playback.
At the same time, continuous mode gets turned on and off about every 20 ms
as frames get queued on to the PBX side channel from the other side of the
Local channel.
Essentially, this module treated continuous mode and a set rate as mutually
exclusive states for the timer to be in. When I dug deep enough, I observed
the following pattern:
1) Set timer to tick every 20 ms.
2) Wait almost 20 ms ...
3) Continuous mode gets turned on for a queued up frame
4) Continuous mode gets turned off
5) The timer goes back to its tick per 20 ms. state but starts counting
at 0 ms.
6) Goto step 2.
Sometimes, res_timing_pthread would make it 20 ms and produce a timer tick,
but not most of the time. This is what produced the choppy sound (or sometimes
no sound at all).
Now, the module treats continuous mode and a set rate as completely independent
timer modes. They can be enabled and disabled independently of each other and
things work as expected.
(closes issue #14412)
Reported by: dome
Patches:
issue14412.diff.txt uploaded by russell (license 2)
issue14412-1.6.1.0.diff.txt uploaded by russell (license 2)
Tested by: DennisD, russell
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@198146 65c4cc65-6c06-0410-ace0-fbb531ad65f3
1) Add module use count handling so that timing modules can be unloaded.
2) Implement unload_module() functions for the timing interface modules.
3) Allow multiple timing modules to be loaded, and use the one with the
highest priority value.
4) Report which timing module is being use in the "timing test" CLI command.
(closes issue #14489)
Reported by: russell
Review: http://reviewboard.digium.com/r/162/
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@176666 65c4cc65-6c06-0410-ace0-fbb531ad65f3
is used while continuous mode was already turned on.
(closes issue #13738)
Reported by: smurfix
Patches:
res.patch.fixed uploaded by smurfix (license 547)
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@163241 65c4cc65-6c06-0410-ace0-fbb531ad65f3
it would be best to maintain API compatibility. Instead, this commit introduces
ao2_callback_data() which is functionally identical to ao2_callback() except
that it allows you to pass arbitrary data to the callback.
Reviewed by Mark Michelson via ReviewBoard:
http://reviewboard.digium.com/r/64
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@158959 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This provides a new timing interface. In order to use it,
you must be running a Linux with a kernel version of
2.6.25 or newer and glibc 2.8 or newer.
This timing interface is a good alternative if a timing
source is necessary (e.g. for IAX trunking) but DAHDI is
otherwise unnecessary for the system.
For now, this commit contains the actual work done in the
res_timing_timerfd branch. There are no notices in the README
or CHANGES files yet, but they will be added in my next commit.
The timing API of Asterisk also needs to have a bit of work done
with regards to choosing which timing interface to use. This commit
makes the choice a build-time decision, by only allowing one of
the timer interfaces to be chosen in menuselect. It would be preferable
if the choice could be made at run-time, however. The preferred timing
interface could be loaded and tested, and if it does not work, choice
number two may be used instead. That sort of thing. That is beyond
the scope of work in this branch though.
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@157820 65c4cc65-6c06-0410-ace0-fbb531ad65f3
ao2_callback and ao2_find). Currently, passing OBJ_POINTER to either
of these mandates that the passed 'arg' is a hashable object, making
searching for an ao2 object based on outside criteria difficult.
Reviewed by Russell and Mark M. via ReviewBoard:
http://reviewboard.digium.com/r/36/
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@155401 65c4cc65-6c06-0410-ace0-fbb531ad65f3
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r140488 | mmichelson | 2008-08-29 12:34:17 -0500 (Fri, 29 Aug 2008) | 22 lines
After working on the ao2_containers branch, I noticed
something a bit strange. In all cases where we provide
a callback function to ao2_container_alloc, the callback
function would only return 0 or CMP_MATCH. After inspecting
the ao2_callback() code carefully, I found that if you're
only looking for one specific item, then you should return
CMP_MATCH | CMP_STOP. Otherwise, astobj2 will continue
traversing the current bucket until the end searching for
more matches.
In cases like chan_iax2 where in 1.4, all the peers are
shoved into a single bucket, this makes for potentially
terrible performance since the entire bucket will be
traversed even if the peer is one of the first ones come
across in the bucket.
All the changes I have made were for cases where the
callback function defined was passed to ao2_container_alloc
so that calls to ao2_find could find a unique instance
of whatever object was being stored in the container.
........
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@140489 65c4cc65-6c06-0410-ace0-fbb531ad65f3
- change ast_settimeout() to honor max rate in edge cases of file playback
(this will make some warning messages go away at the end of playing back
a file)
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@125332 65c4cc65-6c06-0410-ace0-fbb531ad65f3
does not require DAHDI. It's called "pthread" because it uses a pthread
API call in the timing thread for sleeping and ensuring we wake up at
an appropriate time. I wasn't sure what else to call it. :)
The timing API requires a file descriptor that can be polled on. So,
when you open a timer, this module creates a pipe and returns the read
end of the pipe. There is a background thread that wakes up every 10ms
and checks to see if any of the currently open timers need a 'tick' and
writes to the appropriate pipe.
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@122928 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Joshua Colp
Matthew Jordan
Leif Madsen
Russell Bryant
Tilghman Lesher
David Vossel
Sean Bright
Mark Michelson