During some performance testing of Asterisk with AGI, ARI, and lots of Local
channels, we noticed that there's quite a hit in performance during channel
creation and releasing to the dialplan (ARI continue). After investigating
the performance spike that occurs during channel creation, we discovered
that we create a lot of channel snapshots that are technically unnecessary.
This includes creating snapshots during:
* AGI execution
* Returning objects for ARI commands
* During some Local channel operations
* During some dialling operations
* During variable setting
* During some bridging operations
And more.
This patch does the following:
- It removes a number of fields from channel snapshots. These fields were
rarely used, were expensive to have on the snapshot, and hurt performance.
This included formats, translation paths, Log Call ID, callgroup, pickup
group, and all channel variables. As a result, AMI Status,
"core show channel", "core show channelvar", and "pjsip show channel" were
modified to either hit the live channel or not show certain pieces of data.
While this is unfortunate, the performance gain from this patch is worth
the loss in behaviour.
- It adds a mechanism to publish a cached snapshot + blob. A large number of
publications were changed to use this, including:
- During Dial begin
- During Variable assignment (if no AMI variables are emitted - if AMI
variables are set, we have to make snapshots when a variable is changed)
- During channel pickup
- When a channel is put on hold/unhold
- When a DTMF digit is begun/ended
- When creating a bridge snapshot
- When an AOC event is raised
- During Local channel optimization/Local bridging
- When endpoint snapshots are generated
- All AGI events
- All ARI responses that return a channel
- Events in the AgentPool, MeetMe, and some in Queue
- Additionally, some extraneous channel snapshots were being made that were
unnecessary. These were removed.
- The result of ast_hashtab_hash_string is now cached in stasis_cache. This
reduces a large number of calls to ast_hashtab_hash_string, which reduced
the amount of time spent in this function in gprof by around 50%.
#ASTERISK-23811 #close
Reported by: Matt Jordan
Review: https://reviewboard.asterisk.org/r/3568/
........
Merged revisions 416211 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@416216 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This patch is a re-do of r414122.
When r414122 was merged, a major problem with it was uncovered. UNBRIDGE soft
hangup flags have a catastrophic effect on the pbx core if they leak out from
the bridge layer: the channel gets hung up. With the number of threads
involved in a blind transfer, and with the initial patch, it was likely that
this would occur. This caused a large number of test failures
This patch is nearly identical with the one proposed in r414122, save for the
following changes:
- We explicitly clear the UNBRIDGE flag when setting an after goto on a
channel in a bridge
- Defensively, if we encounter an UNBRIDGE flag in the pbx core, we handle it
https://reviewboard.asterisk.org/r/3585/
........
Merged revisions 415443 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@415444 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This patch fixes issues with direct media bridges that occur after a blind
transfer. These issues were caught by the (currently failing)
pjsip/transfers/blind_transfer/caller_direct_media test.
The test currently fails primarily for two reasons:
(1) When Bob and Charlie (the transfer target and the transfer destination)
enter a bridge together, the framehook remains on the transfer target
channel until both channels are in the bridge. As it consumes voice frames,
the initial bridge type is a simple bridge. The framehook is removed when
both channels are in the bridge; however, this does not currently cause the
bridging framework to re-evaluate the bridge. This patch adds a
AST_SOFTHANGUP_UNBRIDGE poke to the transfer target channel when a
framehook is removed so the bridge can re-evaluate itself.
(2) When a channel leaves a native RTP bridge, it may be leaving due to being
hung up. Sending a re-INVITE to a channel that is about to be hung up is
not nice - in fact, there's a good chance we'll send the BYE request before
the channel has had a chance to send back a 200 OK. To be somewhat nicer,
this patch adds a function to channel.h that allows the bridging framework
to query for exactly why a channel is leaving a bridge via the channel's
soft hangup flags. This allows it to only send the re-INVITE if there's a
chance the channel will survive the native bridging experience.
Review: https://reviewboard.asterisk.org/r/3535/
........
Merged revisions 414122 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@414123 65c4cc65-6c06-0410-ace0-fbb531ad65f3
In the past framehooks have had no capability to determine what frame types a hook
is actually interested in consuming. This has meant that code has had to assume they
want all frames, thus preventing native bridging.
This change adds a callback which allows a framehook to be queried for whether it
is consuming a frame of a specific type. The native RTP bridging module has also
been updated to take advantange of this, allowing native bridging to occur when
previously it would not.
ASTERISK-23497 #comment Reported by: Etienne Lessard
ASTERISK-23497 #close
Review: https://reviewboard.asterisk.org/r/3522/
........
Merged revisions 413681 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@413682 65c4cc65-6c06-0410-ace0-fbb531ad65f3
In the past framehooks have had no capability to determine what frame types a hook
is actually interested in consuming. This has meant that code has had to assume they
want all frames, thus preventing native bridging.
This change adds a callback which allows a framehook to be queried for whether it
is consuming a frame of a specific type. The native RTP bridging module has also
been updated to take advantange of this, allowing native bridging to occur when
previously it would not.
ASTERISK-23497 #comment Reported by: Etienne Lessard
ASTERISK-23497 #close
Review: https://reviewboard.asterisk.org/r/3522/
........
Merged revisions 413650 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@413651 65c4cc65-6c06-0410-ace0-fbb531ad65f3
The masquerade supertest frequently fails because either the local channel
chain doesn't completely optimize out or the DTMF handshake doesn't
completely get accross. Local channel optimization requires frames
flowing to trigger when optimization can happen. When optimization
happens the media frame that triggered the optimization is dropped.
Sending DTMF requires frames to flow in the other direction for timing
purposes while sending nothing. If internal timing is not enabled when
MOH is playing, Asterisk switches to received timing when an audio frame
is received. With optimization dropping media frames and MOH not sending
frames unless it receives frames, occasionaly there are no more frames
being passed and the test fails.
* The asterisk command line -I option and the asterisk.conf
internal_timing option are removed. Asterisk now always uses internal
timing when needed if any timing module is loaded. The issue
ASTERISK-14861 did this quite awhile ago in v1.4 but effectively is broken
if other internal timing modules besides DAHDI are used. The
ast_read_generator_actions() now only does received timing if it has no
choice for frame generators like MOH, silence, and playback streaming.
* Cleaned up some code dealing with frame generators in
ast_deactivate_generator(), generator_write_format_change(),
ast_activate_generator(), and ast_channel_stop_silence_generator().
* Removed ast_internal_timing_enabled(), AST_OPT_FLAG_INTERNAL_TIMING, and
ast_opt_internal_timing.
ASTERISK-22846 #close
Reported by: Matt Jordan
Review: https://reviewboard.asterisk.org/r/3414/
........
Merged revisions 411715 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 411716 from http://svn.asterisk.org/svn/asterisk/branches/11
........
Merged revisions 411717 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@411724 65c4cc65-6c06-0410-ace0-fbb531ad65f3
* Assert if a channel is destroyed but has the snapshot staging flag set.
In this case the final channel destruction snapshot would never get taken.
* Assert if what we just got out of the stasis cache is not what we were
looking for. This assert would have saved several days searching for a
bug and a lot of my hair.
* Assert if the music on hold message posts could not find the associated
channel. A crash will happen later when manager tries to send the MOH AMI
message. This assert catches the problem when the stasis message is
posted instead of by the thread processing the defective message.
* Always generate a backtrace when an ast_assert() fails.
Review: https://reviewboard.asterisk.org/r/3411/
........
Merged revisions 411701 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@411702 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Playing back a file to a channel in an ARI bridge would attempt to wait until
the playback concluded before returning. The method used involved signaling the
waiting thread in the ARI custom playback function.
The problem with this is that there were some corner cases that were not accounted for:
* If a bridge channel could not be found, then we never would attempt the playback but
would still attempt to wait for the playback to complete.
* If the bridge playfile action failed to queue, we would still attempt to wait for the
playback to complete.
* If the bridge playfile action were queued but some circumstance caused the playback
not to occur (the bridge dies, the channel is removed from the bridge), then we would
never be notified.
The solution to this is to move the waiting logic into the bridge code. A new bridge
API function is added to queue a synchronous action on a bridge. The waiting thread
is notified when the queued frame has been freed, either due to an error occurring
or due to successful playback. As a failsafe, the waiting thread has a 10 minute
timeout just in case there is a frame leak somewhere.
Review: https://reviewboard.asterisk.org/r/3338
........
Merged revisions 410673 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@410684 65c4cc65-6c06-0410-ace0-fbb531ad65f3
* Made res_musiconhold.c always post the MusicOnHoldStart/MusicOnHoldStop
events when it actually starts/stops the music streams. This allows the
events to always happen when MOH starts/stops. The event posting code was
moved to the MOH alloc/release routines.
* Made channel_do_masquerade() stop any MOH on the original channel before
masquerading so the original channel will get a stop event with correct
information.
* Cleaned up a couple odd codings in moh_files_alloc() and moh_alloc()
dealing with the music state variable.
(issue ASTERISK-23311)
Reported by: Benjamin Keith Ford
Review: https://reviewboard.asterisk.org/r/3306/
........
Merged revisions 410493 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@410494 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Much needed was a way to assign id to objects on creation, and
much change was necessary to accomplish it. Channel uniqueids
and linkedids are split into separate string and creation time
components without breaking linkedid propgation. This allowed
the uniqueid to be specified by the user interface - and those
values are now carried through to channel creation, adding the
assignedids value to every function in the chain including the
channel drivers. For local channels, the second channel can be
specified or left to default to a ;2 suffix of first. In ARI,
bridge, playback, and snoop objects can also be created with a
specified uniqueid.
Along the way, the args order to allocating channels was fixed
in chan_mgcp and chan_gtalk, and linkedid is no longer lost as
masquerade occurs.
(closes issue ASTERISK-23120)
Review: https://reviewboard.asterisk.org/r/3191/
........
Merged revisions 410157 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@410158 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Updated the code to check to see if MOH is playing on the transferor and if
so then start it on the channel that replaces it during a masquerade.
Example scenario of the problem:
Alice calls Bob and then Bob begins the attended transfer process into a queue.
Upon going on hold Alice hears music and so does Bob once he is in the queue.
Bob then transfers Alice into the queue and then music for Alice stops even
though she should be hearing it since has now replaced Bob in the queue.
The problem that was occurring is that once the channel was masqueraded the app
(queues, confbridge, etc...) had no way of knowing that the channel had just
been swapped out thus it did not start music for the present channel.
Credit to Olle Johansson for pointing me in the right direction on this issue.
(closes issue ASTERISK-19499)
Reported by: Timo Teräs
Review: https://reviewboard.asterisk.org/r/3226/
........
Merged revisions 408642 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 408643 from http://svn.asterisk.org/svn/asterisk/branches/11
........
Merged revisions 408644 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@408645 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Implements the following cli commands:
pjsip list aors
pjsip list auths
pjsip list channels
pjsip list contacts
pjsip list endpoints
pjsip show aor(s)
pjsip show auth(s)
pjsip show channels
pjsip show endpoint(s)
Also...
Minor modifications made to the AMI command implementations to facilitate
reuse.
New function ast_variable_list_sort added to config.c and config.h to implement
variable list sorting.
(issue ASTERISK-22610)
patches:
pjsip_cli_v2.patch uploaded by george.joseph (License 6322)
........
Merged revisions 404480 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@404507 65c4cc65-6c06-0410-ace0-fbb531ad65f3
For the explanation, here is a copy-paste of the review board explanation:
Initially, it was discovered that performing an attended transfer of a
multiparty bridge with a PJSIP channel would cause a deadlock. A PBX thread
started a masquerade and reached the point where it was calling the fixup()
callback on the "original" channel. For chan_pjsip, this involves pushing a
synchronous task to the session's serializer. The problem was that a task ahead
of the fixup task was also attempting to perform a channel masquerade. However,
since masquerades are designed in a way to only allow for one to occur at a
time, the task ahead of the fixup could not continue until the masquerade
already in progress had completed. And of course, the masquerade in progress
could not complete until the task ahead of the fixup task had completed.
Deadlock.
The initial fix was to change the fixup task to be asynchronous. While this
prevented the deadlock from occurring, it had the frightful side effect of
potentially allowing for tasks in the session's serializer to operate on a
zombie channel.
Taking a step back from this particular deadlock, it became clear that the
problem was not really this one particular issue but that masquerades
themselves needed to be addressed. A PJSIP attended transfer operation calls
ast_channel_move(), which attempts to both set up and execute a masquerade. The
problem was that after it had set up the masquerade, the PBX thread had swooped
in and tried to actually perform the masquerade. Looking at changes that had
been made to Asterisk 12, it became clear that there never is any time now that
anyone ever wants to set up a masquerade and allow for the channel thread to
actually perform the masquerade. Everyone always is calling ast_channel_move(),
performs the masquerade itself before returning.
In this patch, I have removed all blocks of code from channel.c that will
attempt to perform a masquerade if ast_channel_masq() returns true. Now, there
is no distinction between setting up a masquerade and performing the
masquerade. It is one operation. The only remaining checks for
ast_channel_masq() and ast_channel_masqr() are in ast_hangup() since we do not
want to interrupt a masquerade by hanging up the channel. Instead, now
ast_hangup() will wait for a masquerade to complete before moving forward with
its operation.
The ast_channel_move() function has been modified to basically in-line the
logic that used to be in ast_channel_masquerade(). ast_channel_masquerade() has
been killed off for real. ast_channel_move() now has a lock associated with it
that is used to prevent any simultaneous moves from occurring at once. This
means there is no need to make sure that ast_channel_masq() or
ast_channel_masqr() are already set on a channel when ast_channel_move() is
called. It also means the channel container lock is not pulling double duty by
both keeping the container locked and preventing multiple masquerades from
occurring simultaneously.
The ast_do_masquerade() function has been renamed to do_channel_masquerade()
and is now internal to channel.c. The function now takes explicit arguments of
which channels are involved in the masquerade instead of a single channel.
While it probably is possible to do some further refactoring of this method, I
feel that I would be treading dangerously. Instead, all I did was change some
comments that no longer are true after this changeset.
The other more minor change introduced in this patch is to res_pjsip.c to make
ast_sip_push_task_synchronous() run the task in-place if we are already a SIP
servant thread. This is related to this patch because even when we isolate the
channel masquerade to only running in the SIP servant thread, we would still
deadlock when the fixup() callback is reached since we would essentially be
waiting forever for ourselves to finish before actually running the fixup. This
makes it so the fixup is run without having to push a task into a serializer at
all.
(closes issue ASTERISK-22936)
Reported by Jonathan Rose
Review: https://reviewboard.asterisk.org/r/3069
........
Merged revisions 404356 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@404368 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Original commit message by mmichelson (asterisk 12 r403311):
"This adds channel locks around calls to create channel snapshots as well
as other functions which operate on a channel and then end up
creating a channel snapshot. Functions that expect the channel to be
locked prior to being called have had their documentation updated to
indicate such."
The above was initially committed and then reverted at r403398. The problem
was found to be in core_local.c in the publish_local_bridge_message function.
The ast_unreal_lock_all function locks and adds a reference to the returned
channels and while they were being unlocked they were not being unreffed when
no longer needed. Fixed by unreffing the channels.
Also in bridge.c a lock was obtained on "other->chan", but then an attempt was
made to unlock "other" and not the previously locked channel. Fixed by
unlocking "other->chan"
(closes issue ASTERISK-22709)
Reported by: John Bigelow
........
Merged revisions 404237 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@404260 65c4cc65-6c06-0410-ace0-fbb531ad65f3
The change contains a slightly adjusted patch that was on the issue
(submitted by kmoore). A fix was made by adding in a bridge lock
while calling bridge_start/stop from the framehook callback. Since
the framehook callback is not called from the bridging core the bridge
is not locked, but needs to be before calling bridge_start.
(closes issue ASTERISK-22749)
Reported by: Kinsey Moore
Review: https://reviewboard.asterisk.org/r/3066/
Patches:
lock_inversion.diff uploaded by kmoore (license 6273)
........
Merged revisions 403767 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403768 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This adds channel locks around calls to create channel snapshots as well
as other functions which operate on a channel and then end up
creating a channel snapshot. Functions that expect the channel to be
locked prior to being called have had their documentation updated to
indicate such.
........
Merged revisions 403311 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@403314 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Two variables were being checked for NULLity immediately
after being declared NULL. I moved the NULL check until
after the variables are allocated.
This allows for the "channelvars" option in manager.conf
to work as intended again.
........
Merged revisions 402767 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@402768 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Debug messages aren't free. Even when the debug level is sufficiently low such
that the messages are never evaluated, there is a cost to having to parse
Asterisk logs that contain debug messages that (a) fail to convey sufficient
information or (b) occur so frequently as to be next to meaningless. Based on
having to stare at lots of DEBUG messages, this patch makes the following
changes:
* channel.c: When copying variables from a parent channel to a child channel,
specify the channels involved. Do not log anything for a variable that is not
inherited; the fact that it doesn't have an _ or __ already signifies that it
won't be inherited.
* pbx.c: Specify what function evaluation has occurred that created the result.
* translate.c: Bump up the translator path messages to 10. I've never once had
to use these debug messages, and for each format that is registered (on
startup) and unregistered (on shutdown) the entire f^2 matrix is logged out.
For short tests in the Asterisk Test Suite, this should make finding the
actual test much easier.
* xmldoc.c: The debug message that 'blah' is not found in the tree is expected.
Often, description elements - which are not required - are not provided.
This debug message adds no additional value, as it is not indicative of an
error or helpful in debugging which element did not contain a 'blah' element
as a child. If an element is supposed to contain a child element, then that
XML tree should have failed validation in the first place.
Review: https://reviewboard.asterisk.org/r/2966/
........
Merged revisions 402150 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 402151 from http://svn.asterisk.org/svn/asterisk/branches/11
........
Merged revisions 402154 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@402155 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Most callers of ast_channel_make_compatible() happen before the channels
enter a two party bridge. With the new bridging framework, two party
bridging technologies may also call ast_channel_make_compatible() when
there is more than one thread involved with the two channels.
* Added channel lock protection in set_format() and
ast_channel_make_compatible_helper() when dealing with the channel's
native formats while setting up a translation path.
* Fixed best_src_fmt and best_dst_fmt usage consistency in
ast_channel_make_compatible_helper(). The call to
ast_translator_best_choice() got them backwards.
* Updated some callers of ast_channel_make_compatible() and the function
documentation. There is actually a difference between the two channels
passed in.
* Fixed the deadlock potential in res_fax.c dealing with
ast_channel_make_compatible(). The deadlock potential was already there
anyway because res_fax called ast_channel_make_compatible() with chan
locked.
(closes issue ASTERISK-22542)
Reported by: Matt Jordan
Review: https://reviewboard.asterisk.org/r/2915/
........
Merged revisions 401239 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@401240 65c4cc65-6c06-0410-ace0-fbb531ad65f3
In r378303 the AST_FLAG_DISABLE_DEVSTATE_CACHE flag was added that tells
the devstate system to not cache states for non-real devices. However,
when optimizing away channels (ast_do_masquerade), that flag wasn't
copied.
In my case, using Local devices as queue members created a situation
where the endpoint was considered in use, but the state change of the
device being available again was ignored (not cached). The endpoint
channel was optimized into the (previously) Local channel, but kept
the do-not-cache flag. The end result being that the queue member
apparently stayed in use forever.
(closes issue ASTERISK-22718)
Reported by: Walter Doekes
Review: https://reviewboard.asterisk.org/r/2925/
........
Merged revisions 401178 from http://svn.asterisk.org/svn/asterisk/branches/1.8
........
Merged revisions 401179 from http://svn.asterisk.org/svn/asterisk/branches/11
........
Merged revisions 401180 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@401181 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Channel snapshots have string representations of the channel's native formats.
Prior to this change, the format strings were re-created on ever channel snapshot
creation. Since channel native formats rarely change, this was very wasteful.
Now, string representations of formats may optionally be stored on the ast_format_cap
for cases where string representations may be requested frequently. When formats
are altered, the string cache is marked as invalid. When strings are requested, the
cache validity is checked. If the cache is valid, then the cached strings are copied.
If the cache is invalid, then the string cache is rebuilt and copied, and the cache
is marked as being valid again.
Review: https://reviewboard.asterisk.org/r/2879
........
Merged revisions 400356 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@400363 65c4cc65-6c06-0410-ace0-fbb531ad65f3
........
r399887 | dlee | 2013-09-26 10:41:47 -0500 (Thu, 26 Sep 2013) | 1 line
Minor performance bump by not allocate manager variable struct if we don't need it
........
r400138 | dlee | 2013-09-30 10:24:00 -0500 (Mon, 30 Sep 2013) | 23 lines
Stasis performance improvements
This patch addresses several performance problems that were found in
the initial performance testing of Asterisk 12.
The Stasis dispatch object was allocated as an AO2 object, even though
it has a very confined lifecycle. This was replaced with a straight
ast_malloc().
The Stasis message router was spending an inordinate amount of time
searching hash tables. In this case, most of our routers had 6 or
fewer routes in them to begin with. This was replaced with an array
that's searched linearly for the route.
We more heavily rely on AO2 objects in Asterisk 12, and the memset()
in ao2_ref() actually became noticeable on the profile. This was
#ifdef'ed to only run when AO2_DEBUG was enabled.
After being misled by an erroneous comment in taskprocessor.c during
profiling, the wrong comment was removed.
Review: https://reviewboard.asterisk.org/r/2873/
........
r400178 | dlee | 2013-09-30 13:26:27 -0500 (Mon, 30 Sep 2013) | 24 lines
Taskprocessor optimization; switch Stasis to use taskprocessors
This patch optimizes taskprocessor to use a semaphore for signaling,
which the OS can do a better job at managing contention and waiting
that we can with a mutex and condition.
The taskprocessor execution was also slightly optimized to reduce the
number of locks taken.
The only observable difference in the taskprocessor implementation is
that when the final reference to the taskprocessor goes away, it will
execute all tasks to completion instead of discarding the unexecuted
tasks.
For systems where unnamed semaphores are not supported, a really
simple semaphore implementation is provided. (Which gives identical
performance as the original taskprocessor implementation).
The way we ended up implementing Stasis caused the threadpool to be a
burden instead of a boost to performance. This was switched to just
use taskprocessors directly for subscriptions.
Review: https://reviewboard.asterisk.org/r/2881/
........
r400180 | dlee | 2013-09-30 13:39:34 -0500 (Mon, 30 Sep 2013) | 28 lines
Optimize how Stasis forwards are dispatched
This patch optimizes how forwards are dispatched in Stasis.
Originally, forwards were dispatched as subscriptions that are invoked
on the publishing thread. This did not account for the vast number of
forwards we would end up having in the system, and the amount of work it
would take to walk though the forward subscriptions.
This patch modifies Stasis so that rather than walking the tree of
forwards on every dispatch, when forwards and subscriptions are changed,
the subscriber list for every topic in the tree is changed.
This has a couple of benefits. First, this reduces the workload of
dispatching messages. It also reduces contention when dispatching to
different topics that happen to forward to the same aggregation topic
(as happens with all of the channel, bridge and endpoint topics).
Since forwards are no longer subscriptions, the bulk of this patch is
simply changing stasis_subscription objects to stasis_forward objects
(which, admittedly, I should have done in the first place.)
Since this required me to yet again put in a growing array, I finally
abstracted that out into a set of ast_vector macros in
asterisk/vector.h.
Review: https://reviewboard.asterisk.org/r/2883/
........
r400181 | dlee | 2013-09-30 13:48:57 -0500 (Mon, 30 Sep 2013) | 28 lines
Remove dispatch object allocation from Stasis publishing
While looking for areas for performance improvement, I realized that an
unused feature in Stasis was negatively impacting performance.
When a message is sent to a subscriber, a dispatch object is allocated
for the dispatch, containing the topic the message was published to, the
subscriber the message is being sent to, and the message itself.
The topic is actually unused by any subscriber in Asterisk today. And
the subscriber is associated with the taskprocessor the message is being
dispatched to.
First, this patch removes the unused topic parameter from Stasis
subscription callbacks.
Second, this patch introduces the concept of taskprocessor local data,
data that may be set on a taskprocessor and provided along with the data
pointer when a task is pushed using the ast_taskprocessor_push_local()
call. This allows the task to have both data specific to that
taskprocessor, in addition to data specific to that invocation.
With those two changes, the dispatch object can be removed completely,
and the message is simply refcounted and sent directly to the
taskprocessor.
Review: https://reviewboard.asterisk.org/r/2884/
........
Merged revisions 399887,400138,400178,400180-400181 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@400186 65c4cc65-6c06-0410-ace0-fbb531ad65f3
The Dial, Queue, and FollowMe applications need to inhibit the bridging
initial connected line exchange in order to support the 'I' option.
* Replaced the pass_reference flag on ast_bridge_join() with a flags
parameter to pass other flags defined by enum ast_bridge_join_flags.
* Replaced the independent flag on ast_bridge_impart() with a flags
parameter to pass other flags defined by enum ast_bridge_impart_flags.
* Since the Dial, Queue, and FollowMe applications are now the only
callers of ast_bridge_call() and ast_bridge_call_with_flags(), changed the
calling contract to require the initial COLP exchange to already have been
done by the caller.
* Made all callers of ast_bridge_impart() check the return value. It is
important. As a precaution, I also made the compiler complain now if it
is not checked.
* Did some cleanup in parking_tests.c as a result of checking the
ast_bridge_impart() return value.
An independent, but associated change is:
* Reduce stack usage in ast_indicate_data() and add a dropping redundant
connected line verbose message.
(closes issue ASTERISK-22072)
Reported by: Joshua Colp
Review: https://reviewboard.asterisk.org/r/2845/
........
Merged revisions 399136 from http://svn.asterisk.org/svn/asterisk/branches/12
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@399138 65c4cc65-6c06-0410-ace0-fbb531ad65f3
DTMF start/end and hold/unhold events have state because a DTMF begin
event and hold event must be ended by something.
The following cases need to be handled when a channel is moved around in
the system.
* When a channel leaves a bridge it may owe a DTMF end event to the
bridge.
* When a channel leaves a bridge it may owe an UNHOLD event to the bridge.
(This case is explicitly ignored because things like transfers need
explicit control over this.)
* When a channel leaves the bridging system it may need to simulate a DTMF
end event to the channel.
* When a channel leaves the bridging system it may need to simulate an
UNHOLD event to the channel.
The patch also fixes the following:
* Fixes playing a file and restarting MOH using the latest MOH class used.
(closes issue ASTERISK-22043)
Reported by: Matt Jordan
Review: https://reviewboard.asterisk.org/r/2791/
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397577 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This patch adds pass through support for Opus and VP8. That includes:
* Format attribute negotiation for Opus. Note that unlike some other codecs,
the draft RFC specifies having spaces delimiting the attributes in addition
to ';', so you have "attra=X; attrb=Y". This broke the attribute parsing in
chan_sip, so a small tweak was also included in this patch for that.
* A format attribute negotiation module for Opus, res_format_attr_opus
* Fast picture update for VP8. Since VP8 uses a different RTCP packet number
than FIR, this really is specific to VP8 at this time.
Note that the format attribute negotiation in res_pjsip_sdp_rtp was written
by mjordan. The rest of this patch was written completely by Lorenzo Miniero.
Review: https://reviewboard.asterisk.org/r/2723/
(closes issue ASTERISK-21981)
Reported by: Tzafrir Cohen
patches:
asterisk_opus+vp8_passthrough_20130718.patch uploaded by lminiero (License 6518)
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397526 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This adds a new dialplan application, SayAlphaCase, that performs much
the same function as SayAlpha except that it takes additional options
which allow the user to specify whether the case of each letter should
be announced for uppercase, lowercase, or all letters. Similar
functionality has been added to the SAY ALPHA AGI command via an
optional parameter.
Original Patch by: Kevin Scott Adams
Reported by: Kevin Scott Adams
Review: https://reviewboard.asterisk.org/r/2725/
(closes issue ASTERISK-20782)
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@397493 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This patch does three things:
1. It provides a Surrogate channel technology with a consolidated
"implementation detail flag" on the channel technology. This tells
consumers of Stasis that the creation of this channel is an implementation
detail in Asterisk and can be ignored (if they so choose). This
consolidates the conference recorder/announcer flags as well - these flags
had no additional meaning beyond "ignore this channel please".
2. It modifies allocation of a channel in two ways:
(a) If a channel technology can be determined from the name, we set it
directly in the allocation routine. This prevents the initial
publication of the message from going out with a NULL channel technology
where possible. This lets Stasis consumers get the right channel
technology on the first publication.
(b) It reorganizes allocation to make use of the 'finalized' property on the
channel. This was already used to know that a channel had completely
finished its construction in the masquerade routine; now we also use it
to know whether or not the setting of certain channel properties is
occurring during or post construction. The various set routines were
modified accordingly as well.
3. The masquerade event is now dead, Jim. It no longer served any purpose
whatsoever - if you perform a call pickup you'll get a Pickup event;
if you perform an attended transfer you will still get those events; if you
steal a channel to put it elsewhere you'll get the corresponding NewExten or
BridgeEnter events.
Review: https://reviewboard.asterisk.org/r/2740
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@396392 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This patch implements the controls from ARI recordings. The controls
are:
* DELETE /recordings/live/{recordingName} - stop recording and
discard it
* POST /recordings/live/{recordingName}/stop - stop recording
* POST /recordings/live/{recordingName}/pause - pause recording
* POST /recordings/live/{recordingName}/unpause - resume recording
* POST /recordings/live/{recordingName}/mute - mute recording (record
silence to the file)
* POST /recordings/live/{recordingName}/unmute - unmute recording.
Since this underlying functionality did not already exist, is was
added to app.c by a set of control frames, similar to how playback
control works. The pause/mute control frames are toggles, even though
the ARI controls are idempotent, to be consistent with the playback
control frames.
(closes issue ASTERISK-22181)
Review: https://reviewboard.asterisk.org/r/2697/
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@396331 65c4cc65-6c06-0410-ace0-fbb531ad65f3
The new res_ari_asterisk.so module presents several config options
from asterisk main. Unfortunately, they aren't exported, so the module
won't load on Linux.
This patch renames the variables, adding the ast_ prefix so they will
be exported.
Review: https://reviewboard.asterisk.org/r/2737
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@396166 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This commit is smaller than the initial review placed on review board. This is because
a change to allow for channel drivers to access parking functionality externally was
committed and invalidated quite a few of the changes initially made.
(closes issue ASTERISK-22039)
reported by Matt Jordan
Review: https://reviewboard.asterisk.org/r/2717
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@396103 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This patch renames the bridging* files to bridge*. This may seem pedantic
and silly, but it fits better in line with current Asterisk naming conventions:
* channel is not "channeling"
* monitor is not "monitoring"
etc.
A bridge is an object. It is a first class citizen in Asterisk. "Bridging" is
the act of using a bridge on a set of channels - and the API that fulfills that
role is more than just the action.
(closes issue ASTERISK-22130)
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@395378 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This patch does the following:
* It pulls out bridge_channel and puts it into its own translation unit
* It adds public and protected headers for bridging_channel. Protected
functions are appropriate only for the Bridging API and sub-classes of a
bridge.
(issue ASTERISK-22130)
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@395253 65c4cc65-6c06-0410-ace0-fbb531ad65f3
In previous versions of Asterisk, the zombies roamed freely,
unchecked and uncontrolled. They ravaged Asterisk systems with
their biting and their nashing and their pointy teeth.
Sometimes, you couldn't even hang them up.
Now, zombies are rare. They still *technically* exist in certain
places, but they are controlled. Kind of like a zombie zoo: you can
see them, but you can't touch them, and they can't touch you.
Bring your kids!
Because zombies are now population controlled with a very short lifespan,
there's no reason to rename the channels to '%s<ZOMBIE>'. The channels
are guaranteed to die off quickly; the rename really is just confusing
at this point.
This patch finally removes the renaming. On the plus side: this made
my life easier in CDRs during call pickup and attended transfers to
an Asterisk application. It will make other folks lives easier as well!
Review: https://reviewboard.astierks.org/r/2690/
(closes issue ASTERISK-21699)
Reported by: Matt Jordan
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@395135 65c4cc65-6c06-0410-ace0-fbb531ad65f3
This adds CEL support for blind and attended transfers and call pickup.
During the course of adding this functionality I noticed that
CONF_ENTER, CONF_EXIT, and BRIDGE_TO_CONF events are particularly
useless without a bridge identifier, so I added that as well.
This adds tests for blind transfers, several types of attended
transfers, and call pickup.
The extra field in CEL records now consists of a JSON blob whose fields
are defined on a per-event basis.
Review: https://reviewboard.asterisk.org/r/2658/
(closes issue ASTERISK-21565)
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@394858 65c4cc65-6c06-0410-ace0-fbb531ad65f3
* Made ast_audiohook_detach_list() and ast_audiohook_write_list_empty()
NULL tolerant.
* Made ast_audiohook_detach_list() return void since it is a destructor.
git-svn-id: https://origsvn.digium.com/svn/asterisk/trunk@394836 65c4cc65-6c06-0410-ace0-fbb531ad65f3
Matthew Jordan
Kinsey Moore
Joshua Colp
Richard Mudgett
Mark Michelson
Scott Griepentrog
Kevin Harwell
Russell Bryant
Rusty Newton
David M. Lee
Walter Doekes