AstriDevCon was held on at the Omni Orlando Resort at Championsgate near Orlando, FL. There were approximately 35 attendees on average throughout the day. Lunch was provided by e4strategies.com.
- Matt Fredrickson, Digium, US
- Josh Colp, Digium, US
- George Joseph, Digium, US
- Malcolm Davenport, Digium, US
- Kevin Harrell, Digium, US
- Ben Ford, Digium, US
- David Duffett, Digium, US
- Matt Jordan, Digium, US
- Sean Pimental, Digium, US
- Kyle Kurz, Digium, US
- Torrey Searle, Voxbone, BE
- Gabriel Gontariu, Voxbone, BE
- Clod Patry, Jive, CA
- Lorenzo Emilitri, Loway, CH
- Dan Jenkins, Nimble Ape, UK
- Nir Simionovich, Greenfield, IL
- Eric Klein, Greenfield, IL
- Sean McCord, CyCore, IL
- Jim Van Meggelen, CA
- Louis-Olivier Roff, Jive, CA
- James Finstrom, Sangoma, US
- Jason Parker, Sangoma, US
- Bryan Walters, Sangoma, US
- Andrew Nagy, Sangoma US
- Evan McGee, HiFelix, US
- Emmanuel Rolon, Organic Farms Vitamins, US
- Daniel Mierla, Miconda, DE/RO
- Fred Posner, Qxork, US
- François Blackborn, Wazo, CA
- Sylvain Boily, Wazo, CA
- Ludovic Gasc, ALLOCloud/Eyepea, BE
- Sean Bright, Callshaper, US
- Alex Goodman, Axia Technology Partners, US
- David Al-Khadhairi, USAN, US
- Dan Collins, USAN, US
- Steve Murphy, US
- Corey Farrell, US
- Jared Smith, US
- Corey McFadden, Voneto, US
Notes and highlights
Introductions. See attendees list.
Matt F introduces his background and why he's in the front of the room.
Current releases of 13 and 14 and now 15.0.0.
Asterisk 15 contribution stats: 924 commits, 82 individual contributors, almost 2400 merged code reviews acros all branches on Gerrit in the past 12 months.
Top contributors by # of commits by people outside of Digium:
- 104 Sean Bright, Callshaper
- 42 Corey Farrell
- 39 Alexander Traud
- 20 Alexei Gradinari
- 19 Tzafrir Cohen, Xorcom
- 15 Torrey Searle, Voxbone
- 11 Walter Doekes
- 9 Rodrigo Ramirez Norambuena
- 9 Badalyan Vyacheslav
- 6 Frahaase
- 6 Sebastian Gutierrez
- 6 Michael Kuron
- 5 Daniel Journo
- 4 kkm
- 4 Timo Teras
- 4 Martin Tomec
- 4 Joshua Elson
- 4 Jean Aunis
- 4 Aaron an
What’s new in Asterisk 15?
Miscellaneous Other Improvements and…
Video, WebRTC, and more, Oh My!
GCC 7 fixes
Build fixes for FreeBSD when missing crypt.h
Build fixes for the GNU HURD
Added support to build against BIND8
OpenSSL 1.1 support
Alembic support for MS-SQL
PJPROJECT bundled support is enabled by default
Miscellaneous Other Improvements:
New Asterisk sounds release (1.6)
Google OAuth 2.0 protocol support for XMPP/Motif
Chan_rtp uses ulaw by default now instead of slinear
Binaural audio support patches for confbridge were merged
Video, WebRTC, and more, Oh My!:
Support for RTCP-MUX
‘Webrtc’ endpoint option in res_pjsip.conf
VP9 passthrough support
RTP dynamic payload numbers are now truly dynamic (on a per-call basis)
Extensive work to preserve RTP sequence number gaps / losses across legs in a call (critical for video, makes audio better, too)
ICE interface blacklist optional added to rtp.conf
(Discussion about Sean Bright’s patch on Gerrit for ephemeral keys - that are used in the RTP encryption in DTLS-SRTP.)
(Torrey brought up the point that Kamailio now has ephemeral authentication, as well, so that certificates for authentication can be time-limited, etc.)
Support for more than 32 dynamic RTP payloads now exists.
Abstracted SDP layer was added (and is still being worked on)
Added support within the Asterisk core for multi-audio and multi-video stream media per ast_channel
Added support within the Asterisk core to renegotiate media capabilities on an active call as required
Support for BUNDLE was added
SFU support in app_confbridge
(some discussion about SFU and MCU and the tradeoffs between them)
Asterisk 11( LTS) was released in October of 2012
Asterisk 12 was released in December of 2013
Asterisk 13 (LTS) was released in October of 2014
Asterisk 14 was released Monday, Sep 26 of 2016
Asterisk 15 was released Tuesday, October 3rd of 2017
Asterisk 16 is the next LTS target. There is a lot of additional work that needs to go into the video capabilities of Asterisk 15 before we want to support it as an LTS. The video work in Asterisk 15 is a great MVP, but it needs more functionality to be useful for years to come. So, many changes will occur in 15 towards the goal of 16 as the next LTS.
Chrome decided to require an additional flag be passed in to interoperate with legacy endpoints that lack support for RTCP-MUX in January/February of this year
Dan Jenkins informed the Asterisk project of this issue around that time
RTCP-MUX support was implemented at around that time frame to deal with a potential end of life of that behavior
RTCP-MUX support was merged into Asterisk 13 and 14 branches
Chrome is supposed to completely remove support for RTCP-MUX at sometime around the October timeframe.
11 was already in security-fix only mode and is going to be completely dead in October. Get off that branch! (particularly if you run WebRTC)
Now, Joshua Colp and Kevin Harwell to talk about video SFU in Asterisk.
But first, recognition to those that have built tests for the Asterisk test-suite in the past year.
Now, actually, Josh and Kevin.
Asterisk 15 and video.
Old Media Flow
New Real STreams
Old Media Flow
Single logical flow internally carrying media
Each media frame has a type and format
Conceptually, only 1 stream of each type is possible
Negotiated media formats are all combined together
Flow of a single type of media
Can be one way or two way
Has a name which can have meaning
Can be added, removed, or changed
Has negotiated media formats specific to the stream
New Real STreams
First class stream object
Contains only information specific to the stream
Groups of streams are kept in a container called a topology, indexed based on position number
Channel can have stream added, removed, or changed
Each media frame has a type, format, and stream number
Constructs streams according to negotiated result
Responsible for placing stream topology on channel - not done automatically
Responsible for responding to requests
Only channel driver supporting multiple streams currently
Outgoing uses requested stream topology, adding streams to SDP
Incoming negotiates streams based on configured formats
Can be told to renegotiate to add/remove/change streams
PJSIP has a hard limit right now of 16 streams; you’d have to recompile pjproject in order to change that number
Extended version of the Echo() application
Will request renegotiation to ensure specified number of streams are present
Echoes media receives on first stream of each type to every other stream of that type
See and interact with only a single pipe like before
Can have only 1 stream of each type
Existing APIs create streams automatically as appropriate
Does not have any knowledge of new stream support
Ast_read, ast_write, ast_channel_nativeformats
Required no code changes to legacy useres
Legacy Video Support
Calling between devices (if video is in the initial offer)
Basic video recording
Basic video playback
Conference with single video sent to each participant
Currently two bridging modules support multistream:
Softmix (What confbridge uses)
Other bridge modules, e.g. bridge_native_rtp, unchanged and behave the same
How Simple Bridging Now Works
Channel with fewer stream renegotiated to match other
IF same number then second channel joined gets renegotiated to match first channel that joines
Each stream is mapped 1 to 1
Acts as media forwarder based on stream number
Multiple video streams can now be sent to participants (SFU)
Selective forwarding Unit
Picks a subset of video streams to forward
Currently limited by max number of video streams on channel
No server side transcoding or manipulation is done
In the future, additional policy choices will probably exist.
How Softmix Now Works
Each video stream on a channel is mapped to a bridge specific stream number
Each channel can have a mappping from bridge specific stream number to channel video stream
Audio is still mixed server side to provide same ConfBridge audio experience as previously
Enabled using video_mode=sfu in ConfBridge
Best option for rich ConfBridge SFU experience
Required for Google Chrome to support multiple streams due to Plan B usage
Specification to allow multiple streams to be sent/received over the same transport
Cuts down on ICE and DTLS negotiation time
Now available in PJSIP
Limited example code available (is on Github, MIT license)
JsSIP based client for use with Asterisk
Adds/removes video as participants join/leave conference
Controls to mute/unmute
Firefox and Chrome supported on desktop
Potential VIdeo Support Additions
Adding/removing video mid-call
Better video recording (into containers) and playback (with multiple streams)
Feedback allowing video quality to change due to bandwidth change
Better handling of packet loss and out of order packets
Putting it All Together
…(notes not compiled for this segment)
(A demo of CyberMegaPhone was done.)
Now, it’s time for planning the Agenda.
- Talk by Wazo
- Talk by Ludovic
- Talk by Daniel
- Talk by Greenfield
- Discussion by Nir
- Proposed deprecation of app_macro
- Proposed deprecation of chan_sip
- How do we get to an all ARI solution?
- Getting features into an LTS
- What’s the next evolution of Asterisk?
- How to improve functions in ARI to make it more of a first class citizen?
AMQP and the Stasis Message Bus
Remove direct connection to AMI (no parsin)
And use AJAM to send actions to AMI
Remove external proxy for ARI
We already talked about this feature at the last AstriDevCon
AMQP client for Asterisk
Based on patch from https://reviewboard.asterisk.org/r/4365
Extracted version to have a first asterisk patch
Based on librabbitmq
We only test with rabbitmq
Configuration is on /etc/asterisk/amqp.conf
It doesn’t nothing, only an AMQP connection
Git clone; make; make install
Publish stasis message to AMQP
Depends on res_amqp
Git clone; make; make install
To test on your Asterisk and get messages
Adapt the exchange on the script
Allows you to subscribe on specific events, e.g. just the status of a Queue.
Integration in Asterisk
Submit to gerrit the res_amqp support
Submit to gerrit the res-stasis-amqp support
Your feedback is welcome!
(Some discussion on why the original patches weren’t merged (lack of tests in the CEL and CDR modules) and about where this would end up if it was merged (16 if no tests, or 13 if tests).)
Now, Ludovic to talk about res_calendar!
Asterisk and the calendars, when non-C developers meet Asterisk+libical
Who am I?
Creator of API-House (Daemon framework for Python-AsyncIO)
Creation or aiosip (used by Sangoma to test their phones)
Co-maintainer of Panoramisk (Asterisk binding for AsyncIO)
Small contributor in several AsyncIO libraries (aiohttp…)
Interested by benchmarks to find the bottlenecks
Contributor of https://www.techempower.com/benchmarks
Most simple as possible
Distributed telephony and collaboration
Efficiency is the first class citizen (1500+ simultaneous calls by server)
Full-monty customized solutions
Solutions mainly based on Wazo
Historical business of the company
Google Calendar/Office365 integrations
There are two steps:
Step 1, define a calendar.
Step 2, put the calendar in the callflow
What now? Icalendar is the most obvious format.
It’s used in a lot of products.
It’s a stable standard
And it’s very old; more chances that there are good implementations
But not really...old != stable.
First, they wrote an implementation using icalendar in Python
It was easy to debug and integrate.
But, libical integration in Asterisk looked like a Proof of Concept during an Astricon.
Very few messages from people using it on the Internet
Lack of examples
Need to dig in the original Astricon presentation to understand how to use the diaplan functions.
(they’re not C-developers)
First client? Crashed immediately.
The first challenge: recurrency; something very common with calendar events.
Most libraries parse recurrency fields, but most don’t interpret correctly recurrency data.
They tested lots of libraries, but libical works best.
Plan B: libical integration in Asterisk
They put it into production, and it worked!
But..then comes the daylight savings time in winter and everything’s going to be thrown for a loop.
Libcal has bugs with timezone and DST.
But they’re fixed in libical3
But, libical1 and libical forks are widely distributed.
Libical3 isn’t released and available in Debian or CentOS.
So, they had to import it manually from libical master branch.
One more bug remains...editing of a recurring event, recurrence-id.
A fix was submitted on Gerrit: ASTERISK-27296 / https://gerrit.asterisk.org/#/c/6625/
For now, they have 973 calendars in production.
Right now, there is no file system support for res_calendar..working on a patch but it has memory leaks.
Might do python bindings for libical.
Next, Daniel will talk about SIP Proxy Router and Media Server PBX; Integrate, Interconnect, Innovate.
There are things that aren’t yet possible, but we worry about how to make things easier.
We could integrate log formatting, and have a common prefix for easy correlation, callid, cseq, etc.
It would improve troubleshooting and unit testing.
We could integrate user profile and database structure
That’d give us unified user authentication, user location, and presence.
We need ad-hoc and realtime propogation of information (by headers) so that we don’t have to always worry about having replicated state across nodes.
This would be done in the forwarded/generated request/response.
Next hop address (route)
...so that we can make configuration easier on people. People can forget to make changes in two places and have negative results today.
Could be done via templates that respect special headers that are sent in the signaling. Templates indicate what sorts of parameters to apply to the call/endpoint.
Don’t always wait for specifications between all parties - IETF, ITU, ETSI, etc.
Collaborate with others to define new features and services, like the XMPP model
Amend, or go around existing specs
Avoid useless roundtrips
Why not allow dynamic new server nodes, because you can’t predict their IPs always (Amazon, for example) that are authenticated via an API key that’s a shared secret between server apps? It’s better than IP and digest-based authentication.
Optimize for mobile and IoT
Get rid of “not needed” headers
Use a server-side app to fetch data, like for hard phones.
Security and Privacy
Auto-provisioning of end points/cpe
Federation and open peering. There’s a lot of FUD out there that suggests SIP isn’t designed for federation. But, if we don’t have a trust model that people will use, then they won’t interconnect, because they’re afraid of bad calls.
(Discussion about Asterisk’s existing, new capability to a allow identification of an incoming request by token, as well as discussion around how to pass a SIP call identifier around the internals of Asterisk - today, you can’t.)
(Discussion about the upcoming implementations of SHAKEN/STIR that are going to be mandated on many carriers.)
(Summary of discussion: It would be good to see work in the area of authentication between services.)
What about knowing what pool of servers is available? DNS is currently used, for round-robin environments. What’s the right way in Kamailio though to make it intelligently aware there has been an expansion or contraction of a pool of available media (Asterisk) servers?
Kamailio has a module called RTJSON that allows pushing JSON into Kamailio to tell it about new routes.
(More discussion of the sharing of state and the dangers of replication - you’re only as good as your weakest server.)
Make our community greater.
Wanted to talk about deployments and containers, but that shouldn’t be the topic. Instead, we need to talk about our community.
We write good code (or phenomenal bugs)
We write good tests (or at least we want to believe it)
But, we suck at...providing proper documentation. Most of it isn’t updated.
We also suck at providing best practices; people are still making the same, old stupid mistakes.
Now, Nir spends his time writing product specs, but still finds that he’s not doing a good job of providing documentation.
Training has actually gotten better in the latest syllabus, so that’s good.
Concrete examples of doing things don’t exist.
Who’s willing to sit for a documentation hackathon? Dialplan, for example, is documented to death, but it’s not documented well enough.
Why aren’t people using it enough? It’s probably documentation.
How do we change the state?
Kamailio has moved to accepting markdown. ARI in Asterisk is Swagger, so there’s no markdown. The docs in Asterisk are in the code in XML and aren’t in a good position to include lots of formatting.
Kamailio, like Asterisk, isn’t missing reference documentation, just examples.
There should be templates for various types of articles: HOWTO, Advanced Guide, Beginner Guide, etc.
Finding information in the Wiki is challenging.
There’s no formal way to make a pull request into the Asterisk repo on Github.
Sometimes, while people are happy to contribute code, because it’s a burden to maintain it, they’re not happy to contribute extensive documentation. How do you get people to contribute, where you’re not paying full-time documentors?
Post-DevCon discussion to be had at the Wine event.
Now, the afternoon topics (at 3:30pm)...
Proposed deprecation of app_macro
Gosub has existed for 12 years now and is a suitable replacement, but not 100% compatible.
You exist a Macro by using Goto to any different context
You normally exit a Gosub using the REturn app which sends control bck to the n+1 priority that originally called Gosub.
A Gosub return address can be thrown out using “StackPop,” then you can use Goto with any context.
Documentation for the DIal app would be simpler if the Macro option were excluded.
Macro adds some code/complexity to the pbx core and a few apps.
(Room is in general consensus that it should be proposed.)
Proposed deprecation of chan_sip
Feature Parity - What features are available in chan_sip that are not available in chan_pjsip and what is the level of effort to get us there?
Configuration - sip.conf vs. pjsip.conf vs. pjsip_wizard.conf vs. contrib/scripts/sip_to_pjsip
Stability - Both actual and perceived
Performance- Both actual and perceived
Outside Forces - Is there a business case for keeping chan_sip around?
What 3 features are missing? CCSS, AOC and outbound Subscriptions. The only one that still gets used is CCSS. Asterisk maintains support for CCSS in the core, and pjproject has the necessary bits to handle it; someone just has to tie them together.
What about stability? The FreePBX community has a large thread with users indicating issues with PJSIP that they don’t experience with chan_sip, but no one is filing bugs or presenting actual issues - it appears to be primarily anecdotal.
Is it already defacto deprecated since it’s in extended support and there is no community maintainer? And, are we being setup for something bad by not making it more clear.
What about configuration? The converter isn’t necessarily feature complete; but is written in Python (hint, non-C developers) There’s built-in help in Asterisk’s CLI (config show help res_pjsip_endpoint.so, for example). Is it worthwhile to make PJSIP read sip.conf? (There are problems here as a friend and a peer are different and if you move that to PJSIP under the hood you can end up with weird configuration or vulnerability issues)
We need a plan of attack. Something in 15 (warning on startup, something else in 16 (noload it and unselect it in menuconfig), and deprecate it in 18. All deprecated modules should probably have a warning on startup. When fully booted list all modules that are deprecated.
How do we get to an all ARI solution?
Worry that users could get themselves into trouble here, because their ARI apps could get into trouble.
Counterpoint against this request is that this can be accomplished with just 3 lines of dialplan - use this Stasis app, vs. pre-setting the ARI app in ari.conf.
Proposal to set stasis=xyz on an endpoint so that an incoming call to an endpoint goes straight to a Stasis app. The room really likes this proposal.
But with a pbx_ari you can just map a dialplan context to a Stasis app.
How can an ARI app know more than just what’s in its own app? “Subscribe all” when connecting the web socket.
Getting features from 14 into an LTS.
Discussion about the implications of 15 as a Standard release instead of an LTS.
What’s the next revolution of Asterisk?
There’s going to be a continued focus on video.
There’s going to be less of a focus on Asterisk as a PBX, and a continued focus on Asterisk as a general purpose media application server, which might be a PBX by the time a developer delivers it to an end user, but might be more like a call center, but could be something else entirely.
How to improve functions in ARI to make it more of a first class citizen?
Setting variables on a bridge
Set or get multiple variables on a channel
List global variables
No variables when a channel is hungup
Easily redirect a channel into Stasis
Should Asterisk be packaged up as a ready-made app for certain purposes?
(Discussion that this is complex and not best served by the core development team)
Everyone's work is appreciated! Thanks for coming!