connection failure fallback

[marknelissen]

Hi,

Is there some kind of failover included in the library in case the connection and/or registration fails? Does the library retry to connect/register, or does this need to be piloted externally? If so, how is the best way to interact with the cycle to renew the connection/registration?

What happens on connection interruption? Does the library reconnect automatically on connection loss to the sip server? How to detect this, and fix it manually, if it does not?

I'm talking about the connection to the sip server, not to another sip client.

Regards,

Mark Nelissen

[siniypin]

Hi Mark,

long story short - there is no reliable way to get an instant notice that a connection was broken.
I mean, in general there is no way one can reliably tell whether the connection persists with ms precision, especially within wireless networks.

Long story:

1. SIP is a transport agnostic protocol, meaning that one can 'connect' to SIP server using either UDP or TCP transport protocols.
2. UDP and TCP differ significantly. With UDP transport one have no solid connection, rather a loose flow of outgoing and incoming datagram messages. Their sequence form a virtual connection state. Their timely arrival is a sign of a healthy connection. Matters get a little bit easier with TCP, as there is some sort of solid connection both ends can write to. However, even with TCP one should not expect an instant notice of a broken connection as there still is some sort of ping pong present on a kernel level.

So having that in mind I decided to check several theories:

• SIP over UDP would not notice a server downtime until a registration timeout;
• SIP over TCP should notice a server downtime almost instantaneously and report an account registration state change;
• SIP over UDP with keep alive message should notice a server downtime within a keep alive interval;

So I setup a VM with a FreeSwitch server to connect to and performed tests using the pjsip4net.Console app that comes with a source code. That's what I found:

• [x] SIP over UDP does not notice a server going down indeed - there is nothing to comment here. It is just as it should be.
• [ ] SIP over TCP notices a TCP connection closed, however it raises no callbacks till the registration timeout. - That seems a bug to me. It isn't much surprising if you consider that a current pjsip library version is 2.4.5 as opposed to version 1.4 pjsip4net uses.
• [ ] SIP over UDP with keep alive message seem to have no difference to regular SIP connection. - It seems to be a bug to me as well.
• [ ] Using SIP over TCP I also checked that neither keep alive nor publish presence settings have any effect on behaviour in question.

Nevertheless, there are options:

• SIP over UDP: If you don't need a ms precision you can setup a short registration timeout, e.g. 10 sec. That would mean that within configured_reg_timeout+time_it_takes_to_figure_out_server_is_down seconds you'll get a callback notifying you lost a connection. time_it_takes_to_figure_out_server_is_down usually takes another 10-15 seconds (not configurable in 1.4 version). With relatively small number of accounts that might be a feasible solution;
• SIP over TCP: even though a library does not raise an expected callback as soon as it notices a TCP socket had been closed it does send a log message that can be intercepted leadinf to certain actions to be applied. An example message: 2016-02-13 23:54:17,478 [11] DEBUG - 23:54:17.472 tcps0921AE64 TCP transport destroyed with reason 70016: End of file (PJ_EEOF). And this is how you can handle it: https://gist.github.com/siniypin/a20f0cf37bfb7cbc24b9. I know it is hackish at best, but this is the best you can get if you are required to diagnose disconnects almost instantly.

P.S. I shall encapsulate this logic within 1.4 version wrapper shortly, for everybody to take advantage of fast feedback with TCP transport. Mind it isn't real-time anyways.

P.P.S. You also asked whether library automatically reconnects whenever it looses a connection - it doesn't. Newer versions of pjsip do. So having a wrapper on top of the latest version of pjsip would solve that issue for you. Otherwise you have to do it on your own.

[siniypin]

Reconnecting means recovery from a disconnect. Pjsip periodically refresh registration with healthy connection, so you don't have to worry about that.

[marknelissen]

Hi,

Thank you very much for the detailed and in-depth explanation, as well as the time you are pouring into this.

I don't need ms precision, it is sufficient to know when the registration is not renewed, because of some error. However, it is not very clear to me which callback I should be registering on the wrapper to trigger another registration attempt manually. I did try some stuff, but I ended up triggering a new registration attempt, while the first was still busy. I did this by monitoring the state of the wrapper, which seems to not be the correct way of doing it, because it does not show when the registration is still trying, or else I checked the wrong state.

The use case I'm covering, is that the sip module is only part of a larger application on a public station. As such, the global application should not have to be restarted to recover from connection issues, since the end user will not have sufficient control to perform this operation. I want to put some watch dog function that automatically triggers reconnection (through registering again) in case the registration expires and fails to renew.

Also, are there any plans of upgrading the wrapper to the newer version of pjsip? Is the API much different? I barely understand how the current wrapper works, thus preventing me from offering you a pull request with the upgrade, at least right now.

[siniypin]

Hi,

it looks like a simple AccountStateChanged (https://github.com/siniypin/pjsip4net/blob/master/pjsip4net/Interfaces/IAccountManager.cs#L17) notification would be sufficient in your case. Objects of AccountStateChangedEventArgs class have a StatusCode property where values >= 400 mean something gone wrong.

Unfortunately, I have no time to add support for any newer versions of pjsip.

Cheers

[siniypin]

Hi Mark, did that work out for you?

[marknelissen]

Hi,

I haven't had time to work on the project again, but I will soon test it. I'll keep you informed of the result.

[marknelissen]

Hi,

I've finally gone for the solution to upgrade to the newer version 2.4.5 of pjsip, using the pjsua swig interface. The library is handling the reconnection quite nicely. The documentation is complete, but there are some stuff you need to manually do, which should for me be handled by the wrapper. As such, your wrapper would still offer added value on top of this interface. I hope you get around one day to upgrade the wrapper to support the newer version of pjsip.

Thank you for all the time you spend helping me.

[siniypin]

Hey Mark, good news.

I'd appreciate you sharing your SWIG bindings to pjsip, so anyone (including me) could simply adapt pjsip4net interface to pjsip 2.x.

Cheers

[marknelissen]

I've followed the instructions of MikeGawi on issue 15. Although pjsip natively only has the necessary configurations for java and python, there is only minor adaptation to be done of the generated files.

I only discovered the project at https://github.com/StrikeForceZero/PJSip-CSharp after having done all the work. I don't know how far along it is, and whether it works.

[marknelissen]

pjsua2api.zip contains the swig bindings I used. The documentation at www.pjsip.org/docs/book-latest/html/index.html is rather helpful on how to use the api, even though you have to translate the example code from c++ to c#.

[siniypin]

Awesome, Mark. Thank you very much! ᐧ

Best regards, Boris Tveritnev de.linkedin.com/pub/boris-tveritnev/11/3a4/113/ www.xing.com/profile/Boris_Tveritnev

On 16 March 2016 at 12:53, marknelissen notifications@github.com wrote:

pjsua2api.zip https://github.com/siniypin/pjsip4net/files/175879/pjsua2api.zip contains the swig bindings I used. The documentation at www.pjsip.org/docs/book-latest/html/index.html is rather helpful on how to use the api, even though you have to translate the example code from c++ to c#.

— You are receiving this because you commented. Reply to this email directly or view it on GitHub https://github.com/siniypin/pjsip4net/issues/43#issuecomment-197281369