UA recovers transport even after stopping

[wakamoleguy]

Thanks to Kair on the mailing list for finding this one. In the event of a network failure (e.g., turning Wifi off), the UA continues attempting to recover the transport, with no (good) way of breaking the cycle. Even stopping the UA does not prevent it from trying.

I'm guessing the fix for this is either to stop recovering the transport if the UA is stopped, although there were several other quirks about the network recovery behavior.

[josephfrazier]

I haven't been able to reproduce this on the master branch (e2ab986). Here's how I tried in Chrome/Firefox:

window.ua = new SIP.UA({
  traceSip: true
});

// (turn Wifi off)

ua.register() // triggers recoverTransport after about 20 seconds

// wait for "WebSocket connection error" to be logged, then run:
ua.stop()

// recoverTransport is no longer called

I haven't looked much into the other quirks mentioned on the Google Group, but I think we should have separate issues for them.

[capilkey]

I have a jsfiddle that shows the continuing connect attempts after calling UA.stop(). It has a linked sip.js that is a build of the master branch (same commit as above). A couple of things are different about what I do to reproduce it though. I don't disconnect my ethernet cable and I'm using UA.start() instead of UA.register(). There's a listener that calls UA.stop() when the 'disconnected' event is first fired.

http://jsfiddle.net/kcrfefxe/20/

[josephfrazier]

Thanks for the jsfiddle, @capilkey! Using it, I was able to reproduce the problem, and I'll keep looking into it.

[josephfrazier]

I've pushed f68e2f6, which should stop recovering the transport if the UA is stopped. It works against @capilkey's jsfiddle's code, with the caveat that the configuration variable was missing the autostart: false required to prevent 'disconnected' from being emitted before its listener is added.

[wakamoleguy]

If I'm reading this right, disconnected is emitted synchronously if the autostart fails? That's not great.

[wakamoleguy]

I've confirmed that disconnected is not firing synchronously in that case. The listeners are all set up properly in the example, except that the gUM call delays the test until after the disconnected event has happened. Not a SIP.js bug, just a glitch in the test case.