1177921 - ice_unittest has numerous failures when not running in CI

Status: RESOLVED FIXED

(Core :: WebRTC: Networking, defect, P2)

Description

[Byron Campen [:bwc]]

The tests that require an external STUN server (and are therefore disabled in CI) are extremely unreliable now. It seems that either twilio's STUN servers are broken, or there is some interop problem.

Comment 1

[Eric Rescorla (:ekr)]

0. Can you post an error report here?
1. What happens if you use Googles's STUN servers?

Comment 2

[Byron Campen [:bwc]]

Google's STUN server and our STUN server behave similarly, but still lead to failures since they do not support TCP. Using twilio's STUN server leads to lots of timeouts when trying to gather. I'm going to try increasing the timeout.

Comment 3

[Byron Campen [:bwc]]

Extending timeouts doesn't work. Also, I'm seeing lots of failures in tests that do non-default things like disabling ICE TCP, or enabling loopback, because we now init an NrIceCtx (with the default settings) to back the test STUN TCP server. This inits the registry with the defaults, so subsequent setting don't stick anymore. I have no idea how these tests are passing on CI...

Comment 4

[Byron Campen [:bwc]]

Oh wow, MOZ_WEBRTC_TESTS=1 is no longer set in CI, so the test doesn't run at all...

Comment 5

[Byron Campen [:bwc]]

Huh, we are running this test on OS X, and it works there. This is a real mess.

Comment 6

[Maire Reavy [:mreavy]]

Given the other work on our plates (and until I hire more folks), I think we need to move this down to a P2.  I'll do the same for bug 1178349.

Comment 7

[Eric Rescorla (:ekr)]

Attachment: Fix typo in STUN server name

Comment 8

[Eric Rescorla (:ekr)]

Attachment: MozReview Request: Bug 1177921 - Fix typo in STUN server name. r?drno

Bug 1177921 - Fix typo in STUN server name. r?drno

Comment 9

[Eric Rescorla (:ekr)]

(In reply to Byron Campen [:bwc] from comment #2)
> Google's STUN server and our STUN server behave similarly, but still lead to
> failures since they do not support TCP. Using twilio's STUN server leads to
> lots of timeouts when trying to gather. I'm going to try increasing the
> timeout.

At least part of the problem here is the typo I just fixed. The tests seem to run cleanly on my local machine now.

Comment 10

[Eric Rescorla (:ekr)]

With the above patch, I just successfully ran the tests 10x on my Linux box.

Comment 11

[Byron Campen [:bwc]]

Comment on attachment 8642031 [details]
MozReview Request: Bug 1177921 - Fix typo in STUN server name. r?drno

https://reviewboard.mozilla.org/r/14675/#review13291

Ship It!

Comment 12

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/03f8c335d0e0

Comment 13

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/03f8c335d0e0