1196008 - test_errorhandler.js seems to take a long time and time out often

Status: RESOLVED FIXED

(Firefox :: Sync, defect, P2)

Description

[William Lachance (:wlach)]

services/sync/tests/unit/test_errorhandler.js, which seems to be testing something to do with sync, seems to time out fairly often (and takes a long time even when it does pass). E.g. do a search for `test_errorhandler.js` here:

http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/mozilla-inbound-linux64-debug/1439913153/mozilla-inbound_ubuntu64_vm-debug_test-xpcshell-1-bm120-tests1-linux64-build63.txt.gz

(but you can see the behaviour with just about any log)

This seems to me to indicate there's either (1) something wrong with the test or (2) perhaps that it should be split up into smaller units (so it doesn't time out and we can take better advantage of parallelization).

:markh, :rnewman: Sorry to bother you, but it seems like you were the last people to touch this file... is there anything you can do here?

Comment 1

[Mark Hammond [:markh] [:mhammond]]

Last I looked into this the issue seemed to be that the test spent alot of time waiting for DNS to fail. I agree we should fix this - you mention it often times out - there must already be a bug for that on file, right?

Comment 2

[William Lachance (:wlach)]

(In reply to Mark Hammond [:markh] from comment #1)
> Last I looked into this the issue seemed to be that the test spent alot of
> time waiting for DNS to fail. I agree we should fix this - you mention it
> often times out - there must already be a bug for that on file, right?

I don't know of one, I searched for the test name in the title of this bug on bugzilla and didn't find anything. It's possible we hadn't noticed the timeouts so far as we retry running the test again in the case that it happens when running other stuff in parallel.

Comment 3

[Mark Hammond [:markh] [:mhammond]]

Attachment: 0001-Bug-1196008-have-test_errorhandler.js-use-a-real-ser.patch

The test as written has many entries like:

> 0:04.85 PROCESS_OUTPUT: Thread-1 (pid:17728) "1441269804386	Sync.Tracker.Clients	DEBUG	Saving changed IDs to clients"
> 0:09.09 PROCESS_OUTPUT: Thread-1 (pid:17728) "1441269808627	Sync.Service	DEBUG	verifyLogin failed: Error: NS_ERROR_UNKNOWN_HOST ...

Which is due to:

> const FAKE_SERVER_URL = "http://dummy:9000/";

I changed the test to use a "real" server (with no handlers, so each request still fails to offer a sane response)

Locally the runtime goes from 1.5 minutes to 14s.

There are 2 other tests with similar problems, but they want a "real" network error and only take ~12s each currently - with effort they could also be reduced but I haven't the time to dig into them.

Comment 4

[Pulsebot]

https://hg.mozilla.org/integration/fx-team/rev/2cf5967652d1

Comment 5

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/2cf5967652d1