Open Bug 1392396 Opened 3 years ago Updated 3 years ago

Intermittent browser/components/migration/tests/unit/test_IE_bookmarks.js | - Got all the expected parents - 1 == 0

Categories

(Firefox :: Migration, defect, P5)

defect

Tracking

()

Tracking Status
firefox57 --- disabled
firefox58 --- disabled

People

(Reporter: intermittent-bug-filer, Unassigned)

Details

(Keywords: intermittent-failure, leave-open, Whiteboard: [stockwell disabled])

Attachments

(1 file, 1 obsolete file)

the failure rate continues, onto week 3:
https://brasstacks.mozilla.com/orangefactor/index.html?display=Bug&bugid=1392396

this is specific to windows7 on all configurations.

:mattn, I see you are the triage owner for Firefox::Migration- could you help prioritize this bug and pick up fixing it or find someone who could fix it when reasonable?
Flags: needinfo?(MattN+bmo)
Whiteboard: [stockwell needswork]
Pushed by gbrown@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/18eaf95d0d96
Skip xpcshell test_IE_bookmarks.js on Windows for 200+ failures; r=me,test-only
Keywords: leave-open
Whiteboard: [stockwell disable-recommended] → [stockwell disabled]
Geoff, was there a change to the win7 machines we use for tests around the time this started (end of august, roughly speaking) ? The odd thing is it showed up on inbound, beta and  autoland at roughly the same time. That's very surprising unless this is caused by an infra change. Maybe we switched to different machines to run tests on or something?
Flags: needinfo?(gbrown)
:pmoore, did the win7 vm we use in taskclsuter change around the end of august or early september?
Flags: needinfo?(gbrown) → needinfo?(pmoore)
Also, could we disable the test only on Win7? I'm not sure why the disabling was done globally when the problem only affected one specific version
(In reply to :Gijs from comment #12)
> Geoff, was there a change to the win7 machines we use for tests around the
> time this started (end of august, roughly speaking) ? 

I don't know of anything. pmoore may be more definitive.

> The odd thing is it
> showed up on inbound, beta and  autoland at roughly the same time. 

I don't see that. First autoland was August 21st, first beta was the 27th, right?
(In reply to Marco Bonardo [::mak] from comment #14)
> Also, could we disable the test only on Win7? I'm not sure why the disabling
> was done globally when the problem only affected one specific version

It was already skipped on !win, but it looks like this could be run on Windows 10. I was perhaps heavy handed there...I can add back Windows 10.
(In reply to Geoff Brown [:gbrown] from comment #16)
> It was already skipped on !win, but it looks like this could be run on
> Windows 10. I was perhaps heavy handed there...I can add back Windows 10.

Yes, other platforms don't have IE. Win10 looks good indeed and we'd not lose precious testing.
Attached patch enable on windows 10 (obsolete) — Splinter Review
Is this the best way to enable on Windows 10? (I hate to base annotations on specific version strings).
Attachment #8914864 - Flags: review?(jmaher)
Comment on attachment 8914864 [details] [diff] [review]
enable on windows 10

Review of attachment 8914864 [details] [diff] [review]:
-----------------------------------------------------------------

::: browser/components/migration/tests/unit/xpcshell.ini
@@ +18,4 @@
>  skip-if = os != "win"
>  [test_fx_telemetry.js]
>  [test_IE_bookmarks.js]
> +skip-if = os != "win" || bits == 32 # bug 1392396

I would think:
skip-if = os == "win" && bits == 32

this would be win32 only
Attachment #8914864 - Flags: review?(jmaher) → review-
Discussed on irc...we came up with this.
Attachment #8914864 - Attachment is obsolete: true
Attachment #8914884 - Flags: review+
Pushed by gbrown@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/ba9aea792718
Re-enable test_IE_bookmarks.js on Windows 10; r=jmaher
Rob, see comment 13 onwards. Thanks!
Flags: needinfo?(pmoore) → needinfo?(rthijssen)
(In reply to Geoff Brown [:gbrown] from comment #15)
> (In reply to :Gijs from comment #12)
> > Geoff, was there a change to the win7 machines we use for tests around the
> > time this started (end of august, roughly speaking) ? 
> 
> I don't know of anything. pmoore may be more definitive.
> 
> > The odd thing is it
> > showed up on inbound, beta and  autoland at roughly the same time. 
> 
> I don't see that. First autoland was August 21st, first beta was the 27th,
> right?

Ah, hm, it appeared in the same report, I didn't look that closely. Still, that's surprising, because this was the 56 beta cycle still, so it wasn't like we were uplifting that much. This is the list of patches that would potentially fit that bill:

https://hg.mozilla.org/releases/mozilla-beta/pushloghtml?startdate=2017-08-21&enddate=2017-08-28

and most of that looks like form autofill. None of these patches (checked via:

hg diff browser/components/migration/ toolkit/components/places/ -r 32b732560adc -r bceb296ce64f --stat

)

touched the test or any of the immediately-relevant files for it (the IE migrator, or places code) at all. This, and the fact that the test interfaces with external windows stuff like bookmarks from IE, is why I'm suspicious there was an infra change that might have caused this. Perhaps one that went live for central-based repos before it did for beta? :-)
windows 7 and 10 instances under taskcluster are shared across all repos. the same workertypes run tests for m-b and m-c, etc so if it were an infra issue, we should see failures across all trees at the same time.

the history for gecko-t-win7-32 shows the last change to the ami was on june 29th.
https://github.com/mozilla-releng/OpenCloudConfig/commits/master/userdata/Manifest/gecko-t-win7-32.json
Flags: needinfo?(rthijssen)
Flags: needinfo?(MattN+bmo)
You need to log in before you can comment on or make changes to this bug.