Closed Bug 637838 Opened 10 years ago Closed 10 years ago

Android logs uploaded to the wrong place

Categories

(Release Engineering :: General, defect, P3)

All
Android
defect

Tracking

(Not tracked)

RESOLVED FIXED

People

(Reporter: mstange, Assigned: aki)

References

Details

(Whiteboard: [json][automation])

Attachments

(4 files)

I haven't found logs for Android builds. For example, in http://build.mozilla.org/builds/builds-2011-03-01.js.gz there's a build with id 3818821 from the "Android Tegra 250 mozilla-central opt test browser-chrome" builder, whose log I'd expect at http://ftp.mozilla.org/pub/mozilla.org/mobile/tinderbox-builds/mozilla-central-android-r7/1298992873/mozilla-central_tegra_android_test-browser-chrome-build24.txt.gz - but it's not there.
These logs are actually uploaded, but in the wrong place:
http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/mozilla-central-android-r7/1298992873/
Summary: No logs uploaded for Android runs → Android logs uploaded to the wrong place
Related to bug 636704, or at least fixable at the same time?
OS: All → Android
Priority: -- → P3
See Also: → 636704
Whiteboard: [json] → [json][automation]
I will be handling at least the Try portion of this during bug 655046 (duped bug 636704 to that).
Assignee: nobody → aki
Attached patch test log configsSplinter Review
Add stage_platform to mozilla-tests/config.py.
Also remove branch-extra, which was used to send Tegra test results to the mobile tinderbox page.
Attachment #536687 - Flags: review?(jhford)
Attached patch test log customSplinter Review
Same.
This also involves generateBranchObjects changes, since there are old branch unittests on there. All generateTestBuilder calls are updated with stageProduct and stagePlatform. generateCCBranchObjects calls generateCCTestBuilder, so we're good here.

Also hardcoding generateMobileBranchObjects' stage_product to 'mobile' until we remove this code.
Attachment #536688 - Flags: review?(jhford)
Comment on attachment 536687 [details] [diff] [review]
test log configs

Review of attachment 536687 [details] [diff] [review]:
-----------------------------------------------------------------

lgtm
Attachment #536687 - Flags: review?(jhford) → review+
Comment on attachment 536688 [details] [diff] [review]
test log custom

Review of attachment 536688 [details] [diff] [review]:
-----------------------------------------------------------------

Aside from the missing kwarg, I think this is what we need.  It would be good to see firefox/fennec things shift from substring matching of the platform to being keyed off stage_product.  That scope creep for this bug, so not blocking r+.

::: misc.py
@@ +3055,5 @@
>  
>      logUploadCmd = makeLogUploadCommand(name, config,
>              is_try=config.get('enable_try'),
> +            is_shadow=bool(name=='shadow-central'),
> +            product_prop='product')

This is missing a kwarg of |platform_prop="stage_platform"|.  Without it, it will upload to linux-android instead of android
Attachment #536688 - Flags: review?(jhford) → review-
Attachment #536688 - Flags: review- → review+
(In reply to comment #7)
> Comment on attachment 536688 [details] [diff] [review] [review]
> test log custom
> 
> Review of attachment 536688 [details] [diff] [review] [review]:
> -----------------------------------------------------------------
> ::: misc.py
> @@ +3055,5 @@
> >  
> >      logUploadCmd = makeLogUploadCommand(name, config,
> >              is_try=config.get('enable_try'),
> > +            is_shadow=bool(name=='shadow-central'),
> > +            product_prop='product')
> 
> This is missing a kwarg of |platform_prop="stage_platform"|.  Without it, it
> will upload to linux-android instead of android

[12:49]	<aki>	jhford: the chunk you're referring to is in generateMobileBranchObjects; I think the platform will stay 'android-r7' there.
[12:50]	<aki>	splinter helpfully removes the function name =P
[12:50]	<jhford>	aki: yes, it does :(
This made it to production today.
Status: NEW → RESOLVED
Closed: 10 years ago
Resolution: --- → FIXED
Status: RESOLVED → REOPENED
Resolution: FIXED → ---
1) remove enable_opt_unittests from project_branches.py

This probably worked in mozilla-tests/, but it's also linked in mozilla/, so that enabled old-style, on-buildslave opt unit tests for Android on Tracemonkey.

I moved it to below the project branch loop in mozilla-tests/config.py.

2) As part of sending mobile results to the Firefox tbpl/tinderbox pages, I changed the mobile_tinderbox_tree for mozilla-central to 'Firefox' in mozilla-tests/production_config.py.
Attachment #536973 - Flags: review?(bhearsum)
John Ford was using the properties 'stage_platform' and 'product' in makeLogUploadCommand in generateBranchObjects.

I did the same thing in generateTalosBranchObjects, but I used 'stage_product' and 'stage_platform'.  However, because of old-style, on-buildslave unit tests, generateBranchObjects called generateTestBuilder as well, and the makeLogUploadCommand in generateBranchObjects was expecting 'product', not 'stage_product'.

I've renamed all 'stage_product' properties to 'product' for consistency.

I also wasn't sure if the release makeLogUploadCommand would be affected by the changes this week, so I added a |product=releaseConfig['productName']| to the makeLogUploadCommand in generateReleaseBranchObjects.

Everything clear? =P
Attachment #536975 - Flags: review?(bhearsum)
Blocks: 661749
Attachment #536973 - Flags: review?(bhearsum) → review+
Attachment #536975 - Flags: review?(bhearsum) → review+
Comment on attachment 536975 [details] [diff] [review]
change 'stage_product' to 'product' in properties list

Landed this on default.
Attachment #536975 - Flags: checked-in+
Attachment #536973 - Flags: checked-in+
These hit production today.
Status: REOPENED → RESOLVED
Closed: 10 years ago10 years ago
Resolution: --- → FIXED
Product: mozilla.org → Release Engineering
You need to log in before you can comment on or make changes to this bug.