Perfherder fails to load (stuck at spinner) with `TypeError: l.get(...) is undefined`
Categories
(Tree Management :: Perfherder, defect, P1)
Tracking
(Not tracked)
People
(Reporter: cfallin, Assigned: sclements)
Details
Attachments
(1 file)
When attempting to load this link, I see the following in the JS console:
TypeError: l.get(...) is undefined [Learn More] CompareSubtestsView.jsx:192:7
|
Reporter | |
Comment 1•6 years ago
|
||
Armenzg in #treeherder on Slack asked me to needinfo :sclements for this one.
|
|
Reporter | |
Comment 2•6 years ago
|
||
Upgrading to P1 as this is blocking investigation of a regression (I can't view the subtests passed to me from the build sheriffs).
|
Assignee | |
Comment 3•6 years ago
|
||
I'm looking into it Chris. I was in an important meeting as soon as I came online so I'm only now having a chance to look into it.
|
|
Assignee | |
Comment 4•6 years ago
|
||
For future reference, please upgrade to critical or blocker in severity - we use priority internally.
|
|
Assignee | |
Updated•6 years ago
|
|
|
Reporter | |
Comment 5•6 years ago
|
||
So, a colleague was able to dig out an individual test-results URL so I have what I need to debug the regression; I'm no longer blocking on this so I'm happy if you want to downgrade it (though I suspect it might affect others if it's not a weird corner-case on just this run -- will leave it up to you). Sorry for the priority churn (and thanks for looking into this!).
|
|
Assignee | |
Comment 7•6 years ago
•
|
||
Chris, I'm curious where exactly you got that subtest url from. When I enter the two revisions via the compare chooser (just click on the "compare" tab), these are the test results I see: https://treeherder.mozilla.org/perf.html#/compare?originalProject=autoland&originalRevision=91697065e99f5b88ceaaf92af99ebdbbfc1dda88&newProject=autoland&newRevision=21f755c04005255c5305a13ad9087420e4489b7b&framework=1
None of which have the signature id: 1927927 per the link you posted, which would be used to retrieve the subtests.
:dmajor, can you tell me what link specifically is blocking you and where you got it from?
|
|
Reporter | |
Comment 8•6 years ago
|
||
(In reply to Sarah Clements [:sclements] from comment #7)
Chris, I'm curious where exactly you got that subtest url from. When I enter the two revisions via the compare chooser (just click on the "compare" tab), these are the test results I see: https://treeherder.mozilla.org/perf.html#/compare?originalProject=autoland&originalRevision=91697065e99f5b88ceaaf92af99ebdbbfc1dda88&newProject=autoland&newRevision=21f755c04005255c5305a13ad9087420e4489b7b&framework=1
Here's where I got the link (just tested again now):
- Go to https://bugzilla.mozilla.org/show_bug.cgi?id=1597988
- In first comment, click on: https://treeherder.mozilla.org/perf.html#/alerts?id=24005
- Hover over the "kraken windows10-64-shippable ..." list entry, click on "subtests" link that appears
|
|
Assignee | |
Comment 9•6 years ago
•
|
||
I have a patch to stop the page from loading if there isn't any data from the test results, but investigating why there's a discrepancy between what that alert url is generating and the actual data in the PerformanceDatum model (where the data is retrieved for compare and compare subtests views) will take more time. Could be related to the recent performance/database issues we've been having.
|
||
Comment 10•6 years ago
|
||
|
|
Assignee | |
Comment 11•6 years ago
|
||
I wasn't able to get very far with this investigation. Ionut, you're more familiar with Perfherder and its data than I am. Can you think of why we don't have performance datum for a particular test/revision (per the reassigned alert link in comment 8)? Could this be a pulse ingestion issue?
|
||
Comment 12•6 years ago
•
|
||
I managed to get closer to the culprit. There's a possible bug with the reassignment mechanic.
The kraken windows10-64-shippable...'s subtests hyperlink from this final #24005 summary is:
which differs from that of the original #24002 summary, which is
There's a mismatch between the originalRevision & the newRevision query params.
This latter one is the one which works.
I assume that the subtests hyperlink for each alert (even for reassigned ones) is computed based on the final alert summary. But for reassigned alerts, it shouldn't happen that way; it should be based on their original summary.
|
|
Assignee | |
Comment 13•6 years ago
|
||
I'm not sure I follow. If you look at summary #24005, the first test - kraken windows7-32-shippable opt e10s stylo - has also been reassigned and its originalRevision and newRevision params for the subtests link are same as the test in question - kraken windows10-64-shippable opt e10s stylo - for both that alert and the original alert 24002.
The reason why this link posted in bug 1597988 doesn't work is because there isn't any performance data for revision 21f755c04005255c5305a13ad9087420e4489b7b for kraken windows10-64-shippable opt e10s stylo. Which makes me think we missed ingesting it or somewhere down the line something else happened.
|
|
Assignee | |
Updated•6 years ago
|
|
|
Assignee | |
Comment 14•6 years ago
|
||
:cfallin, :dmajor are you still finding this issue (lack of data) with different pushes/jobs?
|
|
Reporter | |
Comment 15•6 years ago
|
||
I haven't had another regression I've personally needed to debug, but if I search Bugzilla for "regression" and pick a random recent perf alert (e.g., [1]), it seems the links work...
[1] https://treeherder.mozilla.org/perf.html#/alerts?id=23971
|
||
Comment 16•6 years ago
|
||
I've switched away to other tasks so I'm currently not looking at perfherder (so I guess the answer is no, I'm not experiencing problems anymore). Will let you know if I run into the issue again.
|
|
Assignee | |
Comment 17•6 years ago
|
||
Ok, I'll mark this as resolved. Feel free to reopen if needed.
Description
•