Closed
Bug 1279591
Opened 9 years ago
Closed 9 years ago
Perfherder ignores test results with "0" value in compare view
Categories
(Tree Management :: Perfherder, defect)
Tree Management
Perfherder
Tracking
(Not tracked)
RESOLVED
FIXED
People
(Reporter: wlach, Assigned: wlach)
References
Details
Attachments
(1 file)
According to the comments, we do this because "Talos tests may output 0 as an indication of failure". I have never personally seen this, and this behavior caused confusion in bug 1279292. I'm going to suggest we just display/include these new values in the compare view.
Comment 1•9 years ago
|
||
| Assignee | ||
Comment 2•9 years ago
|
||
Comment on attachment 8762141 [details] [review]
[treeherder] wlach:1279591 > mozilla:master
This makes the null / 0 values in:
https://treeherder.mozilla.org/perf.html#/comparesubtest?originalProject=try&originalRevision=b1047ee23ca1&newProject=try&newRevision=b1047ee23ca1&originalSignature=12ea55b0c0e0b98c6b70a890fdc51244b7604e89&newSignature=12ea55b0c0e0b98c6b70a890fdc51244b7604e89&framework=1
display correctly (e.g. tscrollx tiled-fixed.html.CSSOM.checkerboard opt e10s).
Am I missing something or do we really need to exclude 0 results from talos in some cases? I don't recall putting this check in the compare view so I'm guessing it was one of you.
Attachment #8762141 -
Flags: feedback?(jmaher)
Attachment #8762141 -
Flags: feedback?(avihpit)
Comment 3•9 years ago
|
||
Good question. I do recall such notion but I don't recall specific cases it was meant to address.
Joel doesn't recall either, and on IRC we agreed that unless it makes everything explode, showing 0 even if it indicates an error is actually a good thing - this way we can notice it and try to do something about it.
Updated•9 years ago
|
Attachment #8762141 -
Flags: review+
Attachment #8762141 -
Flags: feedback?(jmaher)
Attachment #8762141 -
Flags: feedback+
Comment 4•9 years ago
|
||
Comment on attachment 8762141 [details] [review]
[treeherder] wlach:1279591 > mozilla:master
I do have one minor concern though, and it's that if we do have subtests which report 0, no one would notice them unless they open the subtests page - but the overall test score would seem magically better.
Let's hope this doesn't happen, and that if it does, we catch that on time.
Attachment #8762141 -
Flags: feedback?(avihpit) → feedback+
Comment 5•9 years ago
|
||
Autolander could not locate a review from a user within the suggested reviewer list. Either the patch author or the reviewer should be in the suggested reviewer list.
Comment 6•9 years ago
|
||
Commit pushed to master at https://github.com/mozilla/treeherder
https://github.com/mozilla/treeherder/commit/6fd2ed2ab6af63ea567468e29d7ace327065e1f3
Bug 1279591 - Include 0 / null values in Perfherder compare view (#1585)
This was intended to guard against Talos outputting 0 values in the case of
error, but I don't believe that happens anymore (if it ever did?)
| Assignee | ||
Comment 7•9 years ago
|
||
(not sure what's going on with autolander... merged manually)
Status: NEW → RESOLVED
Closed: 9 years ago
Resolution: --- → FIXED
Comment 8•9 years ago
|
||
I just recalled another thing. Geometric mean is not great with 0, and it's applied to subtests on more than one place I think...
We do have workaround in some places where we add 1 to all values, calculate the geomean, then reduce 1 from the result, but I don't recall where this workaround is deployed, or that I even know all the places it could bite us.
But as wlach said on IRC, it would make it more likely to make us notice issues sooner...
You need to log in
before you can comment on or make changes to this bug.
Description
•