1858112 - performance try push doesn't show jobs in perfherder

Status: RESOLVED FIXED

(Testing :: Performance, defect, P3)

Description

[Emilio Cobos Álvarez (:emilio)]

I had this happen on two jobs from yesterday:

• https://treeherder.mozilla.org/perfherder/compare?originalProject=try&originalRevision=120739cce5873b60be4c60065082bc0ffe1ebe3b&newProject=try&newRevision=c77d1d98561e4e82d56212623272a8208b12ee64&page=1&framework=13, where we have tests on both windows and linux on both runs, but it looks like they're not found by the compare tool
• https://treeherder.mozilla.org/perfherder/compare?originalProject=try&newProject=try&newRevision=09a9fee90d6c9fa4a3128a0e564884d256c026d5&framework=13&originalRevision=120739cce5873b60be4c60065082bc0ffe1ebe3b&page=1 has the linux jobs but not the windows jobs.

Alex mentioned this also happened on perf.compare.

Comment 1

[Emilio Cobos Álvarez (:emilio)]

The push was made with ./mach try perf --rebuild 10 --show-all --query "'shippable 'speedometer3 'firefox !32"

Comment 2

[Acasandrei Beatrice (needinfo me)]

I investigated the links provided and in both cases there are no results in Threeherder's db and that's why nothing in displayed in Compare View.
Here are some links for the results in Compare View which show that the values field is empty (I also manually searched for the results in the performance_datum table from Treeherder using https://sql.telemetry.mozilla.org/queries/new to confirm this).

Results for the first link:

speedometer3 opt fission webrender	Base	New
linux1804-64-shippable-qr	base values	new values
windows10-64-shippable-qr	base values	new values

Results for the second link:

speedometer3 opt fission webrender	Base	New
windows10-64-shippable-qr	base values	new values

Comment 3

[Julien Wajsberg [:julienw]]

Do you know how this could happen Beatrice?

Comment 4

[Acasandrei Beatrice (needinfo me)]

(In reply to Julien Wajsberg [:julienw] from comment #3)

Do you know how this could happen Beatrice?

Not sure, I'm thinking is a tc or ingestion issue - but I don't have much context in that area.

Comment 5

[Julien Wajsberg [:julienw]]

Who could investigate this more?
It would be good to determine if this is a one-off issue or if this happens regularly.

Comment 6

[Alexandru Ionescu (needinfo me) [:alexandrui]]

Beatrice and I are on it :)

Comment 7

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

Perfherder uses its API to fetch this data for the "old push" from comment 0. Only element 3396 has a non-empty values property (the ones for Linux). It looks like element 3401 should be the one for Windows (+ Firefox) but values is empty.

Can you check the data is in the database? A query which groups by machine_platform.platform counts the following performance_datum rows:

gecko-decision
	1
lint
	1
linux1804-64-shippable-qr
	2,080
linux64-shippable
	103
macosx1015-64-shippable
	1
osx-shippable
	38
windows2012-64-shippable
	42

The 42 Windows rows are for the build tasks.

Comment 8

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

The issue is unrelated to performance - the logs for these tasks did not get parsed. I continue the investigation.

Comment 9

[Alexandru Ionescu (needinfo me) [:alexandrui]]

Thank you, :aryx!

Comment 10

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

The logs from comment 0 did not get parsed because of an issue with the deployment at that time. The other (common) case of logs not getting parsed - which did not affect extraction of performance data and failure lines - has a patch in bug 1861045.