1669998 - Try runs are not using correct fetch task artifact

Status: RESOLVED INVALID

(Firefox Build System :: Task Configuration, defect)

Description

[Greg Mierzwinski [:sparky]]

When I modify the fetch-task for chromium in a new patch that is pushed to try, I am able to get the fetch task triggered and everything works fine: https://treeherder.mozilla.org/#/jobs?repo=try&tier=1%2C2%2C3&revision=26e86e7e8ce654dde69287c3d41cda5126abe2ec&searchStr=chromium%2Cosx

Fetch task job: https://treeherder.mozilla.org/#/jobs?repo=try&tier=1%2C2%2C3&revision=26e86e7e8ce654dde69287c3d41cda5126abe2ec&searchStr=chromium%2Cfetch

But when I push this patch in a stack of patches and it isn't modified from the last time I pushed it to try, it doesn't trigger a new fetch task or use the artifact created in the last push (above). It resorts to using the artifact from mozilla-central when it certainly shouldn't since the fetch task was modified: https://treeherder.mozilla.org/#/jobs?repo=try&tier=1%2C2%2C3&revision=67bc15c07ab542feb2415619b2b0e2707d888f61&searchStr=chromium

Could someone look into this please?

Comment 1

[Aki Sasaki (not active)]

It doesn't look like your 2nd push is based on top of your 1st push. You probably want to do that (i.e., rebase your 2nd push on top of 38617f70135677acfd7c93c47a5fa555d062b3e0 )

Comment 2

[Mike Hommey [:glandium]]

The digest data doesn't contain anything that would vary based on the changes you made, so as far as the decision task is concerned, you didn't change anything that would matter for the resulting artifact.
https://searchfox.org/mozilla-central/rev/919607a3610222099fbfb0113c98b77888ebcbfb/taskcluster/taskgraph/transforms/fetch.py#379-383

OTOH, none of the definitions for those fetches actually have a revision, so they're not deterministic in the first place... It seems like what you'd really want is to either make them deterministic or make them never cached. If the former, you'd need to add a revision to taskcluster/ci/fetch/chromium-fetch.yml. If the latter, remove the digest data and adapt make_task to not call add_optimization when there isn't a digest_data.

Comment 3

[Greg Mierzwinski [:sparky]]

Oh I see, thanks a lot!

:glandium, do you think I could add the contents of the fetch script into the digest data to get the behaviour I'm expecting?

Comment 4

[Mike Hommey [:glandium]]

(In reply to Greg Mierzwinski [:sparky] from comment #3)

Oh I see, thanks a lot!

:glandium, do you think I could add the contents of the fetch script into the digest data to get the behaviour I'm expecting?

I think you either want a fixed revision or no optimization at all.

Comment 5

[Andrew Halberstadt [:ahal]]

From reading the comments here, it seems like this was a task configuration issue. I'm assuming it's not still a problem.