Closed Bug 1069641 Opened 5 years ago Closed 3 years ago

Upload to blobber if marionette builds fail

Categories

(Release Engineering :: General, defect)

x86
macOS
defect
Not set

Tracking

(Not tracked)

RESOLVED INCOMPLETE

People

(Reporter: mdas, Unassigned)

References

Details

From Bug 1066187, we added gecko.logs to blobber, but that only happens if the mozharness script isn't killed. It'd be most useful if these get uploaded even when the script is killed due to timeout, as it will help us diagnose why our tests were taking so long/having timeouts, like https://tbpl.mozilla.org/?tree=B2g-Inbound&showall=1&jobname=b2g_ubuntu64_vm%20b2g-inbound%20debug%20test%20gaia-ui-test for example. I'm trying to diagnose what's going on with the red Gip tests.

mtabara suggested that I could have the marionette.py mozharness script listen for a kill signal, and when it is received, it will attempt to upload any existing files from the "MOZ_UPLOAD_DIR" environment variable to blobber.

Is this an acceptable tactic?
catlee said kill -9 is used, so our script can't catch and handle it...
Status: NEW → RESOLVED
Closed: 5 years ago
Resolution: --- → INVALID
Ah, right :( Nothing can survive -9 from kernel. Sorry
Can we add a feature to register a command to run when buildbot is forced to kill a mozharness script? Like a 'post-kill' command where in my case, I'd write a command to find any build artifacts and upload them. It's critical to figuring out busted build problems!
Status: RESOLVED → REOPENED
Resolution: INVALID → ---
Do alwaysRun steps still run if a previous step is timed out? If so, we might be able to use one to do this. The configs might be a bit tricky, because the scripts that are available (and sometimes their location) depend on the specific job type.
Yes, I believe they do, as long as the slave hasn't disconnected.
Status: REOPENED → RESOLVED
Closed: 5 years ago3 years ago
Resolution: --- → INCOMPLETE
Component: General Automation → General
You need to log in before you can comment on or make changes to this bug.