Output list of manifests selected by shadow schedulers
Categories
(Firefox Build System :: Task Configuration, task)
Tracking
(firefox81 fixed)
Tracking | Status | |
---|---|---|
firefox81 | --- | fixed |
People
(Reporter: marco, Assigned: ahal)
References
(Blocks 1 open bug)
Details
Attachments
(1 file)
Needed for showing manifest-level data on the dashboard (bug 1603467 / https://github.com/mozilla/cia-tasks/issues/1).
Assignee | ||
Comment 1•5 years ago
|
||
Do we want all manifests including the collateral damage ones? If not this will be a bit trickier.
Reporter | ||
Comment 2•5 years ago
|
||
I just wanted the list of actually selected manifests, not the collateral damage ones, so I can measure how the schedulers will be doing when we will have manifest-level scheduling.
I can just evaluate on historical data until this is more easily feasible though.
Reporter | ||
Comment 3•4 years ago
|
||
Could we simply make the shadow schedulers output the task graph as JSON instead of as a list of labels, and then let consumers choose how to use the data?
E.g. a consumer could:
- Load the task graph as JSON;
- List all manifests from manifest-level tasks by using MOZHARNESS_TEST_PATHS;
- List all non-manifest-level tasks (by filtering out tasks that have MOZHARNESS_TEST_PATHS).
Assignee | ||
Comment 4•4 years ago
|
||
Yeah, I like that idea.
For now we could just paste the optimized taskgraph as an artifact (would require minimal changes to the shadow-scheduler task). In the future, we could re-do the shadow-schedulers such that they get invoked via decision.py
and thus upload all the same artifacts that a normal decision task would. I'm not sure if that's important or not though.
Assignee | ||
Comment 5•4 years ago
|
||
This will put some extra strain on the ETL, and we'll need to coordinate landing it with Kyle.
Assignee | ||
Comment 6•4 years ago
|
||
Comment 8•4 years ago
|
||
bugherder |
Description
•