Closed
Bug 1204592
Opened 9 years ago
Closed 6 years ago
Support |mach try| when autolanding to try
Categories
(MozReview Graveyard :: General, defect, P3)
MozReview Graveyard
General
Tracking
(Not tracked)
RESOLVED
INVALID
People
(Reporter: dminor, Unassigned)
References
Details
The |mach try| command is soon going to support intelligent, automated test selection. It would be nice to make this available to MozReview users rather than having them cut and paste try syntax each time.
I think we'd make it the default option while still supporting manual try syntax.
Reporter | ||
Updated•9 years ago
|
Assignee: nobody → dminor
Comment 2•9 years ago
|
||
Choices:
1) Run |mach try| from autoland
2) Grab the moz.build files metadata from the JSON API on hg.mozilla.org. https://hg.mozilla.org/mozilla-central/json-mozbuildinfo/<rev>. We'll need to teach this API to expose this data first, of course.
Comment 3•9 years ago
|
||
We were discussing this last week in Toronto, and I think we need to consider a few things before making this a default. There are numerous common cases the prioritization in moz.build will work well, but there are a other cases it's going to disagree with our infrastructure due to test inter-dependence and chunking.
Test inter-dependence is a problem if we select a single directory and there are failures due to dependence on tests that usually run before that directory. We have run-by-dir almost everywhere, but it's not in reftests, and not in every mochitest combination, so this is a potential problem.
If our test prioritization algorithm results in selecting a large number of tests, we need to figure out how to chunk those. The current approach is to run our selection in a single chunk, but buildbot will fail the job if we inadvertently select more than two hours worth of tests (this is bug 1205008).
Unfortunately, completely resolving these issues means having run-by-dir (or another way to isolate tests) everywhere, and a uniform way to chunk a varying selection of tests, which is something that we may need to implement in TaskCluster. So we've deferred some of the more extensive work we discussed on test prioritization.
So as it stand I think this is a useful option to people who are familiar with our infrastructure and willing to learn a little bit about how it works, but any spurious failures caused by the above might cause frustration if this were our default.
Reporter | ||
Updated•9 years ago
|
Assignee: dminor → nobody
Assignee | ||
Updated•9 years ago
|
Product: Developer Services → MozReview
Comment 4•6 years ago
|
||
MozReview is now obsolete. Please use Phabricator instead. Closing this bug.
Status: NEW → RESOLVED
Closed: 6 years ago
Resolution: --- → INVALID
You need to log in
before you can comment on or make changes to this bug.
Description
•