Closed Bug 900630 Opened 7 years ago Closed 7 years ago

Unable to install /builds/slave/test/build/gaia-ui-tests!

Categories

(Firefox OS Graveyard :: General, defect)

defect
Not set

Tracking

(Not tracked)

RESOLVED FIXED

People

(Reporter: jgriffin, Unassigned)

References

Details

Gaia-ui-tests on cedar are now dying with:

13:11:27     INFO -  No distributions at all found for marionette-client==0.5.35 (from gaiatest)
13:11:27     INFO -  Storing complete log in /home/cltbld/.pip/pip.log
13:11:27    ERROR - Return code: 1
13:11:27    FATAL - Unable to install /builds/slave/test/build/gaia-ui-tests!
13:11:27    FATAL - Exiting -1

This happened because in-tree marionette is at 0.5.36, but gaiatest is pinned to 0.5.35.  

There are two possible fixes to this:

1) change gaiatest not to use strict versioning for in-tree components
2) provide an alternate setup.py in gaiatest used by TBPL which doesn't use strict versioning

Zac, do you have a preference?
Flags: needinfo?(zcampbell)
Looks like it has been bumped already anyway by Aaron's commit. 

I'd rather stay pinned (although others always bump it untested anyway), refer our email conversations about the test/release process for the client about a month ago. the recent stability in Marionette has been great lately and I don't think it's worth risking it over small updates over the client bindings. I don't even think TBPL should risk it either.

Could we do >= ? Then, if I understand, TBPL would pick up the tree and we would just get it when it's released to pypi?

It seems odd that gaiatest and the client bindings should have this interdependency anyway, in my experience with WD it's the test framework that uses the client not the testrunner.
Flags: needinfo?(zcampbell)
(In reply to Zac C (:zac) from comment #1)
> Looks like it has been bumped already anyway by Aaron's commit. 
> 
> I'd rather stay pinned (although others always bump it untested anyway),
> refer our email conversations about the test/release process for the client
> about a month ago. the recent stability in Marionette has been great lately
> and I don't think it's worth risking it over small updates over the client
> bindings. I don't even think TBPL should risk it either.

We actually want TBPL to use the latest.  We want to be able to test the effects of changes to the marionette client within TBPL, so that we catch problems right away if they exist.

> 
> Could we do >= ? Then, if I understand, TBPL would pick up the tree and we
> would just get it when it's released to pypi?

Yes, using >= would work fine, and would be least amount of work for us.  In this scheme, we would use TBPL to validate any client changes, meaning that pushing them to pypi would reduce the risk of breakage in Jenkins.

> 
> It seems odd that gaiatest and the client bindings should have this
> interdependency anyway, in my experience with WD it's the test framework
> that uses the client not the testrunner.

The distinction is a little artificial; we'd have this same issue even if we packaged the client bindings separately, since both bindings and runner are changing (albeit not too frequently).
I just thought that one poor review/commit could cause all of TBPL test runs in all environments. The risk is pretty high.

Anyway these are still failing on Cedar so a bump to 0.5.36 does not seem to have resolved it. The issue must lie elsewhere?
We need to update gaia-ui-tests.json in the gaia repo, in order to pick up your changes.  I'll take care of this.
Bumped with https://github.com/mozilla-b2g/gaia/commit/947804b3e190285a66522bb9a3c9ecb5f7b70f40; we'll see how the results change now, after the related gaia.json change gets merged to m-c and then cedar.
This particular problem is fixed; we're now getting another problem about tbpl-manifest.ini missing.
(In reply to Jonathan Griffin (:jgriffin) from comment #6)
> This particular problem is fixed; we're now getting another problem about
> tbpl-manifest.ini missing.

Ugh, this happened because I updated gaia.json with a pointer to v1-train instead of master.  It's not easy at all to distinguish v1-train commits vs master commits in the hg mirror, see: http://hg.mozilla.org/integration/gaia-ui-tests/pushloghtml?changeset=db5033b9467e
Fixed in https://github.com/mozilla-b2g/gaia/commit/a5dbb991443d00ed3f11ab374a3d8826e85d7f0a; we'll have to wait for this to work it's way over to cedar again...
Tests are running again on cedar, but failing due to some test failures that need to be investigated.
Status: NEW → RESOLVED
Closed: 7 years ago
Resolution: --- → FIXED
You need to log in before you can comment on or make changes to this bug.