1133273 - Treeherder tests use treeherder database instead of test_treeherder database after Django 1.7 changes

Status: RESOLVED DUPLICATE of bug 1193836

(Tree Management :: Treeherder, defect, P2)

Description

[James Lal [:lightsofapollo]]

While writing some new tests I could not figure out why the record I was creating was throwing integrity errors it turned out the test database was not being used (or purged) between test runs (perhaps this is intentional?)

Ideally we use the test_treeherder database for tests and drop data on error (I think this was the previous behavior prior to updating django)

Comment 1

[Ed Morley [:emorley]]

The Django 1.7 PR included this removal:
https://github.com/mozilla/treeherder-service/commit/698fd0551907d7c8c29514d9a897718e06b23e28#diff-484462fced51d1a06b1d93b4a44dd535L58

Comment 2

[Cameron Dawson [:camd]]

I'm not going to get to this before my PTO

Comment 3

[Ed Morley [:emorley]]

Did you get a chance to look at this? (regression from bug 1119479) I'm happy to take it on, just don't want to repeat work if you have a WIP locally :-)

Comment 4

[Cameron Dawson [:camd]]

Ed--  Hey there.  I did not get a chance, no.  If you want to take it, that'd be awesome.  Thanks man!

Comment 5

[Ed Morley [:emorley]]

Cool - no problem! :-)

Comment 6

[William Lachance (:wlach)]

Attachment: test failure with database initialization/destruction code re-enabled

(In reply to Ed Morley (UK public holiday 31st Aug) [:emorley] from comment #1)
> The Django 1.7 PR included this removal:
> https://github.com/mozilla/treeherder-service/commit/
> 698fd0551907d7c8c29514d9a897718e06b23e28#diff-
> 484462fced51d1a06b1d93b4a44dd535L58

You might think that just reverting those lines would fix the problem, but I'm afraid it doesn't (see the attached error output). Unfortunately we don't know exactly what's failing because it throws an exception when tearing down the test environment. Oh well.

I suspect our problems probably come from the fact that we call various external commands in the unit tests to initialize the external datasource databases, but these commands aren't inheriting our test settings. I dunno, I could be wrong.

Fixing this bug feels hard, at least to me. Part of me thinks we'd be best off just exorcising datasource and deleting most of conftest.py once everything's in a single database, instead of trying to keep a collection of what are essentially hacks working.

Comment 7

[William Lachance (:wlach)]

This was fixed in bug 1193836.