relationunit_test.go: 2 tests fail with mgo >= 241
Bug #1221705 reported by
Dimiter Naydenov
This bug affects 1 person
Affects | Status | Importance | Assigned to | Milestone | ||
---|---|---|---|---|---|---|
juju-core | Status tracked in Trunk | |||||
1.14 |
Fix Released
|
Critical
|
Unassigned | |||
Trunk |
Fix Released
|
Critical
|
Unassigned | |||
mgo |
Fix Released
|
Undecided
|
Unassigned |
Bug Description
See: https:/
I cannot waste more time on this issue. Let's look
more into it when jam's back.
Proposed a branch that skips those tests: https:/
Related branches
lp:~dimitern/juju-core/124-skip-failing-relationunit-tests
- Juju Engineering: Pending requested
-
Diff: 25 lines (+8/-0)1 file modifiedstate/relationunit_test.go (+8/-0)
lp:~jameinel/juju-core/restore-tests-1221705
- John A Meinel: Approve
-
Diff: 37 lines (+1/-9)2 files modifieddependencies.tsv (+1/-1)
state/relationunit_test.go (+0/-8)
lp:~dave-cheney/juju-core/backport-r1782
- John A Meinel: Approve
-
Diff: 13 lines (+2/-1)1 file modifiedscripts/build-release-tarball/build-release-tarball.bash (+2/-1)
Changed in juju-core: | |
status: | In Progress → Fix Committed |
To post a comment you must log in.
I don't know why, but the bot is running mgo@ revno 243. This includes some of Gustavo's changes to the mgo driver. I did not personally update mgo, but maybe some other trigger did.
If I update to rev 242 of mgo on my local box, these tests fail for me.
I only thought of this because of Gustavo's recent release of mgo: blog.labix. org/2013/ 09/05/mgo- r2013-09- 04-released
http://
Note that one of the mgo changes include supporting "Improved Timeouts" which involves "Socket level deadlines".
So I'm guessing something in the stack added the change, and something about these tests cause stuff to take longer in Mongo.
If I use mgo@revno 238 (the last version I had locally), I see those 2 tests taking 1.5s and 1.4s which is about 10 slower than the other similar tests (all the other ones take <200ms).
So my guess is an updated mgo added socket timeouts, which were hiding whatever was actually going wrong. I don't know what, but I do think it is worth actually investigating. We can disable the tests for now, because we shouldn't block landing code, but it looks like we might have a genuine issue behind them.