386 compilation error: dblogpruner/worker.go:32: constant 4294967296 overflows int

This is due to #1847 - introducing the dblogpruner worker. On i386 because int is int32 4 *1024*1024*1024 actually overflows it. I'll attempt to fix it by using int(uint32(4*1024*1024*1024)-1) instead.

Correction, the PR is #1781 but the issue was there before that, as the 4GB limit wasn't defined as a named constant.

Fix proposed with http://reviews.vapour.ws/r/1188/

Fix changed a bit to reflect reviewer comments and IRC discussion.

It appears the issue exposed some underlying problems with the implementation, which are outlined in the review. I'll reassign this to Tim and defer the decision on how to fix this as he has more information about it.

Fix is merging here: https://github.com/juju/juju/pull/1866

It seems like there are a couple of test failures:
(BTW can we please add gc.Commentf to such bizarre asserts like (X, jc.LessThan, Y) - if it fails it's obvious why it failed as the underlying assumption is not explicitly expressed).

----------------------------------------------------------------------
FAIL: logs_test.go:109: LogsSuite.TestPruneLogsBySize

SetUpTest
[LOG] 0:00.010 INFO juju.state opening state, mongo addresses: ["localhost:40798"]; entity <nil>
[LOG] 0:00.010 DEBUG juju.state dialing mongo
[LOG] 0:00.016 INFO juju.mongo dialled mongo successfully on address "127.0.0.1:40798"
[LOG] 0:00.022 DEBUG juju.state connection established
[LOG] 0:00.122 INFO juju.mongo dialled mongo successfully on address "127.0.0.1:40798"
[LOG] 0:00.122 INFO juju.state initializing environment, owner: "test-admin@local"
[LOG] 0:00.122 INFO juju.state info: &mongo.MongoInfo{Info:mongo.Info{Addrs:[]string{"localhost:40798"}, CACert:"-----BEGIN CERTIFICATE-----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\nh6UeX1gZ7eQ=\n-----END CERTIFICATE-----\n"}, Tag:names.Tag(nil), Password:""}
[LOG] 0:00.122 INFO juju.state starting presence watcher
[LOG] 0:00.133 INFO juju.mongo dialled mongo successfully on address "127.0.0.1:40798"
SetUpTest done
[LOG] 0:00.179 INFO juju.state opening state, mongo addresses: ["localhost:40798"]; entity <nil>
[LOG] 0:00.179 DEBUG juju.state dialing mongo
[LOG] 0:00.188 INFO juju.mongo dialled mongo successfully on address "127.0.0.1:40798"
[LOG] 0:00.189 DEBUG juju.state connection established
[LOG] 0:00.202 INFO juju.mongo dialled mongo successfully on address "127.0.0.1:40798"
[LOG] 0:00.208 INFO juju.mongo dialled mongo successfully on address "127.0.0.1:40798"
[LOG] 0:03.456 INFO juju.state opening state, mongo addresses: ["localhost:40798"]; entity <nil>
[LOG] 0:03.456 DEBUG juju.state dialing mongo
[LOG] 0:03.458 INFO juju.mongo dialled mongo successfully on address "127.0.0.1:40798"
[LOG] 0:03.458 DEBUG juju.state connection established
[LOG] 0:03.465 INFO juju.mongo dialled mongo successfully on address "127.0.0.1:40798"
[LOG] 0:03.465 INFO juju.mongo dialled mongo successfully on address "127.0.0.1:40798"
logs_test.go:132:
    c.Assert(s.countLogs(c, s1), jc.LessThan, 9000)
... obtained int = 10000
... expected int = 9000

[LOG] 0:06.901 INFO juju.testing reset successfully reset admin password

----------------------------------------------------------------------
FAIL: logs_test.go:147: LogsSuite.TestPruneLogsWithSmallSizeThreshold

SetUpTest
[LOG] 0:00.019 INFO juju.state opening state, mongo addresses: ["local...

Sorry that this isn't fixed yet. I was OCR today so didn't have much time for anything else. Part of the problem is that MongoDB on the test servers seems to behave quite differently in terms of disk usage to MongoDB on my development machine so everything passes locally but falis during test runs.

I will fix this tomorrow, perhaps by taking actual MongoDB usage out of the equation for these tests.

Fix committed: 0edbe59

I can see the i386 jobs are compiling again so closing this ticket.