precise unit-tests fail in 1.20

[Curtis Hovey]

Presice unit tests with ommmit 1cd26425cac98adc4e9aa70efa4e23b70f026c1d and fbd45936 do not pass on precise. For each revision the tests were run 5 times. At least one of the runs failed because of archive issuesin in ec2. I updated the test to use a newer AMI, but the tests still failed. The most recent failures are
    http://juju-ci.vapour.ws:8080/job/run-unit-tests-precise-amd64/1328/console
    http://juju-ci.vapour.ws:8080/job/run-unit-tests-precise-amd64/1329/console
    http://juju-ci.vapour.ws:8080/job/run-unit-tests-precise-amd64/1330/console
    http://juju-ci.vapour.ws:8080/job/run-unit-tests-precise-amd64/1331/console

I will look for a common failure and update this bug.

[Curtis Hovey (sinzui)]

In each run, the failing test is unique, though 3 of them are the same suite.
    FAIL: uniter_test.go:664: UniterSuite.TestUniterUpgradeOverwrite
    FAIL: uniter_test.go:1248: UniterSuite.TestUniterSubordinates
    FAIL: uniter_test.go:660: UniterSuite.TestUniterSteadyStateUpgrade

with a near identical time to run TestPackage
    OOPS: 48 passed, 1 FAILED
    --- FAIL: TestPackage (213.87 seconds)

One test was killed getting mongo? I think this is a separate issue:

SIGQUIT: quit
PC=0x538af2

syscall.Syscall(0x0, 0x7, 0xc210196771, 0x88f, 0x6a, ...)
 /usr/lib/go/src/pkg/syscall/asm_linux_amd64.s:18 +0x5 fp=0x2ac58021ca08
syscall.read(0x7, 0xc210196771, 0x88f, 0x88f, 0x6a, ...)
 /usr/lib/go/src/pkg/syscall/zsyscall_linux_amd64.go:838 +0x72 fp=0x2ac58021ca50
syscall.Read(0x7, 0xc210196771, 0x88f, 0x88f, 0x6a, ...)
 /usr/lib/go/src/pkg/syscall/syscall_unix.go:135 +0x5c fp=0x2ac58021ca90
os.(*File).read(0xc2103326a0, 0xc210196771, 0x88f, 0x88f, 0x6a, ...)
 /usr/lib/go/src/pkg/os/file_unix.go:180 +0x62 fp=0x2ac58021cad0
os.(*File).Read(0xc2103326a0, 0xc210196771, 0x88f, 0x88f, 0xc210196707, ...)
 /usr/lib/go/src/pkg/os/file.go:95 +0x98 fp=0x2ac58021cb28
io.(*multiReader).Read(0xc2100e7a80, 0xc210196771, 0x88f, 0x88f, 0xd9e1b0, ...)
 /usr/lib/go/src/pkg/io/multi.go:13 +0xa2 fp=0x2ac58021cb80
bufio.(*Scanner).Scan(0xc210093a80, 0x8ab460)
 /usr/lib/go/src/pkg/bufio/scan.go:165 +0x487 fp=0x2ac58021cca0
github.com/juju/testing.readLastLines(0xc2101e3280, 0xc, 0x2ac5767bde20, 0xc2100e7a80, 0x14, ...)
 /home/ubuntu/juju-core_1.20.2/src/github.com/juju/testing/mgo.go:381 +0x13f fp=0x2ac58021cdf0
github.com/juju/testing.func��012()
 /home/ubuntu/juju-core_1.20.2/src/github.com/juju/testing/mgo.go:258 +0x412 fp=0x2ac58021cfa0
runtime.goexit()
 /usr/lib/go/src/pkg/runtime/proc.c:1394 fp=0x2ac58021cfa8
created by github.com/juju/testing.(*MgoInstance).run
 /home/ubuntu/juju-core_1.20.2/src/github.com/juju/testing/mgo.go:270 +0x9dd

goroutine 1 [chan receive]:
testing.RunTests(0x994a08, 0xd9d780, 0x1, 0x1, 0xc2100f4901)
 /usr/lib/go/src/pkg/testing/testing.go:472 +0x8d5
testing.Main(0x994a08, 0xd9d780, 0x1, 0x1, 0xda7e40, ...)
 /usr/lib/go/src/pkg/testing/testing.go:403 +0x84
main.main()
 github.com/juju/juju/replicaset/_test/_testmain.go:47 +0x9c

goroutine 3 [syscall]:
os/signal.loop()
 /usr/lib/go/src/pkg/os/signal/signal_unix.go:21 +0x1e
created by os/signal.init��1
 /usr/lib/go/src/pkg/os/signal/signal_unix.go:27 +0x31

goroutine 4 [chan receive]:
launchpad.net/gocheck.(*suiteRunner).runTest(0xc2100973f0, 0xc210097a10, 0x0)
 /home/ubuntu/juju-core_1.20.2/src/launchpad.net/gocheck/gocheck.go:773 +0x51
launchpad.net/gocheck.(*suiteRunner).run(0xc2100973f0, 0xc210077f00)
 /home/ubuntu/juju-core_1.20.2/src/launchpad.net/gocheck/gocheck.go:585 +0x1f4
launchpad.net/gocheck.Run(0x88f580, 0xc210077f00, 0xc2100ada40, 0x769ea1)
 /home/ubuntu/juju-core_1.20.2/src/launchpad.net/gocheck/run.go:76 +0x49
launchpad.net/gocheck.RunAll(0xc2100ada40, 0xc2100ada40)
 /home/ubuntu/juju-core_1.20.2/src/launchpad.net/gocheck/run.go:68 +0xc1
launchpad.net/gocheck.TestingT(0xc210058120)
 /home/ubuntu/juju-core_1.20.2/src/launchpad.net/gocheck/run.go:56 +0x2d9
github.com/juju/j...

[Curtis Hovey (sinzui)]

I think this may be fixed by gitbranch:1.20:github.com/juju/juju 2c4db889 . It passed in the first try. I will rerun the tests in jenkins to dismiss "lucky" as the reason for the pass.

[Curtis Hovey (sinzui)]

Andrew's fix for azure also seems to fix this.