Flaky test: DockerSwarmSuite.TestPruneNetwork

[tophj-ibm]

Seen flakily failing on arm and ppc64le.

----------------------------------------------------------------------
FAIL: docker_cli_prune_unix_test.go:26: DockerSwarmSuite.TestPruneNetwork

[d4ebd159337da] waiting for daemon to start
[d4ebd159337da] daemon started
docker_cli_prune_unix_test.go:35:
    c.Assert(err, checker.IsNil)
... value *exec.ExitError = &exec.ExitError{ProcessState:(*os.ProcessState)(0xc4205245a0), Stderr:[]uint8(nil)} ("exit status 1")

[d4ebd159337da] exiting daemon

----------------------------------------------------------------------

unfortunately the Jenkins logs were deleted as I was making this, but I'm seeing it on my local power machine on master as well.

[tophj-ibm]

hmm looks like a network isn't being cleaned up correctly, from failing docker.log here https://gist.github.com/tophj-ibm/bd415b07affffd124eee5e45ac82e408

time="2016-10-28T19:02:59.242218000Z" level=debug msg=""Calling POST /v1.25/networks/create"
time="2016-10-28T19:02:59.242366000Z" level=debug msg="form data: {\"Attachable\":false,\"CheckDuplicate\":true,\"Driver\":\"overlay\",\"EnableIPv6\":false,\"IPAM\":{\"Config\":[],\"Driver\":\"default\",\"Options\":{}},\"Internal\":false,\"Labels\":{},\"Name\":\"n4\",\"Options\":{}}"
time="2016-10-28T19:02:59.243590000Z" level=error msg="Handler for POST /v1.25/networks/create returned error: network with name n4 already exists"

[thaJeztah]

This test was added in #27525

ping @AkihiroSuda ptal

[AkihiroSuda]

https://jenkins.dockerproject.org/job/Docker%20Master%20(arm)/3558/console seems hitting this.

But AFAIK no test other than TestPruneNetwork creates a network named "n4"...

(bundle: https://jenkins.dockerproject.org/job/Docker%20Master%20(arm)/3571/)

$ find bundles -name docker.log | xargs grep '\\"n4\\"'
bundles/1.13.0-dev/test-integration-cli/d8e3bfbf769bf/docker.log:time="2016-10-29T20:02:58.304429896Z" level=debug msg="form data: {\"Attachable\":false,\"CheckDuplicate\":true,\"Driver\":\"overlay\",\"EnableIPv6\":false,\"IPAM\":{\"Config\":[],\"Driver\":\"default\",\"Options\":{}},\"Internal\":false,\"Labels\":{},\"Name\":\"n4\",\"Options\":{}}"
^
|
+---- this is the log for TestPruneNetwork

Maybe the error string "error: network with name n4 already exists" is wrong, but not sure yet.
Trying to reproduce..

[AkihiroSuda]

Reproduced during BUILDFLAGS="-race" TESTFLAGS='-check.f TestPrune -test.count 100' make test-integration-cli, on Ubuntu 16.10, kernel 4.8.0-26-generic, x86_64

So the test failure is

• Not related to any other test
• Not related to arm nor ppc64le

Looking at the log (mine and @tophj-ibm 's #27866 (comment)) , the failure is caused when the ID of the ingress starts with n%d..

e.g.

Allocating IPv4 pools for network ingress (n4whp4a1b4bkgw23akoamblhz)

The ID is generated from here; it occasionally creates a network named n%d: https://github.com/docker/docker/blob/12a4ed03178982fec44b75479c3e63a316503d90/vendor/src/github.com/docker/swarmkit/identity/randomid.go#L44
(cc @aaronlehmann @stevvooe )

[AkihiroSuda]

I'll try to open PR that allows to create a network of which name is the prefix of the ID of the ingress.

[AkihiroSuda]

opened #27938