Add e2e testing of ipvlan, macvlan, and bridge plugins

[dcbw]

Fixes #179

[dcbw]

@steveej @jonboulle I intend to rebase #145 on top of this and use this framework for the testcases that were requested.

[dcbw]

As a bonus it increases test coverage from 21% to 36% overall, and we get this for plugins:

ok github.com/appc/cni/plugins/main/ipvlan 0.014s coverage: 68.9% of statements
ok github.com/appc/cni/plugins/main/macvlan 0.013s coverage: 66.7% of statements
ok github.com/appc/cni/plugins/main/bridge 0.009s coverage: 63.5% of statements

[dcbw]

@danwinship a review from you would be nice since you've done similar-yet-different stuff for openshift-sdn...

[danwinship]

Seems like maybe it would be good to have a way to run the new test programs against the actual netlink implementation, so you can make sure that they actually actually work. (Maybe a second set of tests using the real ops instead of the test ops, but skipped by default?)

[dcbw]

@steveej @jonboulle @danwinship PTAL; latest push enables both mock and live testing. I'm not 100% happy with how the live testing gets done (by running all testcases over with a new env variable) but I can fix that next week if people want.

[dcbw]

BTW, the first commit in this PR is actually #176

[dcbw]

travis failure seems to be a flake?

$ go get ${TOOLS_CMD}/vet

package golang.org/x/tools/cmd/vet: cannot find package "golang.org/x/tools/cmd/vet" in any of:

    /home/travis/.gimme/versions/go1.5.3.linux.amd64/src/golang.org/x/tools/cmd/vet (from $GOROOT)

    /home/travis/gopath/src/golang.org/x/tools/cmd/vet (from $GOPATH)

The command "go get ${TOOLS_CMD}/vet" failed and exited with 1 during .

Your build has been stopped.

[steveej]

missing default case

[rosenhouse]

@dcbw The NewNS function has some problems still. On line 140 you're checking the value of an err variable, which makes it appear that you're trying to handle the error cases from inside the go (func() { call above. But inside that function, on line 111, you shadow the err variable. So the err values set within the func will not be visible at the outer scope.

[rosenhouse]

I'm also unclear on why you're calling origNS.Set() 3 different times, rather than just doing a defer origNS.Set() on line 122.

[dcbw]

@rosenhouse rebased and fixed NewNS. PTAL, thanks!

[dcbw]

@rosenhouse as a side-effect of only calling LockOSThread() in the goroutine for netns.Do(), we now need to make sure that any checks we do in testcases are done wrapped in a Do(), otherwise I see periodic failures where the namespace gets switched back, possibly because the testcase gets switched to the main thread every so often. Just something to be aware of.

[steveej]

@dcbw @rosenhouse I suggest that we split this PR in two. The first one without change in implementation related to tests at all, just adding of tests and fixes in the network locking. The second one can then be rebased on top of the reworked first one. I would really like to get this in by the end of the week as it is the biggest blocker for v0.3 which is highly awaited at this point.

[dcbw]

@steveej could you clarify which parts you'd like split out into separate PRs? I'm not sure which parts you mean by "just adding of tests and fixes in the network locking". Is there anything else I can do about the later patches in the series that add e2e testing that would make you more comfortable with them?

I'm happy to continue reworking patches, but to get things done by the end of the week the review/rework cycle needs to be a lot shorter...

[dcbw]

In the interest of getting the fixes merged faster I've removed the netops stuff and pushed a reduced set of commits in #211. We'll leave this one open to discuss mocking further.

[steveej]

@dcbw I've lost track of this PR's changes. Have they been merged via another PR?

[dcbw]

@steveej the only thing this PR was kept around for was the mock netlink stuff. But I guess we long ago decided not to use that, so I'll close and we can always resurrect later if we want to.