semaphore.sh, go: add semaphore to release-3.1 branch, upgrade release-3.1 to go 1.8.5

[jpbetz]

Pending approval of #8805. This adds semaphore to 3.1 branch, like 2f74456 did for the 3.2 branch.

This also upgrades the 3.1 branch from go 1.8.3 to 1.8.5, primarly because we already have gcr.io/etcd-development/etcd-test:go1.8.5 in the registry and don't see any point adding 1.8.3 when we can more easily upgrade go by a couple patch versions.

[gyuho]

lgtm, after CI greens.

Thanks!

[fanminshi]

lgtm

[gyuho]

Seems like we also need backport 6825ffe for Go 1.8+.

[jpbetz]

Thanks @gyuho, I'll cherry pick that as well.

[gyuho]

Hmm, nvm. Seems like we already have that in release-3.1.

Looks like it's something else

--- FAIL: TestV2NoRetryNoLeader (0.00s)
expected "no leader", got invalid URL port "123.4686.sock"

UPDATE: we need change this as well?

https://github.com/coreos/etcd/blob/0520cb9304cb2385f7e72b8bc02d6e4d3257158a/client/integration/client_test.go#L65-L69

[jpbetz]

Yep, cherry-pick gave me a "empty commit" response-- it's already on 3.1. I'll poke around tomorrow and figure out where .sock came from.

[jpbetz]

Thanks @gyuho Looks like 22c52b6 is what we need.

[gyuho]

lgtm. Thanks!

[gyuho]

Yeah maybe we can merge this first, and cherry-pick those in a separate PR?
UPDATE: nvm, patch is already pushed :)

[jpbetz]

I've split them out as two separate commits in this PR. That way we can verify that semaphore passes before committing any of this.

[jpbetz]

Only remaining test failure:

github.com/coreos/etcd/clientv3:

Too many goroutines running after all test(s).
2 instances of:
net.runtime_pollWait(...)
	/usr/local/go/src/runtime/netpoll.go:164 +0x4b
net.(*pollDesc).wait(...)
	/usr/local/go/src/net/fd_poll_runtime.go:75 +0x2b
net.(*pollDesc).waitRead(...)
	/usr/local/go/src/net/fd_poll_runtime.go:80 +0x29
net.(*netFD).Read(...)
	/usr/local/go/src/net/fd_unix.go:250 +0x154
net.(*conn).Read(...)
	/usr/local/go/src/net/net.go:181 +0x56
net/http.(*persistConn).Read(...)
	/usr/local/go/src/net/http/transport.go:1316 +0x15d
bufio.(*Reader).fill(...)
	/usr/local/go/src/bufio/bufio.go:97 +0xd5
bufio.(*Reader).Peek(...)
	/usr/local/go/src/bufio/bufio.go:129 +0x51
net/http.(*persistConn).readLoop(...)
	/usr/local/go/src/net/http/transport.go:1474 +0x185
created by net/http.(*Transport).dialConn
	/usr/local/go/src/net/http/transport.go:1117 +0x88b
2 instances of:
net/http.(*persistConn).writeLoop(...)
	/usr/local/go/src/net/http/transport.go:1704 +0x37f
created by net/http.(*Transport).dialConn
	/usr/local/go/src/net/http/transport.go:1118 +0x8ac

[gyuho]

Yeah, release-3.1 branch has been flaky with clientv3 tests.
We can address them when we backport clientv3 fixes.
Shouldn't block mvcc patches.

[fanminshi]

the semaphoreci is not running. @gyuho do you know why?

[gyuho]

@fanminshi It's running?

I just reran a few times https://semaphoreci.com/coreos/etcd/branches/pull-request-8807/builds/5, to see if it passes.

[fanminshi]

@gyuho you are right. I didn't noticed that you re-run the test.

[fanminshi]

similar to this failure #6933. not sure if they are related. @gyuho

[xiang90]

@jpbetz thank you.