cherry pick "etcd_cluster_version" metric" (#10257, #11233, #11254, #11265) to release-3.3

[wenjiaswe]

Cherry pick of #10257 on release-3.3.

#10257: etcdserver/*: add "etcd_cluster_version" metric

[wenjiaswe]

/cc @jingyih hmm, failed test said

etcd_release_upgrade_test.go:108: failed get with curl (read /dev/ptmx: input/output error (expected "etcd_cluster_version{cluster_version=\"3.3\"} 1", got ["# HELP etcd_cluster_version Which version is running. 1 for 'cluster_version' label with current cluster version\r\n" "# TYPE etcd_cluster_version gauge\r\n" "etcd_cluster_version{cluster_version=\"3.3.0\"} 1\r\n"

But I brought up an etcd node and checked the metrics it is as expected:
etcd_cluster_version{cluster_version="3.3"} 1

[jingyih]

I can take a look. This test is skipped in current master branch. Might be an actual bug we did not catch in the master before backporting.

[jingyih]

Oh I misread. This is a test fail on a backported test. No wonder I cannot reproduce locally...

[jingyih]

Please also cherry pick #11265 to fix the test failure.

[wenjiaswe]

@jingyih yes, the third commit already includes #11265, failure mentioned earlier is fixed:

=== RUN TestReleaseUpgrade --- PASS: TestReleaseUpgrade (5.22s)

but travis still failed:
Too many goroutines running after all test(s). 1 instances of: os.(*Process).Release(...) /home/travis/.gimme/versions/go1.12.9.linux.amd64/src/os/exec.go:108 +0x5e FAIL github.com/coreos/etcd/tests/e2e 155.956s

[wenjiaswe]

@gyuho @jingyih Test passed now, looks like [this failure] is just flaky. Please help to review. Thanks!

[gyuho]

lgtm, thanks @jingyih @wenjiaswe