Automated cherry pick of #11652 #11670 #11710

[tangcong]

Cherry pick of #11652 #11670 #11710 on release-3.4.

#11652: *: fix auth revision corruption bug
#11670: optimize auth/etcdserver logs to facilitate troubleshooting data inconsistency
#11710: auth: ensure RoleGrantPermission is compatible with older

[tangcong]

@jingyih @gyuho After listening to your suggestions(Refactored code doesn't need be backported),i have submitted two cherry-pick pr #11752 (release-3.4), #11753 (release-3.3) and also updated changlog #11750. PTAL. thanks.

[jingyih]

Could you help take a look at the failed e2e test?

=== RUN   TestCtlV3AuthRoleGet
--- FAIL: TestCtlV3AuthRoleGet (15.76s)
    testutil.go:55: goroutine 734 [running]:
        go.etcd.io/etcd/pkg/testutil.FatalStack(0xc00022c300, 0xc0001bb8c0, 0x18)
        	/go/src/go.etcd.io/etcd/pkg/testutil/testutil.go:54 +0x6d
        go.etcd.io/etcd/tests/e2e.testCtl(0xc00022c300, 0xd4d0f0, 0x0, 0x0, 0x0)
        	/go/src/go.etcd.io/etcd/tests/e2e/ctl_v3_test.go:184 +0x431
        go.etcd.io/etcd/tests/e2e.TestCtlV3AuthRoleGet(0xc00022c300)
        	/go/src/go.etcd.io/etcd/tests/e2e/ctl_v3_auth_test.go:55 +0x48
        testing.tRunner(0xc00022c300, 0xd4ca28)
        	/usr/local/go/src/testing/testing.go:865 +0xc0
        created by testing.(*T).Run

[tangcong]

Could you help take a look at the failed e2e test?

=== RUN   TestCtlV3AuthRoleGet
--- FAIL: TestCtlV3AuthRoleGet (15.76s)
    testutil.go:55: goroutine 734 [running]:
        go.etcd.io/etcd/pkg/testutil.FatalStack(0xc00022c300, 0xc0001bb8c0, 0x18)
        	/go/src/go.etcd.io/etcd/pkg/testutil/testutil.go:54 +0x6d
        go.etcd.io/etcd/tests/e2e.testCtl(0xc00022c300, 0xd4d0f0, 0x0, 0x0, 0x0)
        	/go/src/go.etcd.io/etcd/tests/e2e/ctl_v3_test.go:184 +0x431
        go.etcd.io/etcd/tests/e2e.TestCtlV3AuthRoleGet(0xc00022c300)
        	/go/src/go.etcd.io/etcd/tests/e2e/ctl_v3_auth_test.go:55 +0x48
        testing.tRunner(0xc00022c300, 0xd4ca28)
        	/usr/local/go/src/testing/testing.go:865 +0xc0
        created by testing.(*T).Run

semaphoreci is flaky. travis ci seems ok. can it be retested? local test is ok.

[jingyih]

semaphore ci and travis ci are actually testing different sets of tests. I have not noticed this particular test been flaky in the past. Plus, the test is auth related. Therefore I asked, just want to make sure.

[tangcong]

semaphore ci and travis ci are actually testing different sets of tests. I have not noticed this particular test been flaky in the past. Plus, the test is auth related. Therefore I asked, just want to make sure.

got it thanks. semaphore ci failed to pass due to test timeout. I checked again without changing the auth role logic. I can't trigger it to retest.

[jingyih]

semaphore ci and travis ci are actually testing different sets of tests. I have not noticed this particular test been flaky in the past. Plus, the test is auth related. Therefore I asked, just want to make sure.

got it thanks. semaphore ci failed to pass due to test timeout. I checked again without changing the auth role logic. I can't trigger it to retest.

I do not know how to trigger retest neither. Did you try with the same test command as semaphore ci? e.g. sudo HOST_TMP_DIR=/tmp TEST_OPTS="GOARCH=386 PASSES='build e2e'" make docker-test

[tangcong]

semaphore ci and travis ci are actually testing different sets of tests. I have not noticed this particular test been flaky in the past. Plus, the test is auth related. Therefore I asked, just want to make sure.

got it thanks. semaphore ci failed to pass due to test timeout. I checked again without changing the auth role logic. I can't trigger it to retest.

I do not know how to trigger retest neither. Did you try with the same test command as semaphore ci? e.g. sudo HOST_TMP_DIR=/tmp TEST_OPTS="GOARCH=386 PASSES='build e2e'" make docker-test

thanks. i will take a try. It seems that this PR merge 3.4 will sometimes cause auth operation timeout, and it looks weird, I need to investigate further.

[tangcong]

@jingyih zap logger may be nil in release-3.4 so it will cause etcd crash and command timeout. fixed it. your judgment is accurate, thanks.

[jingyih]

@jingyih zap logger may be nil in release-3.4 so it will cause etcd crash and command timeout. fixed it. your judgment is accurate, thanks.

Good catch! Thanks! I am thinking if there is a way to auto check for this kind of bug in the future when we backport. Do you have any suggestion? Basically 3.5+ eliminated nil checking on logger, but 3.4 still needs it.

[tangcong]

Good catch! Thanks! I am thinking if there is a way to auto check for this kind of bug in the future when we backport. Do you have any suggestion? Basically 3.5+ eliminated nil checking on logger, but 3.4 still needs it.

it needs a tool such as error check. can we modify cherry-pick script to add some tips?

[jingyih]

LGTM

[jingyih]

@gyuho @mitake @spzala This cherrypick is not trivial. It would be great if any of you could also help take a look.

[spzala]

LGTM

[spzala]

Nit, this is a new file, we should have used year 2020 here (instead of 2015) for an easy future reference of when the file was created. Considering this is cherry pick, doing a separate PR in the master and then cherry pick it would be nice. Thanks!

[spzala]

@gyuho @mitake @spzala This cherrypick is not trivial. It would be great if any of you could also help take a look.

@jingyih awesome job, as usual :), reviewing the original changes and @tangcong great job with the PR. I had a quick look and a small inline comment that can be addressed separately. The build failure seems not related. From cherry pick perspective it looks good to me. Thanks both!