[release-3.4] .github/workflows: Remove tests.yaml

[joshjms]

All the tests inside tests.py has already been migrated to Prow 🥳

Ref: kubernetes/test-infra#32754

/cc @ivanvc @abdurrehman107

Please read https://github.com/etcd-io/etcd/blob/main/CONTRIBUTING.md#contribution-flow.

[joshjms]

/hold

Let's wait for kubernetes/test-infra#34916 to be merged

[joshjms]

/retest

[ivanvc]

/retitle [release-3.4] .github/workflows: Remove tests.yaml

[ivanvc]

/close testing

[ivanvc]

/close

[joshjms]

/test pull-etcd-e2e-amd64

[ivanvc]

Noting the --- FAIL: TestConnectionMultiplexing (5.41s) error. Let's retry the job.

/retest

[ivanvc]

@jmhbnz, do you recall any issues you encountered during the initial migration of AMD64 jobs with the e2e tests? It passes with the same arguments in GitHub runners (i.e., I altered the GitHub workflow to match what we have in Prow in this PR: #20129), but fails in the Prow infrastructure.

[joshjms]

The error:

../../bin/etcd-18074: 2025-06-05 19:14:07.393977 I | embed: rejected connection from "127.0.0.1:45890" (error "tls: client requested unsupported application protocols ([http/1.0])", ServerName "")

/usr/bin/curl-18123: curl: (35) OpenSSL/3.0.16: error:0A000460:SSL routines::tlsv1 alert no application protocol

Comparing with main, this seems to be an issue when curl-ing using HTTP 1.0 as 1.0 is not tested in main branch.

In main:

t.Run("curl", func(t *testing.T) {
	for _, httpVersion := range []string{"2", "1.1", ""} {
		tname := "http" + httpVersion
		if httpVersion == "" {
			tname = "default"
		}
		t.Run(tname, func(t *testing.T) {
			assert.NoError(t, fetchGRPCGateway(httpEndpoint, httpVersion, connType))
			assert.NoError(t, fetchMetrics(t, httpEndpoint, httpVersion, connType))
			assert.NoError(t, fetchVersion(httpEndpoint, httpVersion, connType))
			assert.NoError(t, fetchHealth(httpEndpoint, httpVersion, connType))
			assert.NoError(t, fetchDebugVars(httpEndpoint, httpVersion, connType))
		})
	}
})

In release-3.4

t.Run("curl", func(t *testing.T) {
	for _, httpVersion := range []string{"2", "1.1", "1.0", ""} {
		tname := "http" + httpVersion
		if httpVersion == "" {
			tname = "default"
		}
		t.Run(tname, func(t *testing.T) {
			assert.NoError(t, fetchGrpcGateway(httpEndpoint, httpVersion, connType))
			assert.NoError(t, fetchMetrics(httpEndpoint, httpVersion, connType))
			assert.NoError(t, fetchVersion(httpEndpoint, httpVersion, connType))
			assert.NoError(t, fetchHealth(httpEndpoint, httpVersion, connType))
			assert.NoError(t, fetchDebugVars(httpEndpoint, httpVersion, connType))
		})
	}
})

However, this problem only happens in Prow as the Github actions workflow works fine and running it locally also succeeds. Using pj-on-kind shows the same error as the one in Prow (link).

[joshjms]

I think it's related to this #16516

See the conversation here #16513

From @ahrtr (sorry for the tag)

This PR is breaking FAIL: TestConnectionMultiplexing/ServerTLS/ClientTLS/curl/http1.0.

It looks like http 1.0 isn't supported any more in go:1.20-bookworm. Since HTTP 1.0 has already been obsoleted, it's OK to update the test case.

From @jmhbnz

Please remove the "1.0" entry in this line:

etcd/tests/e2e/cmux_test.go

Line 124 in e4e05c5
for _, httpVersion := range []string{"2", "1.1", "1.0", ""} {

[ivanvc]

@joshjms, I can think of two solutions:

1. We run the tests on the Prow infra using a different image, which will require a different job configuration for release-3.4 e2e jobs
2. We keep this job as a GitHub action/workflow

Given the timeline to give support to this version, I lean towards option 2.

That would mean:

1. Keep the tests.yaml file, simplify it, so it only runs e2e tests
2. Revert etcd: Add e2e tests for branch release-3.4 kubernetes/test-infra#34916

@jmhbnz, thoughts?

[joshjms]

@ivanvc I think according to the conversation in the PR, it would be better to obsolete HTTP 1.0 (i.e. a partial backport of #16516). What do you think?

[ivanvc]

@ivanvc I think according to the conversation in the PR, it would be better to obsolete HTTP 1.0 (i.e. a partial backport of #16516). What do you think?

I don't think it makes sense to change a stable release.

[joshjms]

@ivanvc I think according to the conversation in the PR, it would be better to obsolete HTTP 1.0 (i.e. a partial backport of #16516). What do you think?

I don't think it makes sense to change a stable release.

Ah I see, makes sense.

[jmhbnz]

Hey team - We have already backported the fix to 3.5 in #16568 so I would have no issue with backporting to 3.4 if that means we can modernize our testing infrastructure.

[ivanvc]

@joshjms, let's follow James' suggestion. I didn't realize until later that the change is in the tests package.

Let's backport #16568 first, and then we continue with this pull request. Thanks! :)

[joshjms]

@joshjms, let's follow James' suggestion. I didn't realize until later that the change is in the tests package.

Let's backport #16568 first, and then we continue with this pull request. Thanks! :)

Got it

[ivanvc]

/test pull-etcd-e2e-amd64

[ivanvc]

@joshjms, can we enable e2e periodic and postsubmits before we consider merging this pull request? Thanks.

[ivanvc]

e2e periodics look fine: https://testgrid.k8s.io/sig-etcd-periodics#ci-etcd-e2e-release34-amd64.

The rest of the jobs are migrated into Prow.

[ivanvc]

LGTM. Thanks, @joshjms.

/cc @jmhbnz

[jmhbnz]

LGTM - Thanks team

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: ivanvc, jmhbnz, joshjms

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [jmhbnz]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment