refactor: migrate container_run_log_driver_syslog_test.go to nerdtest.Setup

[ogulcanaydogan]

Closes part of #4613.

What

Migrates cmd/nerdctl/container/container_run_log_driver_syslog_test.go from the legacy testutil.Base / testutil.NewBase pattern to the nerdtest.Setup / test.Case framework, following the same approach used in #4897 (multi_platform_linux_test.go) and #4641.

Design notes

The three tests (TestSyslogNetwork, TestSyslogFacilities, TestSyslogFormat) each expand a cross-product of parameters into independent sub-cases via buildSyslogSubTests:

• Each sub-case manages its own syslog server lifecycle (start in Setup, receive in Cleanup) using closure-captured variables (addr, done, closer, containerName, tag, msg), which is necessary because the channel-based frame validation cannot be expressed through data.Labels().
• CA and cert are shared from the outer fixture to sub-cases via pointer-to-pointer (**testca.CA, **testca.Cert) populated in the outer testCase.Setup.
• testca.New requires testing.TB; it is passed as *testing.T through the newSyslogTestCase(t *testing.T) helper rather than helpers.T(), which only implements the narrower tig.T interface.
• Network skip logic respects the rootless case with a distinct message per the existing convention in the repo.

Validator functions (rfc5424Validator, rfc3164Validator, emptyFormatValidator) are extracted as package-level helpers; their logic is identical to the originals.

Test plan

• go vet ./cmd/nerdctl/container/... passes with no errors.
• Logic is a mechanical translation of the existing test; no behaviour changes.

[AkihiroSuda]

=== Failing tests ===
TestLoadStdinFromPipe
TestSyslogNetwork
TestSyslogNetwork/udp_1_rfc5424
TestSyslogNetwork/udp_user_rfc5424
TestSyslogNetwork/unixgram_1_rfc5424
TestSyslogNetwork/unixgram_user_rfc5424

https://github.com/containerd/nerdctl/actions/runs/25862295284/job/75995732134?pr=4898

[ogulcanaydogan]

Thanks for catching this. The root cause: runPacketSyslog has a 4x100ms read window (~400ms total). Running the subtests in parallel (Tigron default) puts the arm runner under load, pushing container startup past that window so the server drains before the log entry arrives. Fixed by setting NoParallel: true on each case, matching the original sequential t.Run behaviour.

[ogulcanaydogan]

Thanks for catching this. The root cause is in runPacketSyslog: UDP and unixgram use a deadline-based read loop with a 4x100ms (400ms) hard cap. On slow ARM runners, container startup plus syslog driver initialization exceeds that budget, so the server closes the socket and sends an empty string on the channel before any datagram arrives.

Fixed in 66c2358 by increasing the pre-packet retry count from 3 to 20 (2s total). Once the first datagram arrives, it resets to 3 retries (400ms drain) to avoid lingering unnecessarily.

[sathiraumesh]

we can just create a combination structure first and then later iterate over and run over them. I think its much readable and better

type params struct {
		network      string
		fPriK        string
		fPriV        syslog.Priority
		fmtK         string
		fmtValidFunc func(string, string, string, string, syslog.Priority, bool) error
	}
	var cobintaions []params
	for _, network := range networks {
		for rFK, rFV := range syslogFacilities {
			for _, fPriK := range []string{rFK, strconv.Itoa(int(rFV) >> 3)} {
				for fmtK, fmtValidFunc := range fmtValidFuncs {
					combos = append(combos, params{network, fPriK, rFV, fmtK, fmtValidFunc})
				}
			}
		}
	}

[sathiraumesh]

Can use exit code constants from the tigron framework