windows: system connection: unexpected ports

[edsantiago]

  start machine with conflict on SSH port
...
[FAILED] Expected
      <[]string | len:4, cap:4>: ["50443", "50443", "50229", "50229"]    <<<<< 50443 is in there
  to contain element matching
      <string>: 50443
  In [It] at: C:/Users/Administrator/AppData/Local/cirrus-ci-build/repo[/pkg/machine/e2e/start_test.go:113](https://github.com/containers/podman/blob/1093ebb72be3117c5312b702ea89ecfcab5d3452/pkg/machine/e2e/start_test.go#L113) @ 05/28/24 12:52:44.223

Maybe it's a windows ^M thing?
My misunderstanding. The error message means "I want 50443 and ONLY 50443, nothing else". The 50229 is what's causing the problem.

• windows : machine-hyperv podman windows rootless host sqlite
  • PR test/system: make some tests faster part 1 #22821
    • 05-28 08:59 in podman machine start C:/Users/Administrator/AppData/Local/cirrus-ci-build/repo/pkg/machine/e2e/start_test.go:16 start machine with conflict on SSH port C:/Users/Administrator/AppData/Local/cirrus-ci-build/repo/pkg/machine/e2e/start_test.go:94
  • PR update golangci-lint to v1.59.0 #22815
    • 05-27 07:03 in podman machine start C:/Users/Administrator/AppData/Local/cirrus-ci-build/repo/pkg/machine/e2e/start_test.go:16 start machine with conflict on SSH port C:/Users/Administrator/AppData/Local/cirrus-ci-build/repo/pkg/machine/e2e/start_test.go:94
  • PR libpod: wait another interval for healthcheck #22764
    • 05-27 08:07 in podman machine start C:/Users/Administrator/AppData/Local/cirrus-ci-build/repo/pkg/machine/e2e/start_test.go:16 start machine with conflict on SSH port C:/Users/Administrator/AppData/Local/cirrus-ci-build/repo/pkg/machine/e2e/start_test.go:94
  • PR [v5.1] applehv: Rosetta support #22757
    • 05-20 11:21 in podman machine start C:/Users/Administrator/AppData/Local/cirrus-ci-build/repo/pkg/machine/e2e/start_test.go:16 start machine with conflict on SSH port C:/Users/Administrator/AppData/Local/cirrus-ci-build/repo/pkg/machine/e2e/start_test.go:94

x	x	x	x	x	x
machine-hyperv(4)	podman(4)	windows(4)	rootless(4)	host(4)	sqlite(4)

[Luap99]

Maybe it's a windows ^M thing?

Mhh yeah some non printable char sounds logical but why would that flake? I guess such things should be consistent.

Still seeing this? I guess I can throw up a PR to trim spaces from the string and then we can see if it sill reproduces in the future.

[Luap99]

Actually I now looked at the code and the issue is not a invisible char. The code matches
Expect(connectionPorts).To(HaveEach(inspectPort))

This means all connections should have the same port. And per the code there we only created a single machine so why do we have connections from another machine still in there? This sounds to be more like an issue with invalid cleanup.

Given that other tests in the log failed with

  Error: unable to connect to Podman socket: failed to read identity "C:\\Users\\ADMINI~1\\AppData\\Local\\Temp\\podman_test2399464602\\.local\\share\\containers\\podman\\machine\\machine": open C:\Users\ADMINI~1\AppData\Local\Temp\podman_test2399464602\.local\share\containers\podman\machine\machine: The system cannot find the path specified.

So I assume something more general is broken in the windows testing, I would think that relates directly to #22843.

[edsantiago]

Still happening as of last week. This is an infrequent one:

• PR Update module github.com/containernetworking/plugins to v1.5.1 #23024
  • 06-17 18:30 in podman machine start start machine with conflict on SSH port

[edsantiago]

Can someone instrument

podman/cmd/podman/machine/init.go

Lines 216 to 217 in 045413b

	newMachineEvent(events.Init, events.Event{Name: initOpts.Name})
	fmt.Println("Machine init complete")

so it spits out the system connection info?

[Luap99]

I can but what is the point? We can add podman system connection list to tests Befre/AfterEach if you want to debug this. But given all logs so far show at least one other failure before I am pretty sure that issue is caused by the failing cleanup in the other tests.

Most likely the best fix is to make use of PODMAN_CONNECTIONS_CONF env to use a new tmp file for each test as connection storage, that way we can ensure it never leaks.

[Luap99]

Looking at it there are more problems in teardown()

One weird failure that is not shown at all because it is not considered fatal

  C> podman.exe machine rm --force ac24eebe6478
  Error: failure advancing enumeration (2147749924)