platform/qemu: detect if QEMU process exits unexpectedly

[jlebon]

Currently, we only try to detect if the QEMU process exited by actually wait()ing for it in the kola qemuexec path. We should do it in the kola testing path as well so that it's easy to tell if e.g. it was killed while the test was running.

Note this doesn't actually stop the test early if QEMU exited. That would require some tricky wiring into the harness. But at least what it prints helps diagnose the issue when we see the test time out on SSH. And the QEMU process won't just hang there as defunct.

[dustymabe]

I observe this properly in a test pipeline run:

[2024-09-03T19:14:08.996Z] 2024-09-03T19:14:06Z platform/machine/qemu: QEMU process finished abnormally: signal: killed

[dustymabe]

Note this doesn't actually stop the test early if QEMU exited. That would require some tricky wiring into the harness. But at least what it prints helps diagnose the issue when we see the test time out on SSH. And the QEMU process won't just hang there as defunct.

I wonder if we should add this as a comment in the code somewhere (i.e. explaining the more ideal future state).

[dustymabe]

I'm guessing we don't need this code from the other day then?

$ git diff mantle/platform/qemu.go
diff --git a/mantle/platform/qemu.go b/mantle/platform/qemu.go
index 6fe76da9f..5106b42b5 100644
--- a/mantle/platform/qemu.go
+++ b/mantle/platform/qemu.go
@@ -208,7 +208,9 @@ func (inst *QemuInstance) SSHAddress() (string, error) {
 
 // Wait for the qemu process to exit
 func (inst *QemuInstance) Wait() error {
-       return inst.qemu.Wait()
+       r := inst.qemu.Wait()
+       plog.Debugf("Waited for qemu. r=%v", r)
+       return r
 }
 
 // WaitIgnitionError will only return if the instance

[jlebon]

I'm guessing we don't need this code from the other day then?

Yeah, we don't need it anymore.

[dustymabe]

LGTM