Trace identifier can clash if same pod is run twice about traceloop

[invidian]

Shouldn't pods have unique IDs? Or maybe use pod ID as identifier? (wild guess, I don't know much how things works)

[pothos]

Ah, yes:

kubectl get pod -o custom-columns=ID:.metadata.uid
ID
c5886398-75b1-11e9-a997-02090d9b1a32

It seems the first part is the only thing that changes for a same-named pod.

[alban]

Straceback can address a pod by cgroup and it should be unique because it contains the metadata.uid. But that's in the http interface and not in inspektor-gadget.

We have to think about usability though. Could we have both options? Specify the pod name, and maybe fail if there are several traces with the same name, and ask about indeed or uid?

[pothos]

Seems it's solved because the list shows unique trace IDs which can be used to get the correct trace when the pod name is the same. So we can close this issue.

If the trace is selected by pod name there is still a clash because lines 869-884 in straceback.go take the first pod found either by explicit matching or found by containerID.

	containerID, _ := sb.podInformer.GetContainerIDFromPod(namespace, podname, containerIndex)

	for i := 0; i < int(C.MaxTracedPrograms); i++ {
		if containerID != "" && sb.tracelets[i].containerID == containerID {
			out, err = sb.DumpProg(uint32(i))
			return
		}
		if sb.tracelets[i].namespace == namespace &&
			sb.tracelets[i].podname == podname &&
			sb.tracelets[i].containeridx == containerIndex {
			out, err = sb.DumpProg(uint32(i))
			return
		}