Chaos injection fails for pod-network-partition experiment

[dashashutosh80]

I have been trying to create a pod-network-partition experiment by following the steps mentioned in https://litmuschaos.github.io/litmus/experiments/categories/pods/pod-network-partition/ and by working on the template present in chaoshub.
There are a couple of issues that I have been facing. I have been going through the codebase and found that:

• The experiment itself creates a NetworkPolicy targeting the pods using POD_SELECTOR and NAMESPACE_SELECTOR environment variables
• But the experiment cannot be run without TARGETS environment variable. Without it, the experiment receives the following error:

time="2025-06-11T07:40:56Z" level=info msg="[PreReq]: Getting the ENV for the pod-network-partition experiment"
panic: runtime error: index out of range [0] with length 0

This is due to

litmus-go/pkg/generic/pod-network-partition/environment/environment.go

Line 43 in 07de11e

return app[0].Namespace, app[0].Kind, app[0].Labels[0]

and

litmus-go/pkg/generic/pod-network-partition/environment/environment.go

Line 36 in 07de11e

experimentDetails.AppNS, experimentDetails.AppKind, experimentDetails.AppLabel = getAppDetails()

. These three variables are already being picked up separately at https://github.com/litmuschaos/litmus-go/blob/07de11eeee8e919b930e738f316a1e490aba3b18/pkg/generic/pod-network-partition/environment/environment.go#L19C2-L21C58 , so one of these is redundant.

• Even after setting the TARGETS variable as per format: kind:namespace:labelKey=labelValue this fails (Note that format kind:namespace:pod1Name,pod2Name also fails since this causes the AppDetails.Labels field to be null which throws nil pointer exception in getAppDetails function execution) chaos fails to get injected as it is unable to determine the target pods even if all the correct details are provided.
• This is due to the fact that in https://github.com/litmuschaos/litmus-go/blob/07de11eeee8e919b930e738f316a1e490aba3b18/chaoslib/litmus/pod-network-partition/lib/pod-network-partition.go#L57C5-L57C14 , empty string as target pods is being passed into GetPodList (probably due to the fact that in this experiment chaos is not directly injected into the pods but a networkpolicy is created to implement the effect) which eventually results in execution of

  litmus-go/pkg/utils/common/pods.go

  Line 168 in 07de11e

  podList, err := GetTargetPodsWhenTargetPodsENVNotSet(podAffPerc, clients, chaosDetails)

  calling GetTargetPodsWhenTargetPodsENVNotSet function.
• In GetTargetPodsWhenTargetPodsENVNotSet(), if we have set the kind to pod, the first switch case is executed at

  litmus-go/pkg/utils/common/pods.go

  Line 276 in 07de11e

  case "pod":

  and since we have passed only labels for AppDetails and not pod names, we receive error ErrorTypeTargetSelection

I am happy to submit a PR for this fix.

[celiawa123]

Hi, What litmus version are you using.
we are also using pod-network-partition experiment and encounter several problems. We created an issue to chaos-operator repo litmuschaos/chaos-operator#509.
Then I build a private chaos-operator image to make it able to input applabels and empty appkind.

But then it failed with below error:
If I set applabels and empty appkind, the appkind is set to "KIND" by chaos-operator.
But in litmus-go when appkind is "KIND", it random select target pods in the ns.

And we can't set appkind to 'pod' with error:
The ChaosEngine "pod-network-partition-chaos" is invalid: spec.appinfo.appkind: Invalid value: "pod": spec.appinfo.appkind in body should match '^(^$|deployment|statefulset|daemonset|deploymentconfig|rollout)$'

[dashashutosh80]

@celiawa123 I have been using the latest version only, but the experiment could not proceed in any way (cause of the deadlock situation mentioned in this issue) as target selection failed hence, the experiment could not proceed with chaos injection stage. Besides, since network partition differs from other experiments where chaos is injected into the workloads, I have raised a PR to optimize the workflow.