Cleanup issue in cluster mode: namespace gets deleted before other resources, leaving dangling objects

[msmohitkumar]

Description:

Hi John,
Thank you for the great work on pilot-load. I’m using the tool in cluster mode by deploying it as a Kubernetes Deployment using the latest image from this repo.

When I stop the load test by running:

kubectl delete -f load-deployment.yaml

the namespace sometimes gets deleted before the other resources (like simulated applications), leaving those resources in a dangling state. This inconsistency leads to errors and incomplete cleanup of the test environment.

Here’s the configuration file I’m using:

`nodeMetadata: {}
jitter:
workloads: "5s"
config: "5s"
namespaces:

• name: pilot-load
  replicas: 10
  applications:
  • name: big
    replicas: 100
    instances: 10
    type: sidecar
    istio:
    peerAuthentication:
    app: big
    namespace: mesh
    modeIndex: 3
    apiScope: application
    virtualService:
    gateways: ["pilot-load/big"]
    subsets:
    - name: v1
    weight: 80
    - name: v2
    weight: 20
  • name: ingress
    type: router
    replicas: 1
    instances: 0
    gateways:
    name: big
    enabled: true
    kubernetes: false
    nodes:
• name: node
  count: 10`

Request:

• Is there a way to run pilot-load for a fixed duration (e.g., 5 minutes) so that it can automatically stop and clean up instead of relying on kubectl delete?
• If not, would it be feasible to add a duration or ttl option to support timed load generation and safe cleanup?

[howardjohn]

For fixed duration I would probably just utilize bash to send a signal after the time. The cleanup should be safe as long as its given enough time between a SIGTERM and a SIGKILL, or at least i haven't seen any times its not.

We could also try to improve the cleanup ordering, some of the reasons it is the way it is are legacy at this point