# In-cluster load generator. Unlike `kubectl port-forward` (which pins to a
# single pod), hitting the ClusterIP Service load-balances across all replicas —
# so scale-out actually drains the queue. 30 parallel pods, each looping
# streaming requests for ~80s.
apiVersion: batch/v1
kind: Job
metadata:
  name: mock-load
  namespace: inference
spec:
  parallelism: 30
  completions: 30
  backoffLimit: 0
  ttlSecondsAfterFinished: 120
  template:
    spec:
      restartPolicy: Never
      containers:
        - name: load
          image: curlimages/curl:8.10.1
          command: ["/bin/sh", "-c"]
          args:
            - |
              end=$(( $(date +%s) + 80 ))
              while [ $(date +%s) -lt $end ]; do
                curl -s -m 60 -X POST http://mock-vllm.inference.svc:8000/v1/chat/completions \
                  -H 'content-type: application/json' \
                  -d '{"messages":[{"role":"user","content":"hi"}],"max_tokens":200,"stream":true}' >/dev/null
              done