driver-looker.template.yaml

# Looker docs: https://github.com/opstrace/opstrace/tree/main/test/test-remote/containers/looker
apiVersion: v1
kind: Secret
metadata:
  name: tenant-auth-__TENANT__
stringData:
  token: '__AUTH_TOKEN__'
---
# We use a StatefulSet solely in order for the pod name to stay constant.
# This avoids creating a new set of metric series in the test data if the pod is restarted or evicted.
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: looker-__TENANT__
spec:
  serviceName: looker-__TENANT__
  replicas: 50
  # We don't care about sequential deploy ordering - just launch pods in parallel
  # (Note: no maxUnavailable for StatefulSet config changes yet, see https://github.com/kubernetes/kubernetes/pull/82162 )
  podManagementPolicy: Parallel
  selector:
    matchLabels:
      name: looker
  template:
    metadata:
      labels:
        name: looker
    spec:
      containers:
      - name: looker
        image: opstrace/looker:8878e85e-dev # with https://github.com/opstrace/opstrace/pull/995
        imagePullPolicy: IfNotPresent
        command:
        - /bin/sh
        - -c
        # invocation-id: Use stable value for series names/labels - across pod restarts
        # n-concurrent-streams: Number of distinct series per instance
        # n-entries-per-stream-fragment: Number of sample points to get from each stream when polling streams
        # n-fragments-per-push-message: Size of chunks to send as separate Push calls, keep reasonable or else Cortex will reject
        # stream-write-n-seconds: Arbitrarily large value to keep running for a long time
        # metrics-time-increment-ms: How wide each fragment should be in milliseconds
        # change-streams-every-n-cycles: Keep metric names/labels constant across the run
        # metrics-past-start-range-min/max-seconds: Start fragments at a fixed 10m10s ago (vs wall clock time).
        #                                           There's currently a 10min age requirement where looker will pause if it gets too close.
        # NOTE: n-entries-per-stream-fragment * metrics-time-increment-ms must be a multiple of 1000
        - >-
          looker https://cortex.${TENANT}.${CLUSTER_NAME}.opstrace.io
          --log-level debug
          --metrics-mode
          --bearer-token-file /var/run/tenant-auth/token
          --skip-read
          --invocation-id ${POD_NAME}
          --n-concurrent-streams 100000
          --n-entries-per-stream-fragment 10
          --n-fragments-per-push-message 5000
          --stream-write-n-seconds 9999999
          --metrics-time-increment-ms 100
          --change-streams-every-n-cycles 0
          --metrics-past-start-range-min-seconds 610
          --metrics-past-start-range-max-seconds 610
        env:
        - name: CLUSTER_NAME
          value: __CLUSTER_NAME__
        - name: TENANT
          value: __TENANT__
        - name: POD_NAME
          valueFrom:
            fieldRef:
              fieldPath: metadata.name
        resources:
          requests:
            memory: 500Mi # no idea what a good number is, just want to set a cap that doesn't OOM
        volumeMounts:
        - name: tenant-auth
          mountPath: /var/run/tenant-auth
          readOnly: true
      volumes:
      - name: tenant-auth
        secret:
          secretName: tenant-auth-__TENANT__