Template in monitor/

position: both
pod-template:
  image: monitor
  imagePullPolicy: Always
  env:
    - name: INTERVAL
      value: "5"
    - name: MONITORS
      value: >
        ['memory_capacity','cpu_usage', 'memory_usage', 'gpu_usage', 'available_bandwidth',  'gpu_flops', 'model_flops', 'model_memory','queue_state']
port-open:
  pos: cloud
  port: 9000

Policies that consume invocation-level queue state use monitor-queue-state.yaml, which reports only queue_state at a shorter interval. Queue state includes ordered waiting identities and the current running/processing phase rather than a single queue-length value.

←Back to Template List