integrations/dcgm/ops_agent_metadata.yaml

platforms:
- type: GCE
  launch_stage: GA
  version: '1'
  install_documentation_url: https://cloud.google.com/stackdriver/docs/solutions/agents/ops-agent/third-party-nvidia
  agent_requirement:
    metrics_minimum_supported_version:
      major: 2
      minor: 38
      patch: 0
  detections:
  - characteristic_metric:
      metric_type: workload.googleapis.com/dcgm.gpu.profiling.sm_utilization
  default_metrics:
  - name: workload.googleapis.com/dcgm.gpu.profiling.sm_utilization
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/dcgm.gpu.profiling.sm_occupancy
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/dcgm.gpu.profiling.pipe_utilization
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - pipe
    - uuid
  - name: workload.googleapis.com/dcgm.gpu.profiling.dram_utilization
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/dcgm.gpu.profiling.pcie_traffic_rate
    value_type: INT64
    kind: GAUGE
    labels:
    - direction
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/dcgm.gpu.profiling.nvlink_traffic_rate
    value_type: INT64
    kind: GAUGE
    labels:
    - direction
    - gpu_number
    - model
    - uuid
- type: GCE
  launch_stage: GA
  version: '2'
  install_documentation_url: https://cloud.google.com/stackdriver/docs/solutions/agents/ops-agent/third-party-nvidia
  agent_requirement:
    metrics_minimum_supported_version:
      major: 2
      minor: 51
      patch: 0
  detections:
  - characteristic_metric:
      metric_type: workload.googleapis.com/gpu.dcgm.memory.bytes_used
  default_metrics:
  - name: workload.googleapis.com/gpu.dcgm.utilization
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.sm.utilization
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.pipe.utilization
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - pipe
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.codec.encoder.utilization
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.codec.decoder.utilization
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.memory.bytes_used
    value_type: INT64
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - state
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.memory.bandwidth_utilization
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.pcie.io
    value_type: INT64
    kind: CUMULATIVE
    labels:
    - direction
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.nvlink.io
    value_type: INT64
    kind: CUMULATIVE
    labels:
    - direction
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.energy_consumption
    value_type: DOUBLE
    kind: CUMULATIVE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.temperature
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.clock.frequency
    value_type: DOUBLE
    kind: GAUGE
    labels:
    - gpu_number
    - model
    - uuid
  - name: workload.googleapis.com/gpu.dcgm.clock.throttle_duration.time
    value_type: DOUBLE
    kind: CUMULATIVE
    labels:
    - gpu_number
    - model
    - uuid
    - violation
  - name: workload.googleapis.com/gpu.dcgm.ecc_errors
    value_type: INT64
    kind: CUMULATIVE
    labels:
    - error_type
    - gpu_number
    - model
    - uuid