integrations/torchserve/prometheus_metadata.yaml (98 lines of code) (raw):
platforms:
- type: GKE
launch_stage: GA
detections:
- characteristic_metric:
metric_type: prometheus.googleapis.com/Requests2XX/counter
- characteristic_metric:
metric_type: prometheus.googleapis.com/ts_inference_requests_total/gauge
exporter_metadata:
name: TorchServe Prometheus Exporter
doc_url: https://pytorch.org/serve/metrics.html
minimum_supported_version: v0.8.1
default_metrics:
- name: prometheus.googleapis.com/CPUUtilization/gauge
prometheus_name: CPUUtilization
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/DiskAvailable/gauge
prometheus_name: DiskAvailable
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/DiskUsage/gauge
prometheus_name: DiskUsage
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/DiskUtilization/gauge
prometheus_name: DiskUtilization
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/GPUMemoryUsed/gauge
prometheus_name: GPUMemoryUsed
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/GPUMemoryUtilization/gauge
prometheus_name: GPUMemoryUtilization
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/GPUUtilization/gauge
prometheus_name: GPUUtilization
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/HandlerTime/gauge
prometheus_name: HandlerTime
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/MemoryAvailable/gauge
prometheus_name: MemoryAvailable
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/MemoryUsed/gauge
prometheus_name: MemoryUsed
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/MemoryUtilization/gauge
prometheus_name: MemoryUtilization
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/PredictionTime/gauge
prometheus_name: PredictionTime
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/QueueTime/gauge
prometheus_name: QueueTime
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/Requests2XX/gauge
prometheus_name: Requests2XX
kind: CUMULATIVE
value_type: DOUBLE
- name: prometheus.googleapis.com/Requests4XX/gauge
prometheus_name: Requests4XX
kind: CUMULATIVE
value_type: DOUBLE
- name: prometheus.googleapis.com/Requests5XX/gauge
prometheus_name: Requests5XX
kind: CUMULATIVE
value_type: DOUBLE
- name: prometheus.googleapis.com/WorkerLoadTime/gauge
prometheus_name: WorkerLoadTime
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/WorkerThreadTime/gauge
prometheus_name: WorkerThreadTime
kind: GAUGE
value_type: DOUBLE
- name: prometheus.googleapis.com/ts_inference_latency_microseconds/gauge
prometheus_name: ts_inference_latency_microseconds
kind: CUMULATIVE
value_type: DOUBLE
- name: prometheus.googleapis.com/ts_inference_requests_total/gauge
prometheus_name: ts_inference_requests_total
kind: CUMULATIVE
value_type: DOUBLE
- name: prometheus.googleapis.com/ts_queue_latency_microseconds/gauge
prometheus_name: ts_queue_latency_microseconds
kind: CUMULATIVE
value_type: DOUBLE
install_documentation_url: https://cloud.google.com/stackdriver/docs/managed-prometheus/exporters/torchserve