integrations/tensorflow-serving/prometheus_metadata.yaml (96 lines of code) (raw):
platforms:
- type: GKE
launch_stage: GA
detections:
- characteristic_metric:
metric_type: prometheus.googleapis.com/:tensorflow:serving:request_count/counter
exporter_metadata:
name: TensorFlow Serving Prometheus Exporter
doc_url: https://www.tensorflow.org/tfx/serving/serving_config#monitoring_configuration
minimum_supported_version: 2.13.1-gpu
default_metrics:
- name: prometheus.googleapis.com/:tensorflow:cc:saved_model:load_attempt_count/counter
prometheus_name: :tensorflow:cc:saved_model:load_attempt_count
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:cc:saved_model:load_latency/counter
prometheus_name: :tensorflow:cc:saved_model:load_latency
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:cc:saved_model:load_latency_by_stage/histogram
prometheus_name: :tensorflow:cc:saved_model:load_latency_by_stage
value_type: DISTRIBUTION
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:direct_session_runs/counter
prometheus_name: :tensorflow:core:direct_session_runs
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:graph_build_calls/counter
prometheus_name: :tensorflow:core:graph_build_calls
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:graph_build_time_usecs/counter
prometheus_name: :tensorflow:core:graph_build_time_usecs
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:graph_optimization_usecs/counter
prometheus_name: :tensorflow:core:graph_optimization_usecs
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:graph_pending_queue_length_histogram/histogram
prometheus_name: :tensorflow:core:graph_pending_queue_length_histogram
value_type: DISTRIBUTION
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:graph_run_input_tensor_bytes/histogram
prometheus_name: :tensorflow:core:graph_run_input_tensor_bytes
value_type: DISTRIBUTION
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:graph_run_output_tensor_bytes/histogram
prometheus_name: :tensorflow:core:graph_run_output_tensor_bytes
value_type: DISTRIBUTION
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:graph_run_time_usecs/counter
prometheus_name: :tensorflow:core:graph_run_time_usecs
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:graph_run_time_usecs_histogram/histogram
prometheus_name: :tensorflow:core:graph_run_time_usecs_histogram
value_type: DISTRIBUTION
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:graph_runs/counter
prometheus_name: :tensorflow:core:graph_runs
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:graph_unused_outputs/counter
prometheus_name: :tensorflow:core:graph_unused_outputs
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:saved_model:read:api/counter
prometheus_name: :tensorflow:core:saved_model:read:api
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:saved_model:read:count/counter
prometheus_name: :tensorflow:core:saved_model:read:count
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:core:saved_model:read:path/gauge
prometheus_name: :tensorflow:core:saved_model:read:path
value_type: DOUBLE
kind: GAUGE
- name: prometheus.googleapis.com/:tensorflow:core:session_created/gauge
prometheus_name: :tensorflow:core:session_created
value_type: DOUBLE
kind: GAUGE
- name: prometheus.googleapis.com/:tensorflow:serving:request_count/counter
prometheus_name: :tensorflow:serving:request_count
value_type: DOUBLE
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:serving:request_latency/histogram
prometheus_name: :tensorflow:serving:request_latency
value_type: DISTRIBUTION
kind: CUMULATIVE
- name: prometheus.googleapis.com/:tensorflow:serving:runtime_latency/histogram
prometheus_name: :tensorflow:serving:runtime_latency
value_type: DISTRIBUTION
kind: CUMULATIVE
install_documentation_url: https://cloud.google.com/stackdriver/docs/managed-prometheus/exporters/tf-serving