Google Cloud Vertex AI

Introduction

Google Cloud Vertex AI is a unified machine learning platform for building, deploying, and scaling machine learning models and generative AI applications. OpsRamp discovers and monitors Vertex AI resources, including endpoints, deployed models, models and model versions, pipeline jobs, feature online stores, indexes, index endpoints, and reasoning engines. It also collects Cloud Monitoring metrics from the aiplatform.googleapis.com service, providing visibility into resource health, performance, and utilization.

Use the Google Cloud integration to discover Vertex AI resources and collect associated performance and operational metrics.

Setup

Before discovering Vertex AI resources, ensure that the OpsRamp Google Cloud (GCP) Integration is configured.

To enable Vertex AI resource discovery, select the Vertex AI resource types in the GCP Integration Discovery Profile.

OpsRamp collects metrics from Google Cloud Monitoring under the aiplatform.googleapis.com metric domain.

Ensure that the integration service account has the following permissions:

aiplatform.*.list
aiplatform.*.get
monitoring.timeSeries.list

Supported resource types

Resource Type (Native Type)	Resource Tag	Description
GCP Vertex AI Endpoint	GCP_VERTEX_ENDPOINT	Online prediction endpoints serving deployed models.
GCP Vertex AI Deployed Model	GCP_VERTEX_DEPLOYED_MODEL	A model deployed to an endpoint for serving.
GCP Vertex AI Model	GCP_VERTEX_MODEL	Registered models in the Vertex AI Model Registry.
GCP Vertex AI Model Version	GCP_VERTEX_MODEL_VERSION	Individual versions of a registered model.
GCP Vertex AI Pipeline Job	GCP_VERTEX_PIPELINE_JOB	Vertex AI Pipelines execution runs.
GCP Vertex AI Feature Online Store	GCP_VERTEX_FEATURE_STORE	Feature online stores for low-latency serving.
GCP Vertex AI Index	GCP_VERTEX_INDEX	Vector Search indexes.
GCP Vertex AI Index Endpoint	GCP_VERTEX_INDEX_ENDPOINT	Endpoints serving deployed Vector Search indexes.
GCP Vertex AI Reasoning Engine	GCP_VERTEX_REASONING_ENGINE	Managed runtime for agent reasoning and memory.

Supported metrics

Endpoint

OpsRamp Metric	GCP Metric	Metric Display Name	Unit	Aggregation Type	Description
google_vertex_endpoint_prediction_online_accelerator_duty_cycle	prediction/online/accelerator/duty_cycle	Accelerator duty cycle	%	Average	Average fraction of time the accelerator(s) were actively processing.
google_vertex_endpoint_prediction_online_accelerator_memory_bytes_used	prediction/online/accelerator/memory/bytes_used	Accelerator memory usage	count	Average	Amount of accelerator memory allocated to the deployed model replica.
google_vertex_endpoint_prediction_online_cpu_utilization	prediction/online/cpu/utilization	CPU Utilization	%	Average	Fraction of CPU allocated by the deployed model replica and currently in use.
google_vertex_endpoint_prediction_online_error_count	prediction/online/error_count	Error count	count	Sum	Number of online prediction errors.
google_vertex_endpoint_prediction_online_machine_count	prediction/online/machine_count	Machine count	count	Average	Number of machines used by the deployed model.
google_vertex_endpoint_prediction_online_memory_bytes_used	prediction/online/memory/bytes_used	Memory usage	count	Average	Amount of memory allocated to the deployed model replica and currently in use.
google_vertex_endpoint_prediction_online_network_received_bytes_count	prediction/online/network/received_bytes_count	Network bytes received	count	Sum	Bytes received over the network by the deployed model replica.
google_vertex_endpoint_prediction_online_network_sent_bytes_count	prediction/online/network/sent_bytes_count	Network bytes sent	count	Sum	Bytes sent over the network by the deployed model replica.
google_vertex_endpoint_prediction_online_prediction_count	prediction/online/prediction_count	Prediction count	count	Sum	Number of online predictions.
google_vertex_endpoint_prediction_online_prediction_latencies	prediction/online/prediction_latencies	Prediction Latencies	ms	Average	Online prediction latency of the deployed model.
google_vertex_endpoint_prediction_online_private_prediction_latencies	prediction/online/private/prediction_latencies	Private endpoint prediction latencies	ms	Average	Online prediction latency of the private deployed model.
google_vertex_endpoint_prediction_online_private_response_count	prediction/online/private/response_count	Private endpoint response count	count	Sum	Number of online prediction responses for the private deployed model.
google_vertex_endpoint_prediction_online_replicas	prediction/online/replicas	Replica count	count	Average	Number of active replicas used by the deployed model.
google_vertex_endpoint_prediction_online_response_count	prediction/online/response_count	Response count	count	Sum	Number of different online prediction response codes.
google_vertex_endpoint_prediction_online_target_replicas	prediction/online/target_replicas	Replica target	count	Average	Target number of active replicas needed for the deployed model.
google_vertex_endpoint_tuned_model_online_serving_consumed_token_throughput	tuned_model/online_serving/consumed_token_throughput	Token Throughput	count	Sum	Overall throughput in terms of tokens for GenAI tuned models.
google_vertex_endpoint_tuned_model_online_serving_first_token_latencies	tuned_model/online_serving/first_token_latencies	First token latencies	ms	Average	Duration from request received to first token sent back for GenAI tuned models.
google_vertex_endpoint_tuned_model_online_serving_model_invocation_latencies	tuned_model/online_serving/model_invocation_latencies	Model invocation latencies	ms	Average	Model invocation latencies for fine-tuned GenAI models.
google_vertex_endpoint_tuned_model_online_serving_tokens	tuned_model/online_serving/tokens	Tokens	count	Average	Input and output token count distribution for GenAI tuned models.

Model

OpsRamp Metric	GCP Metric	Metric Display Name	Unit	Aggregation Type	Description
google_vertex_model_feature_attribution_deviation	model_monitoring/feature_attribution_deviation	Feature Attribution Deviation	count	Average	The feature attribution deviation value of the given feature.
google_vertex_model_feature_drift_deviation	model_monitoring/feature_drift_deviation	Feature Drift Deviation	count	Average	The drift deviation value of the given feature.
google_vertex_model_prediction_output_drift_deviation	model_monitoring/prediction_output_drift_deviation	Prediction Output Drift Deviation	count	Average	The drift deviation value of the prediction output.

Index Endpoint

OpsRamp Metric	GCP Metric	Metric Display Name	Unit	Aggregation Type	Description
google_vertex_matching_engine_query_request_count	matching_engine/query/request_count	Query Request Count	count	Sum	Number of requests received by matching engine.
google_vertex_matching_engine_query_latencies	matching_engine/query/latencies	Query Latencies	ms	Distribution	Server side query latency.
google_vertex_matching_engine_current_replicas	matching_engine/current_replicas	Current Replicas	count	Gauge	Number of active replicas used by the DeployedIndex.
google_vertex_matching_engine_current_shards	matching_engine/current_shards	Current Shards	count	Gauge	Number of shards of the DeployedIndex.
google_vertex_matching_engine_cpu_request_utilization	matching_engine/cpu/request_utilization	CPU Request Utilization	%	Gauge	CPU utilization of match server containers.
google_vertex_matching_engine_memory_used_bytes	matching_engine/memory/used_bytes	Memory Used Bytes	count	Gauge	Memory used by match server container.

Feature Online Store

OpsRamp Metric	GCP Metric	Metric Display Name	Unit	Aggregation Type	Description
google_vertex_featureonlinestore_online_serving_request_count	featureonlinestore/online_serving/request_count	Request Count	count	Sum	Number of online serving requests received.
google_vertex_featureonlinestore_serving_bytes_count	featureonlinestore/online_serving/serving_bytes_count	Serving Bytes Count	count	Sum	Response bytes count from Feature Online Store.
google_vertex_featureonlinestore_online_serving_latencies	featureonlinestore/online_serving/serving_latencies	Online Serving Latencies	ms	Average	Online serving latencies for Feature Online Store.
google_vertex_featureonlinestore_running_sync	featureonlinestore/running_sync	Running Sync	count	Gauge	Number of running syncs at a given time.
google_vertex_featureonlinestore_serving_data_ages	featureonlinestore/serving_data_ages	Serving Data Ages	Distribution	Gauge	Age of serving data in seconds.
google_vertex_featureonlinestore_serving_data_by_sync_time	featureonlinestore/serving_data_by_sync_time	Serving Data By Sync Time	count	Gauge	Breakdown of serving data by sync time.
google_vertex_featureonlinestore_bigtable_cpu_load	featureonlinestore/storage/bigtable_cpu_load	Bigtable CPU Load	%	Gauge	Average CPU load of Bigtable nodes.
google_vertex_featureonlinestore_bigtable_cpu_load_hottest_node	featureonlinestore/storage/bigtable_cpu_load_hottest_node	Bigtable Hottest Node CPU Load	%	Gauge	CPU load of hottest Bigtable node.
google_vertex_featureonlinestore_bigtable_nodes	featureonlinestore/storage/bigtable_nodes	Bigtable Nodes	count	Gauge	Number of Bigtable nodes.
google_vertex_featureonlinestore_multi_region_bigtable_cpu_load	featureonlinestore/storage/multi_region_bigtable_cpu_load	Multi-region Bigtable CPU Load	%	Gauge	CPU load of the multi-region Bigtable instance.
google_vertex_featureonlinestore_multi_region_bigtable_nodes	featureonlinestore/storage/multi_region_bigtable_nodes	Multi-region Bigtable Nodes	count	Gauge	Node count for multi-region Bigtable.
google_vertex_featureonlinestore_optimized_nodes	featureonlinestore/storage/optimized_nodes	Optimized Nodes	count	Gauge	Number of optimized nodes.
google_vertex_featureonlinestore_stored_bytes	featureonlinestore/storage/stored_bytes	Stored Bytes	count	Gauge	Bytes stored in Feature Online Store.

Pipeline Job

OpsRamp Metric	GCP Metric	Metric Display Name	Unit	Aggregation Type	Description
google_vertex_pipelinejob_duration	pipelinejob/duration	Pipeline Job Duration	ms	Average	Pipeline job run duration.
google_vertex_pipelinejob_task_completed_count	pipelinejob/task_completed_count	Pipeline Task Completed Count	count	Sum	Number of completed pipeline tasks.

Index

OpsRamp Metric	GCP Metric	Metric Display Name	Unit	Aggregation Type	Description
google_vertex_index_stream_update_datapoint_count	matching_engine/stream_update/datapoint_count	Datapoint Count	count	Sum	Number of successfully upserted or removed datapoints.
google_vertex_index_stream_update_latencies	matching_engine/stream_update/latencies	Stream Update Latencies	ms	Average	Latency between the user receiving a response and that update taking effect.
google_vertex_index_stream_update_request_count	matching_engine/stream_update/request_count	Request Count	count	Sum	Number of stream update requests.

Reasoning Engine

OpsRamp Metric	GCP Metric	Metric Display Name	Unit	Aggregation Type	Description
google_vertex_reasoning_engine_cpu_allocation_time	reasoning_engine/cpu/allocation_time	Container CPU Allocation	seconds	Sum	Container CPU allocation in seconds.
google_vertex_reasoning_engine_memory_allocation_time	reasoning_engine/memory/allocation_time	Container Memory Allocation	GB-seconds	Sum	Container memory allocation in Gigabytes-seconds.
google_vertex_reasoning_engine_generate_memories_token_count	reasoning_engine/memory_bank/generate_memories_token_count	Generate Memories Token Count	count	Sum	Number of tokens used in memory generation requests per Reasoning Engine.
google_vertex_reasoning_engine_memory_lro_latency	reasoning_engine/memory_bank/memory_lro_latency	Memory LRO Latency	ms	Average	Total latency of memory LROs.
google_vertex_reasoning_engine_memory_mutation_count	reasoning_engine/memory_bank/memory_mutation_count	Memory Mutation Count	count	Sum	Number of memory mutation operations on the Reasoning Engine.
google_vertex_reasoning_engine_memory_retrieval_count	reasoning_engine/memory_bank/memory_retrieval_count	Retrieved Memories Count	count	Sum	Number of memory entities retrieved per Reasoning Engine.
google_vertex_reasoning_engine_request_count	reasoning_engine/request_count	Request Count	count	Sum	Number of requests reaching the instance.
google_vertex_reasoning_engine_request_latencies	reasoning_engine/request_latencies	Request Latency	ms	Average	Distribution of request latency in milliseconds reaching the instance.

External reference

Vertex AI monitoring metrics