AWS Bedrock: Monitoring agents
The metrics for the AWS Bedrock agents are currently in Beta mode and can currently display only certain items. You can monitor and access your agents through the Monitoring and Compliance screen.
You can filter out your AWS Bedrock agents based on Provider, accounts, models, and time frames. By default, the time frame selected is the Last 60 minutes UTC. For more information, read Monitoring and compliance.
Prerequisites
To view the metrics for your AWS Bedrock agents, ensure you have Enabled Bedrock Agent metric data access.
Dimensions and metrics
The metric ingestion into Agent Control tower from AWS Bedrock is split into two categories:
- by Agent ARN and Alias ARN
- by ModelIDs
The following table depicts the AWS metrics based on Alias ARN(Amazon Resource Name) used to calculate the metrics depicted in the Agent Control Tower:
AWS metrics based on Alias ARN(Amazon Resource Name)
| Metric name | Calculated by | Description |
|---|---|---|
| TotalTime | OperationOperation, AgentArnOperation, AgentAliasArn | Total time for an operation to complete. |
| Time to first token (TTFT) | OperationOperation, AgentArnOperation, AgentAliasArn | TTFT for a given Operation, agent ARN across all agents and all model IDs |
| InvocationThrottles | OperationOperation, AgentArnOperation, AgentAliasArn | API Level Throttling for a given Operation across all agents and all model IDs and for a given operation, given agent, given model ID. |
| InvocationServerErrors | OperationOperation, AgentArnOperation, AgentAliasArn | InvocationServerErrors for a given Operation across all agents and all model IDs, for a given operation, given agent, across all model IDs and for a given operation, given agent, given model ID. |
| InvocationClientErrors | OperationOperation, AgentArnOperation, AgentAliasArn | InvocationClientErrors for a given Operation across all agents and all model IDs, for a given operation, given agent, across all model IDs and for a given operation, given agent, given model ID. |
| ModelLatency | OperationOperation, AgentArnOperation, AgentArn, PromptTypeOperation, AgentAliasArnOperation, AgentAliasArn, PromptType | Model invocation latency for a given operation across all agents and all model IDs, for a given operation across all model IDs, all prompts and a given agent ARN and for a given operation across all models for a given agent ARN and given promptType. |
| ModelInvocationCount | OperationOperation, AgentArnOperation, AgentArn, PromptTypeOperation, AgentAliasArnOperation, AgentAliasArn, PromptType | Model invocation count for a given operation across all agents and all model IDs, for a given operation across all model IDs, all prompts and a given agent ARN and for a given operation across all models for a given agent ARN and given promptType. |
| ModelInvocationThrottles | OperationOperation, AgentArnOperation, AgentArn, PromptTypeOperation, AgentAliasArnOperation, AgentAliasArn, PromptType | Model invocation throttles for a given operation across all agents and all model IDs, for a given operation across all model IDs, all prompts and a given agent ARN and for a given operation across all models for a given agent ARN and given promptType. |
| ModelInvocationClientErrors | OperationOperation, AgentArnOperation, AgentArn, PromptTypeOperation, AgentAliasArnOperation, AgentAliasArn, PromptType | Model invocation client errors for a given operation across all agents and all model IDs, for a given operation across all model IDs, all prompts and a given agent ARN and for a given operation across all models for a given agent ARN and given promptType. |
| ModelInvocationServerErrors | OperationOperation, AgentArnOperation, AgentArn, PromptTypeOperation, AgentAliasArnOperation, AgentAliasArn, PromptType | Model invocation server errors for a given operation across all agents and all model IDs, for a given operation across all model IDs, all prompts and a given agent ARN and for a given operation across all models for a given agent ARN and given promptType. |
| InputTokenCount | OperationOperation, AgentArnOperation, AgentArn, PromptTypeOperation, AgentAliasArnOperation, AgentAliasArn, PromptType | Input token count for a given operation across all agents and all model IDs, for a given operation across all model IDs, all prompts and a given agent ARN and for a given operation across all models for a given agent ARN and given promptType. |
| OutputTokenCount | OperationOperation, AgentArnOperation, AgentArn, PromptTypeOperation, AgentAliasArnOperation, AgentAliasArn, PromptType | Output token count for a given operation across all agents and all model IDs, for a given operation across all model IDs, all prompts and a given agent ARN and for a given operation across all models for a given agent ARN and given promptType. |
The following metrics are calculated on basis of the AWS Model ID:
AWS metrics based on Model ID
| Metric name | Calculated by | Description |
|---|---|---|
| TotalTime | Operation, ModelId | API total processing time for a given Operation and a specific model ID (across all agents). |
| ModelLatency | Operation, ModelIdOperation, ModelId, PromptType | Model invocation latency for a given operation across all agents for a given model ID and Model invocation latency for a given operation, given modelID, given promptType across all agents. |
| ModelInvocationCount | Operation, ModelIdOperation, ModelId, PromptType | Model invocation count for a given operation across all agents for a given model ID and Model invocation count for a given operation, given modelID, given promptType across all agents. |
| ModelInvocationThrottles | Operation, ModelIdOperation, ModelId, PromptType | Model invocation throttles for a given operation across all agents for a given model ID and Model invocation throttles for a given operation, given modelID, given promptType across all agents. |
| ModelInvocationClientErrors | Operation, ModelIdOperation, ModelId, PromptType | Model invocation client errors for a given operation across all agents for a given model ID and Model invocation client errors for a given operation, given modelID, given promptType across all agents. |
| ModelInvocationServerErrors | Operation, ModelIdOperation, ModelId, PromptType | Model invocation server errors for a given operation across all agents for a given model ID and Model invocation server errors for a given operation, given modelID, given promptType across all agents. |
| InputTokenCount | Operation, ModelIdOperation, ModelId, PromptType | Input Token Count for a given operation across all agents for a given model ID and Input token count for a given operation, given modelID, given promptType across all agents. |
| OutputTokenCount | Operation, ModelIdOperation, ModelId, PromptType | Output token for a given operation across all agents for a given model ID and Output Token Count for a given operation, given modelID, given promptType across all agents. |
