Background

As part of our next release, we would like to support Bedrock, Bedrock Agent, Bedrock Runtime, and Bedrock Agent Runtime attributes in our span attributes. As such, we need to add a patch to our implementation to support these services as ServiceExtensions. This PR adds ServiceExtensions for all the listed services and provides unit tests to ensure they are functioning appropriately, as well as manual test evidence showing the attributes appear as expected.

This PR is based heavily on this one made in our aws-observability/aws-otel-python-instrumentation repo.

Description of Changes

General

• Added Bedrock, Bedrock Agent, Bedrock Runtime, and Bedrock Agent Runtime clients to our devDependencies for unit testing.
• Ran npm install to generate a new package-lock.json file. This removed many references to licenses; unsure if this is intended. Will follow up offline with those who might know why this happened.

Bedrock

• Extract guardrailId from API response, and add into aws.bedrock.guardrail.id span attribute.

Bedrock Agent

• Logic to ensure only one of the following attributes is added based on the incoming operation.
• Extract agentId from both API request and response, and add into aws.bedrock.agent.id span attribute.
• Extract knowledgeBaseId from API request, and add into aws.bedrock.knowledge_base.id" span attribute.
• Extract dataSourceId from both API request and response,, and add into aws.bedrock.data_source.id span attribute.

Bedrock Agent Runtime

• Extract agentId from API request, and add into aws.bedrock.agent.id span attribute.
• Extract knowledgeBaseId from API request, and add into aws.bedrock.knowledgebase.id span attribute.

According to the Python PR, there should only be one of these attributes available for each possible API call.

Bedrock Runtime

• Extract the following attributes and add into span according to Gen AI semantic-conventions:

gen_ai.request.model
gen_ai.system

Populate RemoteResourceType and RemoteResourceIdentifier

The RemoteResourceType and RemoteResourceIdentifier attributes are populated as follows:

"RemoteResourceType": "AWS::Bedrock::Agent"
"RemoteResourceIdentifier": "<agent_id>"

"RemoteResourceType": "AWS::Bedrock::DataSource"
"RemoteResourceIdentifier": "<datasource_id>"

"RemoteResourceType": "AWS::Bedrock::Guardrail"
"RemoteResourceIdentifier": "<guardrail_id>"

"RemoteResourceType": "AWS::Bedrock::KnowledgeBase"
"RemoteResourceIdentifier": "<knowledgeBase_id>"

"RemoteResourceType": "AWS::Bedrock::Model"
"RemoteResourceIdentifier": "<modelId>"

Testing

For this PR, I modified our NodeJS sample application in our Application Signals test framework to add APIs that result in the following traces and metrics for each of the listed use cases.

Traces

Bedrock Agent with Agent ID present (using ListAgentVersions):

Bedrock Agent with Knowledge Base ID present (using GetKnowledgeBase):

Bedrock Agent with Data Source ID present (using GetDataSource):

Bedrock Agent Runtime with Knowledge Base ID present (using Retrieve):

Bedrock with Guardrail ID present (using GetGuardrail):

Bedrock Runtime with Gen AI attributes present (using InvokeModel):

Metrics

Bedrock Agent with Agent ID present (using ListAgentVersions):

Bedrock Agent with Knowledge Base ID present (using GetKnowledgeBase):

Bedrock Agent with Data Source ID present (using GetDataSource):

Bedrock Agent Runtime with Knowledge Base ID present (using Retrieve):

Bedrock with Guardrail ID present (using GetGuardrail):

Bedrock Runtime with Gen AI attributes present (using InvokeModel):

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.