AWSBedrockInstrumentation: patch clients initialized with boto3.Session

[codefromthecrypt]

Currently, AWSBedrockInstrumentation patches this properly

client = boto3.client("bedrock-runtime")

However, LangChain initializes it a little different, like this:

session = boto3.Session()
client = session.client("bedrock-runtime")

https://github.com/langchain-ai/langchain/blob/master/libs/community/langchain_community/llms/bedrock.py#L403-L426

I think this change will fix it in manual testing, but need second eyes in case I goofed

--- a/src/langtrace_python_sdk/instrumentation/aws_bedrock/instrumentation.py
+++ b/src/langtrace_python_sdk/instrumentation/aws_bedrock/instrumentation.py
@@ -42,6 +42,11 @@ class AWSBedrockInstrumentation(BaseInstrumentor):
             name="client",
             wrapper=patch_aws_bedrock(tracer, version),
         )
+        _W(
+            module="boto3.session",
+            name="Session.client",
+            wrapper=patch_aws_bedrock(tracer, version),
+        )
 
     def _uninstrument(self, **kwargs):
         pass

Once this works, we can resolve langtrace usage inside Elastic chatbot-rag-app here, by removing a workaround:

--- a/example-apps/chatbot-rag-app/api/llm_integrations.py
+++ b/example-apps/chatbot-rag-app/api/llm_integrations.py
@@ -1,6 +1,5 @@
 import os
 
-import boto3
 from langchain_aws import ChatBedrock
 from langchain_cohere import ChatCohere
 from langchain_google_vertexai import ChatVertexAI
@@ -42,7 +41,6 @@ def init_bedrock(temperature):
 
     AWSBedrockInstrumentation().instrument()
     return ChatBedrock(
-        client=boto3.client("bedrock-runtime"),
         model_id=os.getenv("CHAT_MODEL"),
         streaming=True,
         model_kwargs={"temperature": temperature},

[alizenhom]

Thanks @codefromthecrypt for the very detailed issue,

As usual i learn alot from you just pointing out issues -- I get the issue now and it should be a very simple release will keep you posted once i make a new release 🙏

[alizenhom]

Hey @codefromthecrypt,

your approach was correct it indeed worked with langchain wrapping bedrock
kindly check this #460

kindly update the sdk to 3.3.26 and would be more than happy to help!

[codefromthecrypt]

confirmed all good. Thanks!