🐛 Bug Report: Bedrock instrumentation: uninstrument() does not remove tracing from clients created while instrumented

Open
#4,577 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

@Aftabbs is already working on this.

Since Oct 7, 2026.

  • #4578 by @Aftabbs — open

Assessment

Difficulty
3/5
Estimated time
1-2 days
Newbie friendliness
68/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Active
Tech stack
aws, python
Domain
observability

Research direction

Start with _wrap and _uninstrument in packages/opentelemetry-instrumentation-bedrock/opentelemetry/instrumentation/bedrock/__init__.py, then inspect the existing Bedrock instrumentation tests for client creation and teardown coverage. Reproduce with a client created before uninstrument() and verify it no longer emits spans afterward, including the async path if covered by the tests.

Written by the indexing model from the issue text.

Description

Which component is this bug for?

Bedrock Instrumentation

📜 Description

After BedrockInstrumentor().uninstrument(), a bedrock-runtime client that was created while the instrumentation was active keeps creating spans. Only clients created after uninstrument() are untraced. Clients are usually created once at start-up, so in practice uninstrument() has no effect on an application's existing clients (and on test suites that instrument / uninstrument around a shared client).

Cause: _wrap replaces the API methods on each new client instance (client.invoke_model = ..., client.converse = ..., and the streaming variants; same in the aiobotocore path), but _uninstrument only unwraps create_client. The replaced methods on existing clients are never restored, and the wrappers do not check whether the instrumentor is still active.

This is different from #4552 (OpenAI unwrap() being a no-op): here create_client is unwrapped correctly, but clients that already exist stay instrumented.

👟 Reproduction steps
pip install boto3 opentelemetry-instrumentation-bedrock opentelemetry-sdk
python repro.py
import json, logging
from importlib.metadata import version
logging.disable(logging.WARNING)

import boto3
from botocore.awsrequest import AWSResponse
from botocore.config import Config
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor
from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter
from opentelemetry.instrumentation.bedrock import BedrockInstrumentor

exporter = InMemorySpanExporter()
provider = TracerProvider()
provider.add_span_processor(SimpleSpanProcessor(exporter))


class Raw:                                            # a canned Converse response, no network
    def __init__(self, data): self.data = data
    def stream(self, *a, **k): yield self.data
    def read(self, amt=None): return self.data
    def close(self): pass
    release_conn = close


def before_send(request, **_):
    out = {"output": {"message": {"role": "assistant", "content": [{"text": "hi"}]}}, "stopReason": "end_turn",
           "usage": {"inputTokens": 3, "outputTokens": 1, "totalTokens": 4}, "metrics": {"latencyMs": 1}}
    return AWSResponse(request.url, 200, {"content-type": "application/json"}, Raw(json.dumps(out).encode()))


def new_client():
    c = boto3.client("bedrock-runtime", region_name="us-east-1", aws_access_key_id="x", aws_secret_access_key="x",
                     config=Config(retries={"total_max_attempts": 1}))
    c.meta.events.register("before-send.bedrock-runtime.*", before_send)
    return c


def spans_for(client):
    exporter.clear()
    client.converse(modelId="anthropic.claude-3-haiku-20240307-v1:0",
                    messages=[{"role": "user", "content": [{"text": "hi"}]}])
    return len(exporter.get_finished_spans())


print("boto3", version("boto3"), "| opentelemetry-instrumentation-bedrock", version("opentelemetry-instrumentation-bedrock"))
BedrockInstrumentor().instrument(tracer_provider=provider)
early = new_client()                                  # created while instrumented (typical: at app start-up)
print("instrumented:   client created before  ->", spans_for(early), "span(s)")
BedrockInstrumentor().uninstrument()
print("uninstrumented: client created before  ->", spans_for(early), "span(s)   (expected 0)")
print("uninstrumented: client created after   ->", spans_for(new_client()), "span(s)   (expected 0)")
👍 Expected behavior

After uninstrument(), no client creates spans, including clients created earlier.

Possible fixes: have the per-client wrappers call straight through to the original method once the instrumentor is uninstrumented (a flag checked at call time), or keep a weakref.WeakSet of instrumented clients and restore their original methods in _uninstrument.

👎 Actual Behavior with Screenshots

The client created before uninstrument() is still traced:

boto3 1.43.108 | opentelemetry-instrumentation-bedrock 0.62.4
instrumented:   client created before  -> 1 span(s)
uninstrumented: client created before  -> 1 span(s)   (expected 0)
uninstrumented: client created after   -> 0 span(s)   (expected 0)
🤖 Python Version

3.12

📃 Provide any additional context for the Bug.

OS: Ubuntu Linux. boto3 / botocore 1.43.108, opentelemetry-instrumentation-bedrock 0.62.4.

👀 Have you spent some time to check if this bug has been raised before?
  • I checked and didn't find similar issue
Are you willing to submit PR?

Yes I am willing to submit a PR!

Dominant language
Python
Stars
7.5k
Forks
1.2k
Avg merge
2d 6h
Merged PRs (30d)
2

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from traceloop/openllmetry

All issues in traceloop/openllmetry

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.