Sentry not logging traces for FastAPI's StreamingResponse endpoint

Environment

Steps to Reproduce

I initialised sentry_sdk with

sentry_sdk.init(
        dsn=os.environ['SENTRY_DSN'],
        environment=os.getenv("SENTRY_ENV", "production"),
        send_default_pii=True,
        traces_sample_rate=1.0,
    )

In the Trace Explorer, I can see traces from all endpoints of my app, but not below one:

@app.post("/generate_summary")
def generate_summary(params: GenerateSummaryParams,
                     user: str = Depends(get_current_user)):
    conversation = fetch_conversation(user, params.conversation_id)
    return StreamingResponse(do_generate_summary(conversation, params.doc_ids, user),
                             media_type="text/event-stream")

I tried to manually add transaction inside do_generate_summary() generator like below:

    with sentry_sdk.start_transaction(name="my_transaction"):
        with sentry_sdk.start_span(name="my_span"):
            ...

With above code, something even more strange happens. Now, I can see transactions created by sentry_sdk for my /generate_summary endpoint, but my_transaction transaction and my_span span are missing!

Expected Result

Traces for FastAPI endpoints returning a StreamingResponse should be sent to Sentry.

Actual Result

Traces for FastAPI endpoints returning a StreamingResponse are not being sent to Sentry.

Version

sentry-sdk==2.42.1
fastapi==0.119.1
Python 3.12.12

ENG-5765

Assigning to @getsentry/support for routing ⏲️

PY-1928

@Macok

Hi @Macok,

Thank for your report. The last comment you made is really useful. Based on the remark, it sounds like you are able to get transactions for endpoints returning StreamingResponse, but not together with your do_generate_summary() function.

With above code, something even more strange happens. Now, I can see transactions created by sentry_sdk for my /generate_summary endpoint, but my_transaction transaction and my_span span are missing!

You could be reaching size limits on the ingestion. To confirm can you add debug=True when initializing Sentry, like so

sentry_sdk.init(
      dsn=os.environ['SENTRY_DSN'],
      environment=os.getenv("SENTRY_ENV", "production"),
      send_default_pii=True,
      traces_sample_rate=1.0,
+     debug=True,
  )

Then you should see a message like the following if the serialized transaction is too large:

ERROR: Unexpected status code: 400 (body: b'{"detail":"envelope exceeded size limits for type \'event\' (https://develop.sentry.dev/sdk/envelopes/#size-limits)"}')

That was it! Thanks for quick help!

Inside do_generate_summary, I'm streaming response from OpenAI chunk by chunk, and streaming the chunks to my frontend:

    response = openai.chat.completions.create(
        model=model,
        messages=messages,
        stream=True
    )
    for chunk in response:
        ...
        yield f'event: CHUNK\ndata: {chunk.choices[0].delta.content}'

Apparently, this creates hundreds of spans within Sentry transaction, leading to envelope exceeded size limits error.

Can you suggest some solution?
Maybe there's a way to have my chunk processing logic grouped as a single span which doesn't create any child spans?
Something like:

with sentry_sdk.create_span(name='chunk_processing', allow_child_spans=False):
    # chunk processing logic showed above goes here

Thanks for the answer! But I'm already on 2.42.1.
I was able to fix envelope exceeded size limits by removing the yield line from my code.

    response = openai.chat.completions.create(
        model=model,
        messages=messages,
        stream=True
    )
    for chunk in response:
        ...
        yield f'event: CHUNK\ndata: {chunk.choices[0].delta.content}'   # <-- WORKS FINE WITHOUT THIS LINE

It seems like sentry_sdk creates a span for each chunk of data streamed from my backend to the frontend.

Is this something the Sentry team would consider fixing? Streaming chunks to the frontend is rather common for LLM-powered apps.
For now, do you see any walkaround for me?

@alexander-alderman-webb

@alexander-alderman-webb
Did you manage to take a look at this?

@Macok

Hi @Macok,

We are aware of the problem and there is an initiative on the way that solves the size restriction in the long-term.

In terms of a work-around, you must cut down the size of the transaction holding the spans generated by our integrations for FastAPI and OpenAI.

What you choose to omit depends on which telemetry you care about most. Storing the LLM's response text likely accounts for most of the transaction's size, so I would write a before_send_transaction() callback that truncates the response text.

You can experiment more with the before_send_transaction() callback. Only the return value of the function will be emitted from sentry_sdk.

import sentry_sdk
from sentry_sdk.consts import SPANDATA, OP

 def before_send_transaction(transaction, hint):
    for span in transaction["spans"]:
        if span["op"] == OP.GEN_AI_CHAT:
            span["data"][SPANDATA.GEN_AI_RESPONSE_TEXT] = "Redacted"            

    return transaction

sentry_sdk.init(
      dsn=os.environ['SENTRY_DSN'],
      environment=os.getenv("SENTRY_ENV", "production"),
      send_default_pii=True,
      traces_sample_rate=1.0,
+     debug=True,
+     before_send_transaction=before_send_transaction,
  )

Ok, I'll experiment with before_send_transaction 👍
Thanks for help!

Work is underway to solve this out of the box, in the meantime cutting down on the event size in before_send_transaction is the way to go -- I'll close.

@alexander-alderman-webb

@alexander-alderman-webb Thank you, do we have a possible fix on sentry.io anytime soon?

Navigation Menu

Repository navigation

Sentry not logging traces for FastAPI's StreamingResponse endpoint #5002

Environment

Steps to Reproduce

Expected Result

Actual Result

Version

4 remaining items

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Sentry not logging traces for FastAPI's StreamingResponse endpoint #5002

Description

Environment

Steps to Reproduce

Expected Result

Actual Result

Version

Activity

linear commented on Oct 23, 2025

getsantry commented on Oct 23, 2025

linear commented on Oct 23, 2025

alexander-alderman-webb commented on Oct 24, 2025

Macok commented on Oct 24, 2025

4 remaining items

Macok commented on Oct 24, 2025

Macok commented on Oct 29, 2025

alexander-alderman-webb commented on Oct 29, 2025

Macok commented on Oct 31, 2025

sentrivana commented on Nov 11, 2025

zycon commented on Mar 18, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions