fix(python-sdk): keep streamed request bodies unbuffered across retries

[devin-ai-integration]

Summary

pyqwest's retry middleware keeps a request replayable by mirroring a non-bytes body into memory as it is sent, so a streamed upload through the shared retrying transports (files.write of a file-like object, volume.write_file) reached the wire in chunks yet accumulated its whole body in RAM. curioswitch/pyqwest#219 (released in pyqwest 0.10.0) adds RetryMode.UNBUFFERED, which drops that copy: a streamed body is handed to the next attempt only while nothing has been read from it.

That is exactly what the SDK's connect-only retry policy needs — pyqwest raises the builtin ConnectionError (the only thing should_retry_response retries) only before the request body was written, so the stream is still untouched on every failure we retry. bytes bodies (unary RPCs, in-memory writes) stay replayable in either mode.

Both ConnectionRetryTransports (sync + async) now declare it:

def should_retry_request(self, request: Request) -> RetryMode:
    return RetryMode.UNBUFFERED

The pyqwest pin moves to >=0.10.0,<0.11, the release shipping RetryMode.

Tests (tests/test_retry_stream_buffering.py):

• both transports override should_retry_request with RetryMode.UNBUFFERED (the inherited hook returns True — buffered)
• untouched streams are retried after a simulated connect failure (sync + async)
• peak allocation stays far below body size while streaming (tracemalloc, 16 MiB body), and a stream that failed after its first chunk is not replayed
• the httpx→pyqwest adapters hand the body to the retry middleware as a stream, not flattened bytes

Link to Devin session: https://app.devin.ai/sessions/895a6967064e425abfc84b8cfbc32910
Requested by: @mishushakov

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[changeset-bot]

🦋 Changeset detected

Latest commit: 8bd0e82

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@e2b/python-sdk	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[github-actions]

Package Artifacts

Built from c2b95c9. Download artifacts from this workflow run.

JS SDK (e2b@2.44.2-devin-1787165767-unbuffered-retry-streams.0):

npm install ./e2b-2.44.2-devin-1787165767-unbuffered-retry-streams.0.tgz

CLI (@e2b/cli@2.16.4-devin-1787165767-unbuffered-retry-streams.0):

npm install ./e2b-cli-2.16.4-devin-1787165767-unbuffered-retry-streams.0.tgz

Python SDK (e2b==2.44.0+devin.1787165767.unbuffered.retry.streams):

pip install ./e2b-2.44.0+devin.1787165767.unbuffered.retry.streams-py3-none-any.whl

[mishushakov]

@devin-ai-integration instead of any assume we're on the version of pyqwest that ships the retrymode

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a2d63c7613

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[devin-ai-integration]

Done in fe00f13 — dropped the Any-typed fallback: both transports now from pyqwest.middleware.retry import RetryMode and return RetryMode.UNBUFFERED (typed -> RetryMode), and the buffered-fallback tests are gone.

One dependency note: no PyPI release ships RetryMode yet (latest is 0.9.0), so with the current pyqwest>=0.9.0,<0.10 pin the import fails at runtime/CI. I verified everything against a source build of pyqwest main (lint/ty/tests green). Once the release with curioswitch/pyqwest#219 is out, the pin needs to move — happy to bump it here when you tell me the version.

[linear-code]

SDK-332

[mishushakov]

pyqwest v0.10.0 was released

[devin-ai-integration]

Bumped the pin to pyqwest>=0.10.0,<0.11 in 6c380c4 — lint/ty/tests green locally against the released 0.10.0.