fix(python-sdk): use thread-local API transports

[matthewlouisbrockman]

makes the sync python python API http transport cache thread-local to handle unsafe usage of the shared transport under pressure (e.g. concurrent template builds). uses the same logic that we were using for envd.

test test_sync_api_transport_cache_reuses_within_thread_and_isolates_across_threads fails on main, passes on branch.

[changeset-bot]

🦋 Changeset detected

Latest commit: 1ae0150

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

[linear-code]

ENG-4272

[cursor]

PR Summary

Medium Risk
Changes how sync HTTP clients are created in multi-threaded apps; intended as a correctness fix but could alter connection pooling behavior per thread.

Overview
The sync Python SDK cached TransportWithLogger in a class-level _instances map, so every thread reused the same httpx transport and HTTP/2 pool. get_transport now stores transports on TransportWithLogger._thread_local.instances, matching EnvdTransportWithLogger.

Tests clear thread-local caches instead of _instances, and add coverage that the API transport is reused in one thread but not shared across threads.

^{Reviewed by Cursor Bugbot for commit 1ae0150. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Package Artifacts

Built from 4bb8f50. Download artifacts from this workflow run.

JS SDK (e2b@2.28.1-threaded-api-calls-on-python-have-http2-collisions-eng-4272.0):

npm install ./e2b-2.28.1-threaded-api-calls-on-python-have-http2-collisions-eng-4272.0.tgz

CLI (@e2b/cli@2.10.5-threaded-api-calls-on-python-have-http2-collisions-eng-4272.0):

npm install ./e2b-cli-2.10.5-threaded-api-calls-on-python-have-http2-collisions-eng-4272.0.tgz

Python SDK (e2b==2.26.0+threaded-api-calls-on-python-have-http2-collisions-eng-4272):

pip install ./e2b-2.26.0+threaded.api.calls.on.python.have.http2.collisions.eng.4272-py3-none-any.whl

[claude]

LGTM — thread-local sync API transports mirror the existing envd pattern; the dead else branch flagged inline is a nit.

Extended reasoning...

Overview

The PR makes the sync Python SDK's API HTTP transport cache thread-local by switching TransportWithLogger._instances (a shared class-level dict) to TransportWithLogger._thread_local = threading.local(), mirroring the pattern already used for EnvdTransportWithLogger. The change touches three files: a changeset markdown, packages/python-sdk/e2b/api/client_sync/__init__.py (the transport caching logic), and packages/python-sdk/tests/test_api_client_transport.py (helper extraction and a new regression test).

Security risks

None meaningful. This narrows visibility of cached transports rather than widening it — moving from process-global to thread-local actually reduces the chance of accidental state sharing across thread boundaries. The fix addresses HTTP/2 connection-state collisions between threads (the Linear ticket ENG-4272 / PR title), which is itself a correctness/safety improvement.

Level of scrutiny

Low. This is a small, mechanical change copying an existing, already-reviewed pattern (envd transports) onto an adjacent class. The transformation is straightforward (read-from / write-to a threading.local() attribute via getattr with a fresh-dict default), and the test suite was extended with a direct regression test (test_sync_api_transport_cache_reuses_within_thread_and_isolates_across_threads) that the author confirms fails on main and passes on this branch.

Other factors

The only finding from the bug hunting pass is a nit: a defensive else branch in the new reset_sync_api_transports test helper is unreachable (the class attribute always exists) and references the now-removed _instances attribute. It has no behavioral impact — it's dead code in a test helper — and was likely intended as migration scaffolding. The inline comment already conveys the suggested cleanup, so the author can collapse it in a follow-up commit at their discretion.