Fix performance regression when setting env vars on first execution

[mishushakov]

Changes

• Setting env vars now appends a generated code snippet to the user code with no extra request to Jupyter.
• Resetting env vars happens in a separate request in background after response so that the return of the last cell stays unmodified.
• Because resetting happens in background but the new execution will wait until the previous cleanup task has been completed (to ensure correctness), fast subsequent requests might get delayed due that reason

Remarks

• I suppose, we can cut even more ms when starting new kernels by making change_current_directory append to the user code on first request like the env vars.

Benchmark

Tested locally in debug mode.

Sample code used:

from e2b_code_interpreter import Sandbox
import time

sbx = Sandbox(debug=True)

# Start time measure
start_time = time.time()
sbx.run_code("os.environ['hello']", envs={"hello": "world"})
end_time = time.time()
print(f"Python kernel time taken: {(end_time - start_time) * 1000:.2f} milliseconds")

Before

Python kernel time taken: 116.38 milliseconds

After

Python kernel time taken: 18.64 milliseconds

[linear]

E2B-2850 Performance regression when setting env vars on first request

[mishushakov]

@cursor review

[mishushakov]

@cursor review

[mishushakov]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no bugs!

Comment @cursor review or bugbot run to trigger another review on this PR