Retry OpenCodeSessionFactory.create() with backoff - #1009
Open
sergiopaniego wants to merge 2 commits into
Open
Conversation
Session creation spins up a sandbox, installs opencode, and starts the proxy + agent, the flakiest step in a rollout; a single transient failure used to drop the rollout as unscorable, shrinking the effective group size. create() now retries create_attempts times with exponential backoff. The per-attempt logic moved to _create_once, which already tears its own sandbox down on failure, so a retry never leaks a sandbox.
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds retry-with-backoff to
OpenCodeSessionFactory.create().Why
Session creation spins up a sandbox, installs opencode, and starts the proxy + agent. It is the flakiest step in a rollout (sandbox-API blips, cold install, proxy start). Today a single transient failure drops the whole rollout as unscorable, which shrinks the effective group size and silently weakens the training signal. This is one of the
TODO(@openenv)gaps flagged in TRL's loop-owning harness work (huggingface/trl#6420).Change
create()now retries up tocreate_attempts(default 3) with exponential backoff (create_backoff_s * 2**i, default base 2s), matching the existing_exec_with_retrystyle._create_once, which already tears its own sandbox down on any post-provision failure, so a retry never leaks a sandbox.Tests
The two existing teardown tests now pin
create_attempts=1(they test a single attempt's cleanup). Two new tests cover the wrapper: retries-then-succeeds, and raises-after-exhausting-attempts.pi_env'sPiSessionFactory.create()mirrors this and gets the same change on its own PR (#999).AI-assisted.
Note
Medium Risk
Changes rollout/session startup behavior and can add delay on repeated failures, but scope is limited to factory create retries with existing per-attempt cleanup.
Overview
OpenCodeSessionFactory.create()now retries flaky session provisioning instead of failing a rollout on the first transient error.The factory accepts
create_attempts(default 3) andcreate_backoff_s(default 2s).create()loops with exponential backoff (create_backoff_s * 2**i), logs warnings between tries, and re-raises the last error when attempts are exhausted. The previous single-shot logic lives unchanged in_create_once, which still kills the sandbox on post-provision failure so retries do not leak sandboxes.Lifecycle tests pin
create_attempts=1so teardown behavior stays single-attempt; new tests cover succeed-on-third-try and fail-after-all-attempts.Reviewed by Cursor Bugbot for commit 0cd39bf. Bugbot is set up for automated code reviews on this repo. Configure here.