fix(code-host): recover transient reads without retrying writes - #49
Merged
kitlangton merged 1 commit intoAug 28, 2026
Merged
Conversation
Copy link
Copy Markdown
Owner
Sorry, something went wrong.
Add bounded jittered backoff at explicit read boundaries, avoid caching failed GitLab project lookups, and reuse known titles while treating optional history enrichment as best effort. Drain child process output concurrently to prevent stderr pipe deadlocks. Cover retries, cancellation, mutation safety, enrichment boundaries, and real pipe backpressure.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Why
A single code-host read timeout currently aborts a stack operation, even when another attempt could succeed. GitLab also rereads already-known titles and retains failed source-project lookups in its cache, so optional work can fail a repair and later reads can repeat an old failure without contacting the host.
Investigation also reproduced a subprocess deadlock: a child filling stderr before closing stdout hangs while stack drains the pipes sequentially.
What Changes
Retry Boundaries
Retries wrap only explicit read call sites: lists, details, merge-state polling, and GitLab source-project lookup. Decoding remains outside the retry boundary. A retry stays inside its existing concurrency slot, and interruption cancels the delay rather than launching another request.
Scope
Related to #43, taking an automatic-recovery-first approach. This does not add stack.codeHostConcurrency, change the default of four jobs, or claim to reproduce the reporter's particular environment. Keep #43 open pending their timeout details and verification; persistent overload may still need a separate concurrency-policy decision.
Verification
GIT_CONFIG_GLOBAL=/dev/null GIT_CONFIG_SYSTEM=/dev/null bun run test bun run typecheck bun run format:check bun run lint bun run package:smoke bun run changeset:status --since origin/main bun src/cli.ts --help bun src/cli.ts sync --help git diff --check200 tests pass. New provider tests use injected CLI failures and TestClock to cover recovery, bounded attempts, cancellation, mutation exclusion, cache behavior, and concurrency. Stack regressions reproduce duplicate metadata reads and failed optional enrichment, while protecting mandatory-read and interruption boundaries. Latch-backed pipe tests and a real child writing 2 MiB to stderr reproduce the deadlock before the fix and pass afterward. No live-provider load tests were run.