Retry on errors, with exponential interval! #41

Activity

1. OpenCodeEngineer commented on Mar 26, 2026

   

   OpenCodeEngineer

   on Mar 26, 2026

   More actions

   Investigating now. The current retry loop already backs off on retryable stream failures, but it only gives a minimal recovery path when stream_max_retries = 0. I’m patching codex-core so transient dropped streaming requests keep replaying with exponential backoff using the provider retry budget, and I’m adding regression coverage for repeated Azure-style disconnects before opening the PR.
2. dzianisv commented on Mar 26, 2026

   

   dzianisv

   on Mar 26, 2026

   OwnerAuthor

   More actions

   Implemented on issue-41-retry-transport-disconnects and opened PR #42.

   What changed:

   • reuse the provider request retry budget when stream_max_retries = 0 and the failure is a transient transport disconnect
   • keep zero-retry behavior for non-transport stream errors and keep honoring any nonzero explicit stream retry budget
   • add regressions for repeated incomplete SSE streams in both the in-process core path and the real codex exec CLI path

   Validated locally with:

   • cargo test -p codex-core effective_stream_retry_budget --color never
   • cargo test -p codex-core --test all suite::stream_no_completed::retries_replayed_stream_request_when_stream_retries_are_disabled --color never -- --exact
   • cargo build -p codex-cli --bin codex --color never
   • cargo test -p codex-core --test all suite::cli_stream::responses_mode_stream_cli_retries_replayed_stream_request_when_stream_retries_disabled --color never -- --exact
3. dzianisv

   closed this as completedin #42on Mar 28, 2026
4. dzianisv

   added a commit that references this issue on Mar 28, 2026

   fix(core): retry repeated transient stream disconnects (#41) (#42)

   ...

   5afed61