[FEATURE]:Add retry logic for subagents on TPM/RPM rate limits #59533

Activity

1. opencode-agent commented on Oct 27, 2026

   

   opencode-agentBot

   on Oct 27, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the request! To restate it: you'd like subagents to handle TPM/RPM rate limits gracefully, by waiting for the suggested delay and retrying a configurable number of times, instead of failing and losing their progress.

   Some context from the current v2 code: a subagent's session goes through the same retry logic as the primary agent. It retries rate-limit errors up to 10 times, backing off from 2s to 10s between attempts and waiting at least as long as the provider's retry hint. The "You exceeded your current quota… Please retry in …" message comes from Gemini's free tier. Once its daily quota is used up, every retry until the reset gets another 429, so the subagent can run out of retries and fail. If you see the primary agent keep retrying on the very same error while a subagent fails right away, please reply with your opencode --version, provider and model. That would make this a bug rather than a feature request.

   At the moment this can't be done fully as a plugin. The session.retry hook can change the delay of the built-in retries, but it can't add attempts beyond them. Maintainers will consider both the option and that plugin hook gap.

   Related issues:

   • [FEATURE]:Automatically pause and resume tasks when rate limits have a known reset time #43126: pausing a rate-limited session and resuming it automatically when the limit resets
   • Concurrent sessions retry a failing provider independently — no shared backoff or circuit breaker (2.0) #55075: sharing retry state per provider across sessions and subagents
   • [Feature Request / Bug] Handle Rate Limit (Quota Exceeded 429) automatically with retry delay #49039: the same Gemini free-tier error, reported earlier, with an explanation of the per-minute and per-day quotas
2. monbooo090 commented on Oct 27, 2026

   

   monbooo090

   on Oct 27, 2026

   Last edited by monbooo090

   Author

   More actions

   Thanks for the detailed explanation!

   It does seem to be a bug rather than the retries simply being exhausted. Here
   are my environment details and observations:

   • Version: opencode/latest/2.2.0/desktop
   • Provider: Google
   • Model: Gemini 3.5 flash lite

   Observed behavior:

   • When a subagent hits the TPM/RPM limit, it does not wait for the
     cooldown/retry delay at all—it immediately terminates the session with an
     error on the very first hit (rather than backing off and retrying up to 10
     times).
   • Subagent sessions that do not hit TPM/RPM limits complete successfully
     without any issues.
   • Under the exact same rate-limit error, the primary agent behaves as
     expected: it waits for the configured retry interval and retries properly.

   It looks like the retry logic isn't being triggered for subagents in this setup.
   Please let me know if you need any additional logs or details to help reproduce
   this.

   Note that in Version 1, even when the sub-agent Gemini encounters a limit-exceeded restriction, it is able to retry without any issues after the countdown. Also, I am not using any plugins at all.