Function runResumeLegWithRetry

apps/sim/lib/copilot/request/lifecycle/run.ts:354–392

(
  url: string,
  body: Record<string, unknown>,
  leg: StreamingContext,
  execContext: ExecutionContext,
  options: CopilotLifecycleOptions
)

// failure cancelling this leg, see driveSubagentChains) is non-retryable and
// propagates immediately.
async function runResumeLegWithRetry(
  url: string,
  body: Record<string, unknown>,
  leg: StreamingContext,
  execContext: ExecutionContext,
  options: CopilotLifecycleOptions
): Promise<void> {
  let attempt = 0
  for (;;) {
    const errorsBeforeAttempt = leg.errors.length
    const willRetryOnStreamError = attempt < MAX_RESUME_ATTEMPTS - 1
    const legBody = willRetryOnStreamError ? { ...body, willRetryOnStreamError: true } : body
    try {
      await runStreamLoop(
        url,
        { method: 'POST', headers: mothershipRequestHeaders(), body: JSON.stringify(legBody) },
        leg,
        execContext,
        options
      )
      return
    } catch (error) {
      if (isRetryableStreamError(error) && attempt < MAX_RESUME_ATTEMPTS - 1) {
        leg.errors.length = errorsBeforeAttempt
        attempt++
        const backoff = RESUME_BACKOFF_MS[attempt - 1] ?? 1000
        logger.warn('Child resume leg failed, retrying', {
          attempt: attempt + 1,
          maxAttempts: MAX_RESUME_ATTEMPTS,
          backoffMs: backoff,
          error: toError(error).message,
        })
        await sleepWithAbort(backoff, options.abortSignal)
        continue
      }
      throw error
    }
  }
}

// driveOneChildChain resumes a single subagent's checkpoint chain to its end:
// resume -> (re-pause -> resume)* -> fold into join. Returns the orchestrator's