## TL;DR
The audit made too many Fivetran API calls too fast and got throttled mid-run, failing the whole audit. Budget the calls: fetch the connector list once, page deliberately, back off on 429, and save partial results as you go. An audit that dies at 80 percent with nothing saved is worse than a slower audit that finishes.

```text
fivetran API rate limit hit mid-sync-audit, audit failed
```

## Use this when
- An agent's Fivetran audit fails on API 429s partway through.
- The audit worked on small accounts but fails on large ones.
- No partial results were saved.

## Not for this skill when
- The API returns 401. That is auth, not rate limiting.
- The audit logic is wrong. Fix the logic.

## Steps
1. Count the API calls the audit makes and identify the burstiest phase. Note the budget. Verify: the call pattern is measured
2. Add backoff with jitter on 429 and a per-minute call cap. Note the throttle. Verify: the audit paces itself
3. Save partial results incrementally so a throttled run keeps its progress. Note the checkpointing. Verify: progress survives throttling
4. Reduce calls: fetch lists once, cache connector details, skip unchanged connectors. Note the leaner audit. Verify: the call count drops
5. Re-run the audit and confirm it completes under the limit. Note the green run. Verify: the audit finishes

## Variant phrasings
### fivetran api 429 backoff
The technique phrasing.
### fivetran api rate limit audit
The use-case search.

Compatibility: Fivetran REST API rate limits. Budgeting and backoff are client-side patterns.

## Why it happens
Audit tooling is written against small accounts and deployed against large ones: the call count scales with connectors, and the rate limit does not scale with anything. A 429 mid-audit is the API defending itself, and an audit without backoff or checkpointing turns one throttled response into a total failure. The fix is treating the API budget as a first-class constraint: count calls, pace them, and never let a throttle erase completed work.

## Edge cases / pitfalls
- Retrying a 429 immediately without backoff extends the throttle. Always back off; the Retry-After header tells you how long.
- Different Fivetran API endpoints share the limit pool. Budget across the whole audit, not per endpoint.
- An audit that runs while syncs are active competes with operational calls. Schedule audits off-peak.
- Cached connector details go stale. Cache with a TTL that fits how fast the account changes.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_WeQcEUU_X7OafW_Z-wIuLw