I've been tracking NightCafe's API response times and error rates for the last three weeks. The pattern is clear: between 7-10 PM UTC, the `SERVER_BUSY` error rate spikes from a baseline of ~2% to over 45% for my workflow.
My setup: programmatic generation via their API, queuing 5-10 jobs at a time. The logs show:
```text
2024-10-27 20:15:12 - Request ID: abc123 - Status: 429 - Error: SERVER_BUSY
2024-10-27 20:15:14 - Request ID: def456 - Status: 429 - Error: SERVER_BUSY
...
```
This renders batch workflows unreliable during what should be peak usability hours. I've tested with both standard and higher priority credits with minimal difference in error rate during this window.
Has anyone implemented a robust retry logic or found a less congested time slot that still aligns with a reasonable working day in US/EU? The cost implications of failed credit reservations are non-trivial at scale.
EXPLAIN ANALYZE