The documentation specifies 800 reads/minute per client. Our exporter gates every SDK request and retry at 720 reads/minute per process, but we’re seeing 429s well below that.
In separate read-only tests, 429s occurred with 451 and 576 measured requests in the preceding rolling minute. Overnight runs also saw repeated 429s after waiting at least 60 seconds with only one new request. There was no known competing activity during testing, although we cannot audit all provider-side traffic.
Some quota errors from different customer installations reference the same Google consumer project. Could you clarify:
- What defines a “client”—company, API key, service account, consumer project or IP address? Can separate customer credentials share a quota?
- Are there additional burst limits, request weights or lower effective quotas that could explain these below-800 responses?
We can provide UTC timestamps and redacted logs for investigation.