Rate limits
What the server enforces on event ingestion, errors, spans and the waitlist, what a 429 looks like, and the size caps on a batch.
| Surface | Limit | Over it |
|---|---|---|
POST /api/v1/events/log, per app | 6,000 events per minute (counts events, not requests) | The whole batch is refused: 429 rate_limited, with a Retry-After header and retry_after in the body, both in seconds. Nothing from it is stored; send it again after that wait. |
POST /api/v1/events/log, per device | 600 events per minute per device | The batch is accepted, the excess at its end is dropped, and the 200 reports how many in throttled. There is no retry signal: a device sending this fast is usually stuck in a loop. |
POST /api/v1/errors | 600 error events per minute per app, 60 per device | The whole batch is refused: 429 rate_limited, with Retry-After, retry_after and scope (app or device). Below those limits, each issue stores at most 20 events a minute; the rest are counted, not stored. |
POST /v1/traces | 30,000 spans per minute per app | The whole export is refused: 429 with Retry-After, in OTLP's error shape. |
POST /api/v1/waitlist | 10 requests per minute per IP | 429 rate_limited. |
The per-app ingestion limit is shared across every Fictura server, so spreading requests does not raise it. If your app needs more than 6,000 events a minute, ask and it can be raised for that app.
{
"ok": false,
"error": "rate_limited",
"retry_after": 12
}The SDK handles both: it waits out Retry-After and keeps the batch on the device until it is accepted. A server sender should do the same.
SDK config, entitlements, the Links API and the Data APIs carry no request-rate limit today. Handle a 429 gracefully everywhere anyway.
Size caps
- JSON bodies: 1 MB across the API. Key-authenticated routes that allow more: session replay 12 MB, traces 5 MB, source maps 60 MB, native symbols 80 MB.
- Event batches: at most 500 events per request (
400 too_many_eventsover that). The SDK flushes at 25 events by default; server senders should batch similarly rather than one request per event. - One event: a name of 1 to 200 characters, at most 255 property keys, and properties up to 64 KB serialized. An event over a cap is refused on its own and listed in the response's
failed; the rest of the batch is stored. - Error ingestion: at most 20 events per request; stacks are accepted up to 64,000 characters and stored truncated to 32,000.
- Bulk link creation: 500 links per request, executed in one transaction.
- Link listing:
limitcapped at 500. - Events and Persons API pages:
limitfrom 1 to 1,000, default 100. - Bulk export: at most 366 days per job.
- Traces: at most 5,000 spans per request.
- Targeting attributes: 100 keys per batch, string values ≤ 1,024 characters. A key over a cap is dropped with a warning; the batch is still stored.
Truncation is loud, not silent
Where a read endpoint caps a result (CSV export, activity feeds), the response says so (X-Fictura-Row-Count and X-Fictura-Truncated on exports), so a capped result never masquerades as a complete one.