Skip to documentation content
Dataset APIDocs

Limits and performance

Use the limits actually guaranteed by the v1 contract.

In this guide
  1. 1Bound queries
  2. 2Respect Retry-After
  3. 3Stream downloads
In this guide
  1. 1Bound queries
  2. 2Respect Retry-After
  3. 3Stream downloads

Declared v1 limits

InputContract
Query limit1–1,000 rows; default 100
Projected fieldsAt most 100 field names
Download formatjsonl, csv, or parquet
API major headerX-API-Version defaults to 1

Let the response control retry timing

A 429 response may include Retry-After in seconds. Wait at least that long and add jitter. The public contract does not promise fixed workspace request quotas, so do not hard-code an assumed requests-per-second value.

Bound concurrency at the client

Use a small worker pool and queue excess work. Separate build polling from query traffic so a degraded build does not consume all client capacity. Apply an overall deadline in addition to individual request timeouts.

Choose the correct transport

NeedUse
Interactive previewBounded query with selected fields
Complete release importStreaming download
Build statusGET one build with bounded polling
Catalog refreshConditional scheduled GET in your application
Was this page helpful?