How-to guideApplies to: Unified engine deployments
List endpoints and query-result endpoints have different paging units. Read each operation’s query parameters and response cursor before fetching another page.
Page resource lists
Many lists use limit and continuation_token. Some notebook and assistant APIs use after, before, or sequence cursors instead. Pass the returned cursor to the same endpoint with the same filters.
A server maximum can reject an oversized page. Nested response envelopes matter: a token might be inside workloads, relations, or another page object rather than at the root.
Read SQL result pages
Completed query results use base64 Arrow IPC over JSON:
Result limit counts stored Arrow batches, not rows. The HTTP result endpoint defaults to one batch. Follow continuation_token to read later batches. Paging reads an existing result and does not execute its SQL again.
Decode Arrow batches with a compatible client. Preserve duplicate column names as separate columns. Display truncation in a client and execution truncation are different limits.
MCP can offer a JSON-row adapter. Its result_format, max_rows, and row_offset arguments are not HTTP query parameters.
Limits and retention
Request a smaller page when needed. The server has configured request, response, and page limits.
Results can expire while operation or run history remains. A write-only task can succeed without retained rows. Verify a durable destination through a query and commit evidence.Last modified on November 4, 2026
