design-api-pagination

API pagination using traditional OFFSET logic breaks during concurrent writes because newly inserted rows shift the offset window, causing duplicate or missing records. Keyset-based cursor navigation solves this by anchoring pagination to stable unique tie-break keys rather than positional offsets.

Implement keyset pagination by using opaque base64url-encoded cursors derived from unique tie-break keys, applying limit+1 probing for efficient has_more detection, and filtering rows greater than the cursor value to achieve O(1) query complexity for REST and GraphQL endpoints.

The best way to handle deep-page SQL query performance degradation is replacing OFFSET with keyset pagination, which maintains O(1) query complexity regardless of page depth by using indexed cursor columns instead of scanning and skipping offset rows.

Build GraphQL Relay-compliant cursor connections by implementing opaque base64url-encoded cursors mapped to unique tie-break keys, using limit+1 probing to detect has_more, and returning standardized connection edges and page info for stable infinite scroll feeds.

You should not use keyset cursor pagination when your API requires arbitrary page jumping or random access to specific page numbers, as keyset cursors only support sequential forward and backward navigation anchored to specific row values rather than computed offsets.