Add cursor based pagination for get_task_instances endpoint by pierrejeambrun · Pull Request #64845 · apache/airflow

pierrejeambrun · 2026-04-07T14:54:16Z

related: #62027

Performance improvement. Add API support for the get_task_instances endpoint for 'cursor' based pagination.

At scale, just counting the number of rows is taking too long (60s because the joins prevent using the indexes and trigger parallel full scans, also just a bare count on the table is 2-18s, which is too long). Switching to cursor based pagination will allow to plug the UI on this and reduce this delay. Getting a page takes now <100ms while it was taking 60s on my local setup with 40M TIs.

This keeps backward compatibility, defaulting the list endpoint to 'offset' based pagination. While still allowing passing down a cursor to get a cursor based one.

Was generative AI tooling used to co-author this PR?

Yes (please specify the tool below)

Generated-by: [Cursor] following the guidelines

Read the Pull Request Guidelines for more information. Note: commit author/co-author name and email in commits become permanently public when merged.
For fundamental code changes, an Airflow Improvement Proposal (AIP) is needed.
When adding dependency, check compliance with the ASF 3rd Party License Policy.
For significant user-facing changes create newsfragment: {pr_number}.significant.rst, in airflow-core/newsfragments. You can add this file in a follow-up commit after the PR is created so you know the PR number.

- Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_task_instances endpoint. Offset pagination remains the default and is not deprecated globally. - Response uses a discriminated union: offset responses include total_entries, cursor responses include next_cursor and previous_cursor. - Refactor SortParam to lazily cache column resolution instead of mutating state in to_orm. - Move cursor helpers (encode/decode/apply) to dedicated common/db/cursors.py module. - Cleanly separate cursor vs offset code paths in the endpoint handler.

- Remove order_by from cursor token (now just a list of values) - Support empty string cursor for first page (no fake sentinel needed) - Drop order_by consistency check between cursor and query param

pierrejeambrun · 2026-04-07T14:54:35Z

WIP, needs some cleaning and a few iterations.

Encode type information directly into cursor tokens as {"type": ..., "value": ...} objects, removing the fragile column-based type guessing during deserialization. Narrow return types for endpoints that only return offset pagination (patch, clear, batch, mapped) so the OpenAPI spec and generated UI client reflect the correct types. Only get_task_instances retains the discriminated union response. Update UI components to use the narrowed types from the spec.

ashb

Love to see cursor based pagination though!

pierrejeambrun · 2026-04-16T14:15:33Z

I think you're right, this gave me the impression it was working but the actual items returned are not correct on walking backward.

Let me fix it and add a test for it

…4845) * Add cursor-based pagination to get_task_instances endpoint - Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_task_instances endpoint. Offset pagination remains the default and is not deprecated globally. - Response uses a discriminated union: offset responses include total_entries, cursor responses include next_cursor and previous_cursor. - Refactor SortParam to lazily cache column resolution instead of mutating state in to_orm. - Move cursor helpers (encode/decode/apply) to dedicated common/db/cursors.py module. - Cleanly separate cursor vs offset code paths in the endpoint handler. * Simplify cursor token and support first page without sentinel - Remove order_by from cursor token (now just a list of values) - Support empty string cursor for first page (no fake sentinel needed) - Drop order_by consistency check between cursor and query param * Small adjustments * Adjustments * Narrow endpoint return types and encode cursor value types Encode type information directly into cursor tokens as {"type": ..., "value": ...} objects, removing the fragile column-based type guessing during deserialization. Narrow return types for endpoints that only return offset pagination (patch, clear, batch, mapped) so the OpenAPI spec and generated UI client reflect the correct types. Only get_task_instances retains the discriminated union response. Update UI components to use the narrowed types from the spec. * Use msgpack for cursor tokens and nested keyset predicate Switch cursor encoding from typed JSON to msgpack for compactness. Replace flat OR-of-prefix-equalities with nested and/or keyset predicate for better composite index range scans. Always use ascending PK as the final tie-breaker for stable pagination. * Flatten TaskInstanceCollectionRes ponse to avoid oneOf codegen issues Replace the discriminated union (offset | cursor response types) with a single flat model using optional fields. OpenAPI oneOf + discriminator is not handled correctly by hey-api/openapi-ts (apache#1613, apache#3270): return types degrade to unknown in generated TypeScript code. * Fix UI * Fix CI * Fix cursor pagination boundary detection and error handling - Fetch limit+1 rows to accurately detect last page, returning next_cursor=null when no more results exist - Return previous_cursor=null on the first page (when no cursor was provided) - Use LimitFilter in apply_filters_to_select for the +1 limit instead of a manual .limit() call - Raise HTTP 400 on invalid UUID in cursor token instead of silently passing the invalid value - Update endpoint docs and add boundary-condition test * Fix backward cursor based pagination (cherry picked from commit e11c603)

pierrejeambrun · 2026-04-17T09:09:34Z

Manual backport #65405

…4845) * Add cursor-based pagination to get_task_instances endpoint - Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_task_instances endpoint. Offset pagination remains the default and is not deprecated globally. - Response uses a discriminated union: offset responses include total_entries, cursor responses include next_cursor and previous_cursor. - Refactor SortParam to lazily cache column resolution instead of mutating state in to_orm. - Move cursor helpers (encode/decode/apply) to dedicated common/db/cursors.py module. - Cleanly separate cursor vs offset code paths in the endpoint handler. * Simplify cursor token and support first page without sentinel - Remove order_by from cursor token (now just a list of values) - Support empty string cursor for first page (no fake sentinel needed) - Drop order_by consistency check between cursor and query param * Small adjustments * Adjustments * Narrow endpoint return types and encode cursor value types Encode type information directly into cursor tokens as {"type": ..., "value": ...} objects, removing the fragile column-based type guessing during deserialization. Narrow return types for endpoints that only return offset pagination (patch, clear, batch, mapped) so the OpenAPI spec and generated UI client reflect the correct types. Only get_task_instances retains the discriminated union response. Update UI components to use the narrowed types from the spec. * Use msgpack for cursor tokens and nested keyset predicate Switch cursor encoding from typed JSON to msgpack for compactness. Replace flat OR-of-prefix-equalities with nested and/or keyset predicate for better composite index range scans. Always use ascending PK as the final tie-breaker for stable pagination. * Flatten TaskInstanceCollectionRes ponse to avoid oneOf codegen issues Replace the discriminated union (offset | cursor response types) with a single flat model using optional fields. OpenAPI oneOf + discriminator is not handled correctly by hey-api/openapi-ts (apache#1613, apache#3270): return types degrade to unknown in generated TypeScript code. * Fix UI * Fix CI * Fix cursor pagination boundary detection and error handling - Fetch limit+1 rows to accurately detect last page, returning next_cursor=null when no more results exist - Return previous_cursor=null on the first page (when no cursor was provided) - Use LimitFilter in apply_filters_to_select for the +1 limit instead of a manual .limit() call - Raise HTTP 400 on invalid UUID in cursor token instead of silently passing the invalid value - Update endpoint docs and add boundary-condition test * Fix backward cursor based pagination (cherry picked from commit e11c603)

…65405) * Add cursor-based pagination to get_task_instances endpoint - Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_task_instances endpoint. Offset pagination remains the default and is not deprecated globally. - Response uses a discriminated union: offset responses include total_entries, cursor responses include next_cursor and previous_cursor. - Refactor SortParam to lazily cache column resolution instead of mutating state in to_orm. - Move cursor helpers (encode/decode/apply) to dedicated common/db/cursors.py module. - Cleanly separate cursor vs offset code paths in the endpoint handler. * Simplify cursor token and support first page without sentinel - Remove order_by from cursor token (now just a list of values) - Support empty string cursor for first page (no fake sentinel needed) - Drop order_by consistency check between cursor and query param * Small adjustments * Adjustments * Narrow endpoint return types and encode cursor value types Encode type information directly into cursor tokens as {"type": ..., "value": ...} objects, removing the fragile column-based type guessing during deserialization. Narrow return types for endpoints that only return offset pagination (patch, clear, batch, mapped) so the OpenAPI spec and generated UI client reflect the correct types. Only get_task_instances retains the discriminated union response. Update UI components to use the narrowed types from the spec. * Use msgpack for cursor tokens and nested keyset predicate Switch cursor encoding from typed JSON to msgpack for compactness. Replace flat OR-of-prefix-equalities with nested and/or keyset predicate for better composite index range scans. Always use ascending PK as the final tie-breaker for stable pagination. * Flatten TaskInstanceCollectionRes ponse to avoid oneOf codegen issues Replace the discriminated union (offset | cursor response types) with a single flat model using optional fields. OpenAPI oneOf + discriminator is not handled correctly by hey-api/openapi-ts (#1613, #3270): return types degrade to unknown in generated TypeScript code. * Fix UI * Fix CI * Fix cursor pagination boundary detection and error handling - Fetch limit+1 rows to accurately detect last page, returning next_cursor=null when no more results exist - Return previous_cursor=null on the first page (when no cursor was provided) - Use LimitFilter in apply_filters_to_select for the +1 limit instead of a manual .limit() call - Raise HTTP 400 on invalid UUID in cursor token instead of silently passing the invalid value - Update endpoint docs and add boundary-condition test * Fix backward cursor based pagination (cherry picked from commit e11c603)

Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_dag_runs endpoint. Offset pagination remains the default and is not deprecated. Mirrors the pattern introduced for get_task_instances in apache#64845. - Response schema adds next_cursor / previous_cursor and makes total_entries optional (populated only for offset pagination). - Fetch limit+1 rows to detect last page and return next_cursor=null when no more results exist. - Rewire the DagRuns listing page to use cursor pagination via the DataTable's cursorPagination prop, which eliminates the COUNT(*) query for large datasets.

…65405) * Add cursor-based pagination to get_task_instances endpoint - Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_task_instances endpoint. Offset pagination remains the default and is not deprecated globally. - Response uses a discriminated union: offset responses include total_entries, cursor responses include next_cursor and previous_cursor. - Refactor SortParam to lazily cache column resolution instead of mutating state in to_orm. - Move cursor helpers (encode/decode/apply) to dedicated common/db/cursors.py module. - Cleanly separate cursor vs offset code paths in the endpoint handler. * Simplify cursor token and support first page without sentinel - Remove order_by from cursor token (now just a list of values) - Support empty string cursor for first page (no fake sentinel needed) - Drop order_by consistency check between cursor and query param * Small adjustments * Adjustments * Narrow endpoint return types and encode cursor value types Encode type information directly into cursor tokens as {"type": ..., "value": ...} objects, removing the fragile column-based type guessing during deserialization. Narrow return types for endpoints that only return offset pagination (patch, clear, batch, mapped) so the OpenAPI spec and generated UI client reflect the correct types. Only get_task_instances retains the discriminated union response. Update UI components to use the narrowed types from the spec. * Use msgpack for cursor tokens and nested keyset predicate Switch cursor encoding from typed JSON to msgpack for compactness. Replace flat OR-of-prefix-equalities with nested and/or keyset predicate for better composite index range scans. Always use ascending PK as the final tie-breaker for stable pagination. * Flatten TaskInstanceCollectionRes ponse to avoid oneOf codegen issues Replace the discriminated union (offset | cursor response types) with a single flat model using optional fields. OpenAPI oneOf + discriminator is not handled correctly by hey-api/openapi-ts (#1613, #3270): return types degrade to unknown in generated TypeScript code. * Fix UI * Fix CI * Fix cursor pagination boundary detection and error handling - Fetch limit+1 rows to accurately detect last page, returning next_cursor=null when no more results exist - Return previous_cursor=null on the first page (when no cursor was provided) - Use LimitFilter in apply_filters_to_select for the +1 limit instead of a manual .limit() call - Raise HTTP 400 on invalid UUID in cursor token instead of silently passing the invalid value - Update endpoint docs and add boundary-condition test * Fix backward cursor based pagination (cherry picked from commit e11c603)

* Add cursor based pagination for get_dag_runs endpoint Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_dag_runs endpoint. Offset pagination remains the default and is not deprecated. Mirrors the pattern introduced for get_task_instances in #64845. - Response schema adds next_cursor / previous_cursor and makes total_entries optional (populated only for offset pagination). - Fetch limit+1 rows to detect last page and return next_cursor=null when no more results exist. - Rewire the DagRuns listing page to use cursor pagination via the DataTable's cursorPagination prop, which eliminates the COUNT(*) query for large datasets. * Small Adjustments * Small Adjustments

…5604) * Add cursor based pagination for get_dag_runs endpoint Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_dag_runs endpoint. Offset pagination remains the default and is not deprecated. Mirrors the pattern introduced for get_task_instances in #64845. - Response schema adds next_cursor / previous_cursor and makes total_entries optional (populated only for offset pagination). - Fetch limit+1 rows to detect last page and return next_cursor=null when no more results exist. - Rewire the DagRuns listing page to use cursor pagination via the DataTable's cursorPagination prop, which eliminates the COUNT(*) query for large datasets. * Small Adjustments * Small Adjustments (cherry picked from commit c57abb9) Co-authored-by: Pierre Jeambrun <pierrejbrun@gmail.com>

…5604) (#65746) * Add cursor based pagination for get_dag_runs endpoint Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_dag_runs endpoint. Offset pagination remains the default and is not deprecated. Mirrors the pattern introduced for get_task_instances in #64845. - Response schema adds next_cursor / previous_cursor and makes total_entries optional (populated only for offset pagination). - Fetch limit+1 rows to detect last page and return next_cursor=null when no more results exist. - Rewire the DagRuns listing page to use cursor pagination via the DataTable's cursorPagination prop, which eliminates the COUNT(*) query for large datasets. * Small Adjustments * Small Adjustments (cherry picked from commit c57abb9) Co-authored-by: Pierre Jeambrun <pierrejbrun@gmail.com>

…65405) * Add cursor-based pagination to get_task_instances endpoint - Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_task_instances endpoint. Offset pagination remains the default and is not deprecated globally. - Response uses a discriminated union: offset responses include total_entries, cursor responses include next_cursor and previous_cursor. - Refactor SortParam to lazily cache column resolution instead of mutating state in to_orm. - Move cursor helpers (encode/decode/apply) to dedicated common/db/cursors.py module. - Cleanly separate cursor vs offset code paths in the endpoint handler. * Simplify cursor token and support first page without sentinel - Remove order_by from cursor token (now just a list of values) - Support empty string cursor for first page (no fake sentinel needed) - Drop order_by consistency check between cursor and query param * Small adjustments * Adjustments * Narrow endpoint return types and encode cursor value types Encode type information directly into cursor tokens as {"type": ..., "value": ...} objects, removing the fragile column-based type guessing during deserialization. Narrow return types for endpoints that only return offset pagination (patch, clear, batch, mapped) so the OpenAPI spec and generated UI client reflect the correct types. Only get_task_instances retains the discriminated union response. Update UI components to use the narrowed types from the spec. * Use msgpack for cursor tokens and nested keyset predicate Switch cursor encoding from typed JSON to msgpack for compactness. Replace flat OR-of-prefix-equalities with nested and/or keyset predicate for better composite index range scans. Always use ascending PK as the final tie-breaker for stable pagination. * Flatten TaskInstanceCollectionRes ponse to avoid oneOf codegen issues Replace the discriminated union (offset | cursor response types) with a single flat model using optional fields. OpenAPI oneOf + discriminator is not handled correctly by hey-api/openapi-ts (#1613, #3270): return types degrade to unknown in generated TypeScript code. * Fix UI * Fix CI * Fix cursor pagination boundary detection and error handling - Fetch limit+1 rows to accurately detect last page, returning next_cursor=null when no more results exist - Return previous_cursor=null on the first page (when no cursor was provided) - Use LimitFilter in apply_filters_to_select for the +1 limit instead of a manual .limit() call - Raise HTTP 400 on invalid UUID in cursor token instead of silently passing the invalid value - Update endpoint docs and add boundary-condition test * Fix backward cursor based pagination (cherry picked from commit e11c603)

…5604) (#65746) * Add cursor based pagination for get_dag_runs endpoint Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_dag_runs endpoint. Offset pagination remains the default and is not deprecated. Mirrors the pattern introduced for get_task_instances in #64845. - Response schema adds next_cursor / previous_cursor and makes total_entries optional (populated only for offset pagination). - Fetch limit+1 rows to detect last page and return next_cursor=null when no more results exist. - Rewire the DagRuns listing page to use cursor pagination via the DataTable's cursorPagination prop, which eliminates the COUNT(*) query for large datasets. * Small Adjustments * Small Adjustments (cherry picked from commit c57abb9) Co-authored-by: Pierre Jeambrun <pierrejbrun@gmail.com>

…65405) * Add cursor-based pagination to get_task_instances endpoint - Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_task_instances endpoint. Offset pagination remains the default and is not deprecated globally. - Response uses a discriminated union: offset responses include total_entries, cursor responses include next_cursor and previous_cursor. - Refactor SortParam to lazily cache column resolution instead of mutating state in to_orm. - Move cursor helpers (encode/decode/apply) to dedicated common/db/cursors.py module. - Cleanly separate cursor vs offset code paths in the endpoint handler. * Simplify cursor token and support first page without sentinel - Remove order_by from cursor token (now just a list of values) - Support empty string cursor for first page (no fake sentinel needed) - Drop order_by consistency check between cursor and query param * Small adjustments * Adjustments * Narrow endpoint return types and encode cursor value types Encode type information directly into cursor tokens as {"type": ..., "value": ...} objects, removing the fragile column-based type guessing during deserialization. Narrow return types for endpoints that only return offset pagination (patch, clear, batch, mapped) so the OpenAPI spec and generated UI client reflect the correct types. Only get_task_instances retains the discriminated union response. Update UI components to use the narrowed types from the spec. * Use msgpack for cursor tokens and nested keyset predicate Switch cursor encoding from typed JSON to msgpack for compactness. Replace flat OR-of-prefix-equalities with nested and/or keyset predicate for better composite index range scans. Always use ascending PK as the final tie-breaker for stable pagination. * Flatten TaskInstanceCollectionRes ponse to avoid oneOf codegen issues Replace the discriminated union (offset | cursor response types) with a single flat model using optional fields. OpenAPI oneOf + discriminator is not handled correctly by hey-api/openapi-ts (#1613, #3270): return types degrade to unknown in generated TypeScript code. * Fix UI * Fix CI * Fix cursor pagination boundary detection and error handling - Fetch limit+1 rows to accurately detect last page, returning next_cursor=null when no more results exist - Return previous_cursor=null on the first page (when no cursor was provided) - Use LimitFilter in apply_filters_to_select for the +1 limit instead of a manual .limit() call - Raise HTTP 400 on invalid UUID in cursor token instead of silently passing the invalid value - Update endpoint docs and add boundary-condition test * Fix backward cursor based pagination (cherry picked from commit e11c603)

…5604) (#65746) * Add cursor based pagination for get_dag_runs endpoint Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_dag_runs endpoint. Offset pagination remains the default and is not deprecated. Mirrors the pattern introduced for get_task_instances in #64845. - Response schema adds next_cursor / previous_cursor and makes total_entries optional (populated only for offset pagination). - Fetch limit+1 rows to detect last page and return next_cursor=null when no more results exist. - Rewire the DagRuns listing page to use cursor pagination via the DataTable's cursorPagination prop, which eliminates the COUNT(*) query for large datasets. * Small Adjustments * Small Adjustments (cherry picked from commit c57abb9) Co-authored-by: Pierre Jeambrun <pierrejbrun@gmail.com>

pierrejeambrun · 2026-04-07T14:54:16Z

related: #62027

Performance improvement. Add API support for the get_task_instances endpoint for 'cursor' based pagination.

At scale, just counting the number of rows is taking too long (60s because the joins prevent using the indexes and trigger parallel full scans, also just a bare count on the table is 2-18s, which is too long). Switching to cursor based pagination will allow to plug the UI on this and reduce this delay. Getting a page takes now <100ms while it was taking 60s on my local setup with 40M TIs.

This keeps backward compatibility, defaulting the list endpoint to 'offset' based pagination. While still allowing passing down a cursor to get a cursor based one.

Was generative AI tooling used to co-author this PR?

Yes (please specify the tool below)

Generated-by: [Cursor] following the guidelines

Read the Pull Request Guidelines for more information. Note: commit author/co-author name and email in commits become permanently public when merged.
For fundamental code changes, an Airflow Improvement Proposal (AIP) is needed.
When adding dependency, check compliance with the ASF 3rd Party License Policy.
For significant user-facing changes create newsfragment: {pr_number}.significant.rst, in airflow-core/newsfragments. You can add this file in a follow-up commit after the PR is created so you know the PR number.

- Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_task_instances endpoint. Offset pagination remains the default and is not deprecated globally. - Response uses a discriminated union: offset responses include total_entries, cursor responses include next_cursor and previous_cursor. - Refactor SortParam to lazily cache column resolution instead of mutating state in to_orm. - Move cursor helpers (encode/decode/apply) to dedicated common/db/cursors.py module. - Cleanly separate cursor vs offset code paths in the endpoint handler.

- Remove order_by from cursor token (now just a list of values) - Support empty string cursor for first page (no fake sentinel needed) - Drop order_by consistency check between cursor and query param

pierrejeambrun · 2026-04-07T14:54:35Z

WIP, needs some cleaning and a few iterations.

Encode type information directly into cursor tokens as {"type": ..., "value": ...} objects, removing the fragile column-based type guessing during deserialization. Narrow return types for endpoints that only return offset pagination (patch, clear, batch, mapped) so the OpenAPI spec and generated UI client reflect the correct types. Only get_task_instances retains the discriminated union response. Update UI components to use the narrowed types from the spec.

ashb

Love to see cursor based pagination though!

pierrejeambrun · 2026-04-16T14:15:33Z

I think you're right, this gave me the impression it was working but the actual items returned are not correct on walking backward.

Let me fix it and add a test for it

…4845) * Add cursor-based pagination to get_task_instances endpoint - Add cursor-based (keyset) pagination as an alternative to offset-based pagination on the get_task_instances endpoint. Offset pagination remains the default and is not deprecated globally. - Response uses a discriminated union: offset responses include total_entries, cursor responses include next_cursor and previous_cursor. - Refactor SortParam to lazily cache column resolution instead of mutating state in to_orm. - Move cursor helpers (encode/decode/apply) to dedicated common/db/cursors.py module. - Cleanly separate cursor vs offset code paths in the endpoint handler. * Simplify cursor token and support first page without sentinel - Remove order_by from cursor token (now just a list of values) - Support empty string cursor for first page (no fake sentinel needed) - Drop order_by consistency check between cursor and query param * Small adjustments * Adjustments * Narrow endpoint return types and encode cursor value types Encode type information directly into cursor tokens as {"type": ..., "value": ...} objects, removing the fragile column-based type guessing during deserialization. Narrow return types for endpoints that only return offset pagination (patch, clear, batch, mapped) so the OpenAPI spec and generated UI client reflect the correct types. Only get_task_instances retains the discriminated union response. Update UI components to use the narrowed types from the spec. * Use msgpack for cursor tokens and nested keyset predicate Switch cursor encoding from typed JSON to msgpack for compactness. Replace flat OR-of-prefix-equalities with nested and/or keyset predicate for better composite index range scans. Always use ascending PK as the final tie-breaker for stable pagination. * Flatten TaskInstanceCollectionRes ponse to avoid oneOf codegen issues Replace the discriminated union (offset | cursor response types) with a single flat model using optional fields. OpenAPI oneOf + discriminator is not handled correctly by hey-api/openapi-ts (apache#1613, apache#3270): return types degrade to unknown in generated TypeScript code. * Fix UI * Fix CI * Fix cursor pagination boundary detection and error handling - Fetch limit+1 rows to accurately detect last page, returning next_cursor=null when no more results exist - Return previous_cursor=null on the first page (when no cursor was provided) - Use LimitFilter in apply_filters_to_select for the +1 limit instead of a manual .limit() call - Raise HTTP 400 on invalid UUID in cursor token instead of silently passing the invalid value - Update endpoint docs and add boundary-condition test * Fix backward cursor based pagination (cherry picked from commit e11c603)

pierrejeambrun · 2026-04-17T09:09:34Z