The manually-written empty report-alpha.api.md didn't match what API
Extractor generates for an empty alpha entry point. Use the correct
generated output to pass the CI api-reports check.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
- Update copyright year from 2024 to 2026 across all new source files
- Clear alpha entry point (no alpha exports needed for a net-new package)
and regenerate report-alpha.api.md accordingly
- Delete CHANGELOG.md (auto-generated by the release process)
- Change changeset bump from patch to minor so the first release is 0.1.0,
and remove the redundant "New package:" title line
- Reset package version to 0.0.0 (release process sets the real version)
- Add incremental ingestion section to docs/integrations/azure/org.md
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
A throwing custom groupTransformer on a child group member would fail
the entire groups-phase burst. Apply the same try/catch + debug/warn
logging pattern already used for user member transformation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
- Guard getUserPhotoGated with user.id presence to prevent requesting
users/undefined/photo when Graph omits the id field
- Log a warning when groupIncludeSubGroups is configured, matching the
existing warning for unsupported userGroupMember* options
- Skip spec.children population when groupFilter/groupSearch is active,
since child groups may not pass the filter and would create dangling
references in the catalog
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
Update all references to "up to 999 items" to accurately reflect that
users are fetched in pages of 999 while groups use a smaller page of
100. Also document groupIncludeSubGroups as unsupported in the JSDoc
@remarks, README, and comparison table.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
These modules are not needed by all Backstage users, so registering them
unconditionally in the example backend adds unnecessary routes, services,
and runtime dependencies. The backend.add(...) lines are documented in
the plugin README instead.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
Without $select, Microsoft Graph returns only id and @odata.type for
group members, which causes defaultUserTransformer/defaultGroupTransformer
to return undefined and silently drop membership refs. Requesting the
minimum fields needed by both transformers ensures member objects always
have enough data to produce stable entity refs.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
@backstage/plugin-catalog-backend-module-msgraph is only used
transitively via msgraph-incremental; no direct import exists in
packages/backend/src/index.ts.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
- Wrap userTransformer call for group members in try/catch so a sparse
object that causes the transformer to throw logs a warning and skips
that member rather than failing the entire groups page
- Split PAGE_SIZE into USER_PAGE_SIZE (999) and GROUP_PAGE_SIZE (100)
so each burst stays within its time budget despite per-group member
fetching in the groups phase
- Always send $count=true in advanced mode even when no query object is
provided (use empty object spread instead of short-circuit on query)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
- getUserPhotoGated now throws for non-404 errors (401/403/5xx) so
callers can distinguish a missing photo from a real API failure
- Log a debug message when photo loading fails instead of swallowing
the error silently
- Merge transformer-pre-populated spec.members/spec.children with
fetched Graph membership rather than overwriting them
- deriveRestLength now also excludes manual-trigger schedule objects
(not just cron expressions) from being cast as HumanDuration
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
- Use undefined instead of null for MSGraphCursor.nextLink (repo style)
- Avoid mutating caller-provided query object in requestOnePage
- Improve error handling for non-JSON Graph API error responses
- Build entity spec immutably instead of mutating entity.spec directly
- Log debug message when a sparse group member cannot be transformed
- Warn when schedule.frequency is a cron expression (restLength falls back to 8h)
- Use proper MSGraphContext generic instead of unknown in addProvider
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: pillaris <pillaris@adobe.com>
Add workspace:^ alias for the new plugin so the lockfile is consistent
with the packages/backend dependency reference. Remove stale pg@^8.11.0
range no longer referenced by any workspace package.
Signed-off-by: pillaris <pillaris@adobe.com>
Microsoft Graph requires \$count=true whenever the ConsistencyLevel: eventual
header is present, including plain listing requests with no \$filter or \$search.
The previous condition only added \$count when a filter or search was present,
causing the /groups endpoint to silently return an empty value array when
queryMode is set to advanced without a group filter.
Signed-off-by: pillaris <pillaris@adobe.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Introduces a new Backstage backend module that incrementally ingests
users and groups from Microsoft Graph one page at a time, using the
incremental ingestion framework.
Unlike MicrosoftGraphOrgEntityProvider, this module never holds the
full dataset in memory. Each burst processes a single page (up to 999
items), making it suitable for large Azure AD tenants where the
full-scan provider causes memory pressure or OOM failures.
The @odata.nextLink cursor is persisted in the incremental ingestion
marks table, so a pod restart during ingestion resumes from the last
completed page rather than starting over.
Signed-off-by: pillaris <pillaris@adobe.com>
Add `isPending` prop to Alert, Button, ButtonIcon, Table, and
TableRoot, aligning with React Aria naming conventions. The
`loading` prop is deprecated but remains functional as an alias.
CSS selectors now target `data-ispending` instead of `data-loading`
for pending state styling. The `data-loading` attribute is still
emitted for backward compatibility.
Internal Table hooks (`PaginationResult`, `UsePageCacheResult`)
renamed `loading` to `isPending`. The `useTable` hook returns both
`isPending` and `loading` on `tableProps` to preserve backward
compatibility.
Updated docs-ui documentation and stories accordingly.
Signed-off-by: Johan Persson <johanopersson@gmail.com>
The NOT IN guard against null final_entity rows was preventing the
query planner from using parallel workers and (with a future covering
index) index-only scans. Without the covering index it caused a 2.6x
regression on the no-filter path (7s -> 18.4s).
The filtered path already excludes unstitched entities via the
whereNotNull('final_entities.final_entity') in the inner
entityIdSubquery, so no guard is needed there.
The no-filter path now matches 1.49.x behavior. A followup migration
adding a covering index can re-introduce the guard efficiently.
Signed-off-by: Fredrik Adelöw <freben@spotify.com>
Made-with: Cursor
Remove `export` from five internal helper functions that are not part of
the published API and are only used within their own files. The `sleep`
function in GitLabIntegration is kept exported for test access but marked
as `@internal`.
Signed-off-by: Patrik Oldsberg <poldsberg@gmail.com>
Made-with: Cursor
Instead of checking IS NOT NULL against ~520k rows (which adds a
2.6x regression on the no-filter path), exclude the tiny set of
entities where final_entity IS NULL (~3,700 rows). The anti-join
is nearly free and allows the guard to be applied unconditionally,
so facets results consistently exclude not-yet-stitched (or future
tombstoned) entities regardless of whether filters are present.
Also addresses review feedback:
- Add regression test for unstitched entity exclusion
- Extract setupFacetsCatalog helper to reduce test boilerplate
- Tighten assertions: use exact arrays instead of arrayContaining
Signed-off-by: Fredrik Adelöw <freben@spotify.com>
Made-with: Cursor
Benchmarking on a production-like database shows that always applying
the final_entities subquery (even without filters) causes a 2.6x
regression on the no-filter path (5.2s -> 13.5s) due to ~530k
memoized index lookups against final_entities. The FK cascade from
search -> final_entities already guarantees search rows only exist
for entities with a final_entities row, so the constraint is only
needed when filters route through final_entities.
Signed-off-by: Fredrik Adelöw <freben@spotify.com>
Made-with: Cursor
The whereNotNull('final_entities.final_entity') was copied from other
code paths that select the final_entity column, but the facets subquery
only needs entity_id. Removing it preserves exact v1.49 semantics
(search rows only exist for stitched entities anyway, due to the FK
cascade from search -> final_entities) and avoids an inconsistency
where the no-filter path did not exclude unstitched entities while the
with-filter path did.
Signed-off-by: Fredrik Adelöw <freben@spotify.com>
Made-with: Cursor
Adds standalone test coverage for the facets method in
DefaultEntitiesCatalog to ensure it works correctly with:
- Predicate query filtering (simple key match)
- Predicate query filtering using $in operator
- Compound allOf filters
- Compound anyOf filters
- Both filter and query combined
Signed-off-by: Fredrik Adelöw <freben@spotify.com>
Made-with: Cursor
Route the EXISTS-based filters through final_entities (one row per
entity) instead of correlating against the search table directly.
This avoids the pathological case where correlated subqueries scan
the much larger search table for every row in the outer facets query.
Signed-off-by: Fredrik Adelöw <freben@spotify.com>
Made-with: Cursor
- Add aria-label="Open calendar" and aria-hidden on icon for the calendar trigger button
- Add aria-hidden on nav button icons (React Aria provides built-in labels for prev/next slots)
- Update React Aria docs link to react-aria.adobe.com format
Signed-off-by: Charles de Dreuille <charles.dedreuille@gmail.com>
Made-with: Cursor
- Fixed size: 1.5rem (small), 2rem (medium) in rem units
- No padding, flex-centered icon
- Inset focus ring using 1px box-shadow with --bui-ring token
- Border radius bumped to radius-2
- Medium size padding-inline-end reduced to space-1
Signed-off-by: Charles de Dreuille <charles.dedreuille@gmail.com>
Made-with: Cursor
- Switch DateRangePicker to use the BUI Popover component (with hideArrow)
instead of the raw react-aria-components Popover
- Add WithUnavailableDates story showing isDateUnavailable with weekends
blocked and non-contiguous ranges disallowed (default behaviour)
Signed-off-by: Charles de Dreuille <charles.dedreuille@gmail.com>
Made-with: Cursor
Adds a new DateRangePicker component to @backstage/ui built on React Aria,
featuring a custom field group with two DateInput fields and a calendar
trigger, a RangeCalendar popover with proper range selection visuals
(solid circles for start/end, transparent solid fill for in-range cells
with row-boundary rounding), and full BUI token usage including bg
consumer auto-increment. Includes Storybook stories and a docs-ui page.
Signed-off-by: Charles de Dreuille <charles.dedreuille@gmail.com>
Made-with: Cursor
pillaris