* Add Context Center: Migrate Knowledge Center , Images/ PDFs document support
* Add Context Center: Migrate Knowledge Center , Images/ PDFs document support
* Address PR #27558 review comments
- KnowledgePageRepository: null-safe pageType in getHierarchyWithSearch
and getHierarchyWithSearchForActivePage so the /search/hierarchy
endpoint no longer NPEs when the pageType query param is omitted. The
ES/OS client helpers already skip the pageType term when the value is
null or empty, so this is a pure null-guard.
- ContextFileResource.uploadFile: when a failure happens after the
ContextFileContent row is created (e.g. inside extractionService.submit),
the cleanup path now hard-deletes that content row so the DB is not
left with an orphaned record.
- ContextFileResource: replace the raw Content-Disposition string with a
buildContentDisposition helper that emits both the legacy quoted
filename= and the RFC 5987 filename*=UTF-8'' parameter with
percent-encoded bytes, so international filenames round-trip while
staying header-injection safe. sanitizeFileName also falls back to
"download" on null/blank input.
- ContextFileResourceTest: new cases for sanitizeFileName null/blank
fallbacks and for buildContentDisposition ASCII/unicode/space/injection
behaviour (18 tests, all passing).
* Address copilot review comments on PR #27558
- AssetRepository.getByFqnPrefix: swap arguments so (assetType, fqnPrefix)
matches the DAO signature — previous ordering always missed the index.
- FolderResource / ContextFileResource getEntitySpecificOperations: return
List.of() instead of null so callers iterating the returned list cannot
NPE.
- SearchUtils.getPageHierarchy: replace UUID.fromString with a parseUuid
helper that returns null for missing/malformed values and logs a warning
instead of failing the whole hierarchy response.
- DaoListFilter: qualify the pageType column with the caller-provided
tableName, rename getArticleCondition to getPageTypeCondition (legacy
no-arg method kept as @Deprecated wrapper for compatibility).
- Elastic/OpenSearch client processPageHierarchyHits: replace the per-hit
getChildrenCountForPage search (N+1) with a single pass over the batch
that derives childrenCount from pages whose parent is in the same
result set. Also drops the now-unused helper and its throws clause.
- openmetadata-sdk/pom.xml: mark JWT, JAX-RS client, Apache HttpClient,
jakarta.json, parsson, and JUnit Jupiter as <optional>true</optional>
so they don't leak into SDK consumers that only use the core client.
- InMemoryAssetService: use the shared AsyncService executor for upload
/read/delete instead of the JVM common ForkJoinPool.
- sample-pricing.xlsx: replace the plain-text placeholder with a real
minimal XLSX workbook so detection-based and extraction-based code
paths see a valid Microsoft Excel 2007+ file.
* Use one filters aggregation for page hierarchy childrenCount
Follow-up to b8458e2868. The previous fix derived childrenCount from
pages whose parent appeared in the same batch — that worked for
listPageHierarchyForActivePage (which fetches all depths) but always
returned 0 on the plain listPageHierarchy path (which only fetches one
depth), so top-level listings lost the count semantically.
Replace with a single `filters` aggregation keyed by page id: each
named bucket matches descendants via a fullyQualifiedName prefix query
against the page's FQN. That gives accurate direct-descendant counts
for every returned page in one aggregation round-trip, still O(1)
additional search requests regardless of batch size.
* Add allowedFields entries for contextFile, folder, page
Fixes SearchSettingsHandlerTest.testEveryAssetTypeHasCorrespondingAllowedFields.
searchSettings.json already had assetTypeConfigurations for contextFile,
folder, and page but no matching allowedFields entries, so the test that
asserts every assetType has a corresponding allowedFields block failed
with 'Asset type contextFile has no corresponding allowedFields entry'.
Adds the three missing blocks with the fields that each index actually
exposes — name / displayName (with .keyword and .ngram variants),
description, fqn, fqnParts, tags/tier/domains/dataProducts, plus
entity-specific fields (fileType/contentType/extractedText for
contextFile, parent.displayName for folder/page, pageType for page).
* Fix ui checkstyle
* Fix Java checkstyle
* Address PR #27558 copilot review round 2
- ES/OS populateChildrenCounts: add fqnDepth == parentDepth + 1 to the
per-page filter so childrenCount is direct children only, matching the
field name and the UI's isLeaf check semantics. Previously matched all
descendants.
- ES/OS buildPageNestedSearchHierarchy: filter out hits with a null id
before Collectors.toMap, which would otherwise NPE when SearchUtils
drops a malformed UUID.
- SearchUtils.getPageHierarchy: wrap PageType.fromValue in a parsePageType
helper that logs and returns null on unknown values, so a single bad
hit can no longer break the whole hierarchy response.
- TestSuiteBootstrap.setupMinIO: pin minio/minio to
RELEASE.2024-01-16T16-07-38Z instead of :latest so a newly-published
image cannot break integration tests without a code change.
- createContextFile.json: rewrite the assetId description to be provider
agnostic (S3 / Azure Blob / in-memory / no-op) and flag it as the legacy
path, preferring headContentId / ContextFileContent.
* Update generated TypeScript types
* Address PR #27558 copilot review round 3
- bootstrap/sql/migrations/native/2.0.0/mysql/schemaChanges.sql:
- asset_entity: add PRIMARY KEY (id); mark all generated columns STORED
for consistency with the other drive/knowledge tables in the same
migration; compute deleted as a real boolean via
IFNULL(JSON_EXTRACT(json, '$.deleted'), FALSE) so the boolean index
behaves correctly.
- knowledge_center: mark name, updatedAt, updatedBy, pageType as STORED
and apply the same deleted expression so the existing indexes on
name and (fqnHash, deleted) are reliable on fresh installs.
- drive_folder / context_file / context_file_content: update the
deleted generated column to use the same boolean-safe expression.
- ElasticSearch/OpenSearch hierarchy search: add an explicit sort on
fullyQualifiedName ASC with _id ASC as tiebreaker so from/size
pagination is deterministic and cannot skip/duplicate pages between
requests.
* Fix UI checkstyle
* Address PR #27558 copilot review round 4
- createPage.json: rewrite the field descriptions for name, displayName,
owners, reviewers, and entityStatus. They were copy/pasted from other
schemas ('query', 'tag') and were misleading in generated docs and
clients.
- NoOpAssetService.generateDownloadUrlWithExpiry: return asset.getUrl()
instead of a synthetic 'https://cdn.example.com/...' URL. The old
behaviour let clients attempt downloads that would never resolve when
object storage was disabled; returning the asset's own (empty) URL
surfaces the misconfiguration cleanly.
- AzureAssetService: normalize the prefix path the same way S3 does.
Previously a null/blank prefix produced the literal 'null/' prefix,
writing blobs under the wrong key. New formatPrefix returns "" for
null/blank and ensures exactly one trailing '/' for a real prefix.
- AssetRepository.getByFQN: treat null *or* empty list as 'not found',
matching getByFqnPrefix. Callers previously received an empty list
silently when the DAO returned [] instead of a 404.
* Update generated TypeScript types
* Fix UI checkstyle
* Address PR #27558 copilot review round 5
- AssetDAO.update / AssetRepository.update: switch the UPDATE target from
fqnHash to id. Two assets can share the same fullyQualifiedName (e.g.
successive revisions of the same context file), so the old SQL could
silently update sibling rows.
- ContextFileExtractionService: run the extraction pipeline on a
dedicated fixed thread pool instead of AsyncService.getExecutorService.
process() blocks on assetService.read(...).join(), and S3/Azure reads
are themselves scheduled on AsyncService — sharing the same bounded
pool risks starving those reads (and deadlocking) once every thread is
busy running extractions.
- postgres/schemaChanges.sql: wrap the generated deleted column in
COALESCE((json ->> 'deleted')::boolean, false) (and the asset_entity
CAST variant) so an absent 'deleted' key is stored as FALSE, not NULL.
Otherwise "non-deleted" filters based on the boolean index drop rows
silently. Matches the MySQL IFNULL(..., FALSE) side of the migration.
- ContextFileUploadSupport.sanitizeEntityName: treat null/blank input as
'file' instead of NPE-ing on replaceAll. Multipart uploads can arrive
without filename metadata; the upload should still succeed with a
stable generated name.
* Remove macOS-only @rollup/rollup-darwin-arm64 dev dep
I pinned this during local troubleshooting to get a Vite dev server
running on macOS (rollup's optional native binary was missing). CI runs
on Linux, where yarn install --frozen-lockfile refuses the package
('The platform \"linux\" is incompatible with this module'), which
broke license-header, lint-src, lint-playwright, i18n-sync, app-docs,
and ui-coverage-tests for PR #27558.
rollup re-resolves its native binary per platform — there's no reason
to pin the darwin one. Remove it from package.json and drop the
matching '@rollup/rollup-darwin-arm64@^4.60.2' block from yarn.lock.
* Re-declare optional SDK test deps on integration-tests classpath
KnowledgeCenterIT failed in CI with
'java.lang.NoClassDefFoundError: org/glassfish/jersey/apache/connector/ApacheConnectorProvider'
after I marked the JAX-RS client stack in openmetadata-sdk as
<optional>true</optional> during review round 2. That change stops the
deps from leaking to every SDK consumer, but integration-tests actually
uses org.openmetadata.sdk.test.util.RestClient, so the optional deps
must be re-declared on its own classpath.
Adds jakarta.ws.rs-api, jersey-client, jersey-apache-connector,
httpclient, jakarta.json-api, and parsson to
openmetadata-integration-tests/pom.xml as <scope>test</scope>.
* Fix IT failures from CI integration-tests-mysql-elasticsearch
1. MySQL deleted column: revert the IFNULL wrapper to plain
(json -> '$.deleted'). My earlier
IFNULL(JSON_EXTRACT(json, '$.deleted'), FALSE) hit
'Incorrect integer value: false for column deleted' on fresh installs
because MySQL cannot coerce the resulting JSON scalar into TINYINT(1)
when the column is STORED. The bare '(json -> '$.deleted')' form is
what other OM tables already use, and MySQL converts JSON true/false
to 1/0 directly for the BOOLEAN column. STORED + PRIMARY KEY stay
in place.
2. DriveFileUploadIT: raise the four short atMost(5s) awaits to 20s
with explicit pollDelay(ZERO) + pollInterval(200ms).
K8sOMJobOperatorIT sets a global Awaitility pollInterval of 5s at
class setup; any subsequent test with atMost <= 5s hits
'Timeout must be greater than the poll delay'. Overriding the
per-call poll settings insulates these asserts from the global
leak.
* Document SDK test-utility optional deps
In review round 2 we marked jersey-client, jersey-apache-connector,
jakarta.ws.rs-api, httpclient, jakarta.json-api, parsson, java-jwt, and
junit-jupiter-api as <optional>true</optional> on openmetadata-sdk so
that core SDK consumers don't inherit a heavy JAX-RS + JUnit stack.
openmetadata-integration-tests hit this immediately with
NoClassDefFoundError from RestClient; its own pom now re-declares the
deps.
Add a "Test utilities" section to the SDK README that lists the
optional deps downstream test-utility consumers must re-declare (with
the concrete <scope>test</scope> XML snippet) and explains the error
they'd otherwise see.
* NoOpAssetService: never return null from generateDownloadUrlWithExpiry
In review round 4 I changed this method to return asset.getUrl() when
the asset is non-null. But Asset.url is optional in the schema, so
asset.getUrl() itself can be null — which breaks the implied "never
returns null" contract downstream callers rely on (AttachmentResource
only null-checks defensively).
Normalize null and blank URLs to an empty string so the method's
non-null, non-blank contract holds even when storage is disabled and
the asset was never populated with a URL.
* AssetServiceFactory: swap to NoOp when re-initialized with storage off
init(...) previously only assigned NoOpAssetService when instance was
null. On a re-init with object storage toggled off (config reload, test
teardown, etc.), the previously wired S3/Azure/InMemory provider stayed
live and kept serving real IO against a backend the operator thought
was disabled.
Replace the instance with a fresh NoOp when storage is disabled unless
the instance is already a NoOp (idempotent on repeated disabled
inits).
* Type create-request domains arrays as fullyQualifiedEntityName
The three new KC/Drive create schemas (createFolder, createContextFile,
createPage) had domains as an array of unconstrained strings. The rest
of the OM API models domain references as FQNs, and the shared
basic.json#/definitions/fullyQualifiedEntityName is the convention for
this.
Point all three items refs at fullyQualifiedEntityName so generated
clients see a consistent FQN type and requests get validated for
non-empty length/format rather than any string.
* Update generated TypeScript types
* Address PR #27558 copilot review 4144965142
- ContextFileExtractionService: switch the default thread pool to
a static final DEFAULT_EXECUTOR, so every production instance of the
service reuses the same pool instead of leaking a fresh fixed pool
per construction (tests especially create multiple instances).
Threads remain daemons, so the pool never blocks JVM shutdown.
- ObjectDeleteQueueService: when queueCapacity is 0, use a
SynchronousQueue so "reject-if-all-workers-busy, no buffering" holds.
Previous Math.max(1, queueCapacity) silently allocated a 1-slot
ArrayBlockingQueue, contradicting the caller's stated capacity and
potentially buffering one task past the semaphore's accounting.
Not fixing:
- SearchUtils @Slf4j 'LOG' vs 'log'. OM's openmetadata-service/lombok.config
sets 'lombok.log.fieldName = LOG', so @Slf4j correctly generates
'LOG' for every class in this module. The reviewer's concern only
applies to projects without that directive. Verified clean compile.
* Address PR #27558 copilot review 4144917449
- knowledgeCenterTags.json: change mutuallyExclusive from the string
"false" to the JSON boolean false. The Classification schema declares
this as `"type": "boolean"`; jackson's lenient string->boolean
coercion masked it until now, but strict validators would reject and
the other OM bootstrap tag files that use the correct boolean
(piiTagsWithRecognizers.json) model what this should look like.
- ContextFileExtractionService.process: guard the updateContent
updater with the same head-content check already used in
updateFile. Previously, if headContentId flipped between the
initial check and the status writes, updateFile would no-op while
updateContent still marked the now-stale content "Analyzing",
leaving it stuck once the later early-return fires.
- AzureAssetService.upload: stream the InputStream straight to the
blob using the known asset.getSize() instead of reading the whole
payload into a byte[] via IOUtils.toByteArray. Matches the S3
streaming behaviour and avoids full-file heap pressure / OOM risk
on larger files. Buffered fallback retained when size is unknown.
- Size fields modeled as integer: flip fileSize / size on
createContextFile.json, contextFile.json, asset.json,
createAsset.json, and contextFileContent.json from
"type": "number" to "type": "integer" with "format": "int64" and
"minimum": 0. Byte counts are inherently whole numbers; floating
point loses precision above 2^53 and makes validation murky.
Update the (double) call sites in ContextFileResource,
ContextFileUploadSupport, and AttachmentResource to match.
Not fixing:
- ContextEntityPromptService "unused Authorizer import" — false
positive, the class uses it in the constructor.
- NoOpAssetService.generateDownloadUrlWithExpiry null return — already
fixed earlier in commit a4a2dcc91d (returns "" when url is
null/blank).
* AssetService.read: run inline instead of hopping through AsyncService
Every caller of AssetService.read(...) immediately .join()s on the
returned future:
- ContextFileExtractionService.process reads + extracts
- ContextFileResource.downloadFile reads + streams back
- AttachmentResource.serveAsset reads + streams back
- QueuedDeleteAssetService just delegates
None of them exploit the async nature, but the S3/Azure/InMemory
implementations all wrapped the blocking fetch in
AsyncService.executeAsync or CompletableFuture.supplyAsync on a
bounded pool. That created a starvation path when any caller thread
was already running on AsyncService (or could monopolize it under
load) — join() would block the caller while the submitted read
task fought for a free worker.
Switch S3, Azure, and InMemory read() to execute on the caller's
thread and return CompletableFuture.completedFuture(...). Interface
is unchanged so existing .join() callers keep working; the extra
thread hop and the potential for AsyncService starvation are both
gone. Combined with the dedicated context-file-extraction pool, the
extraction pipeline no longer touches AsyncService for any
asset-read step.
* Address PR #27558 copilot review 4151211562
- FolderIndex / ContextFileIndex: stop re-setting entityType, deleted,
owners, totalVotes inside buildSearchIndexDocInternal. Those common
fields are populated by populateCommonFields in the SearchIndex
template method (Phase 1) before Phase 3 calls the entity-specific
internal builder, so the explicit puts were redundant and silently
overrode the template output. Aligns with PageIndex convention and
updates the unit tests to assert the internal builder sets only
entity-specific fields.
- ContextFileTextExtractor: bound the Tika BodyContentHandler at
MAX_CANONICAL_TEXT_LENGTH instead of passing -1 (unbounded) so a
pathological image cannot drive OCR to accumulate arbitrary output
on the heap.
- ContextFileExtractionService: replace the unbounded
Executors.newFixedThreadPool backing queue with a ThreadPoolExecutor
using an ArrayBlockingQueue + AbortPolicy. Without a bounded queue
the RejectedExecutionException handling in submit(...) was dead
code; with it, an overloaded server surfaces a "retry later"
failure status instead of silently accumulating work.
- S3AssetService / AssetService / AssetServiceFactory /
QueuedDeleteAssetService: make AssetService extend AutoCloseable
with a default no-op, override close() in S3AssetService to release
the S3Client and S3Presigner connection pools, and register a
shutdown hook in AssetServiceFactory that closes the current
provider on JVM exit (and on re-init when the provider changes).
- bootstrap 2.0.0 MySQL schemaChanges: change the deleted generated
column from (json -> '$.deleted') to
(JSON_EXTRACT(json, '$.deleted') IS TRUE) so rows where the JSON
key is absent resolve to FALSE instead of NULL. Avoids filter misses
on the composite (fqnHash, deleted) index.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Fix Java checkstyle
* Fix integration test compile + S3 generateDownloadURL
ContextFileIT / DriveFileUploadIT compile failures came from the
fileSize schema switch to integer/int64 — the generated setter/getter
is now Integer. Replace the double literals with ints and the
assertEquals(double, ...) sites with intValue() so the (int, int)
overload resolves unambiguously.
Also override S3AssetService.generateDownloadURL to return a
short-lived presigned URL (mirroring AzureAssetService) instead of
inheriting the default, which would return the raw S3 key from
asset.url. Addresses review 4151282021.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Revert MySQL deleted column back to bare json -> expression
The JSON_EXTRACT(...) IS TRUE form broke integration tests — GET after
create started returning 404, consistent with MySQL evaluating the
IS TRUE predicate against the JSON scalar in a way that stored 1
instead of 0 for freshly-created rows (deleted=false).
Restoring the bare (json -> '$.deleted') expression used pre-review.
Rows with the key missing will store NULL on the generated column,
which is a theoretical concern the review flagged but does not affect
current code paths (all inserts write json.deleted explicitly).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Fix Transi18next import path in KnowledgeCenter components
Two KnowledgeCenter files imported Transi18next from
'utils/CommonUtils', which is where Collate's UI re-exports it from.
OpenMetadata core exports Transi18next from 'utils/i18next/LocalUtil'
(same path every other core file uses). The Collate-style import
broke the production Vite/Rollup build.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Harden ContextFileIT.testFileAppearsInSearch against async indexing
The test used a fixed Thread.sleep(2000) then a single assertEquals
on the status code. That was flaky two ways: ES indexing is async
and the 2s window is not always enough, and on a fresh cluster the
context_file_search_index itself may not exist yet at first query
(yielding 500).
Replace with an await() loop that polls every 200ms for up to 30s
and asserts both status==200 AND that the newly-created file's UUID
appears in the response. Matches the assertSearchContainsFile
helper in DriveFileUploadIT.
Also URL-encode the namespaced query string so the uniqueName
does not break the query parsing.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Make playwright editor shortcuts platform-aware
The SHORTCUTS constant in playwright/constant/KnowledgeCenter.constant.ts
hard-coded "Meta+b" / "Meta+z" / etc. On macOS Meta is Cmd and those
shortcuts trigger bold / undo / copy as expected, but on the Linux CI
runners Meta is the Super (Windows) key — so every ProseMirror
formatting and history test just pressed Super+b, which does nothing,
and the test then fails waiting for the <strong>…</strong> element
(or for the undone text to disappear).
Detect the runner platform and use Meta on macOS, Control everywhere
else — matching the same pattern in src/constants/KnowledgeCenter.constant.ts.
Unblocks the 6 KnowledgeCenterTextEditor failures across Admin / Data
Consumer / Data Steward roles (Text Formatting + Undo/Redo). Slash
commands keep passing because they don't depend on modifier keys.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Run prettier on DateTimeUtils.ts
CI's lint-src job fails because ESLint+Prettier --fix produces a
non-empty diff against the committed tree. Local prettier pass
trimmed the indentation and added a trailing comma in the imports
block. No behavioral change.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Fix Knowledge Page entity-link + DAO filter regressions from the port
Downloaded the failing playwright traces from the PR's postgres e2e run
and walked each one. Three distinct bugs, all present because the
Collate-side overrides (overrides/EntityUtilClassCollate.ts and the
DaoExtension.KnowledgeExtensionDAO custom SQL) were not carried over
into OpenMetadata core when KnowledgeCenter was merged up.
1) CollectionDAO.KnowledgePageDAO: override listCount / listBefore /
listAfter (plus helper SQL queries) so that
`GET /v1/knowledgeCenter?entityId=X&entityType=topic` actually INNER
JOINs entity_relationship and returns only pages whose
relatedEntities contains the target entity. Without this the base
EntityDAO ignored entityId/entityType entirely and returned every
page, which is why the "Knowledge Articles" widget on a data asset
page showed the 15 fixture articles instead of the one just attached
— and why updateDataAsset timed out waiting for the linked article.
Uses OWNS relation for user/team filters (same semantics Collate
uses) and HAS for every other entity type.
2) EntityUtilClassBase + EntityUtils.getEntityLinkFromType: add
EntityType.KNOWLEDGE_PAGE cases that route to getKnowledgePagePath.
Before this, mention notifications for Knowledge Pages fell through
to the default `/table/<fqn>` branch (confirmed in the captured
page-snapshot: the mention link pointed at `/table/Article_eEqrWeeU`),
which 404'd on the Table API and rendered an error page — so the
entity-header-display-name textarea never appeared and the User
Mentions test timed out. Search results on Explore had the same
problem, rendering every Knowledge Page result card with href="/".
3) EntityUtilClassBase.getEntityByFqn / ENTITY_PATCH_API_MAP /
getResourceEntityFromEntityType: handle KNOWLEDGE_PAGE end-to-end so
the detail-page fetch, patches, and policy lookups all route through
the knowledgeCenter REST API rather than falling back to the generic
entity utilities (which don't know about the 'page' entity type).
Verified against the real trace artifacts from CI run 24790718035:
- shard 3 Knowledge Center page test — widget shows 10 unrelated
"Article_*" fixture items instead of the created one → root cause
is the missing DAO JOIN (#1).
- shard 3 User Mentions test — notification link is /table/, not
/knowledge-center/ (#2).
- shard 3 Reviewer Workflow — data consumer's knowledge-center goto
renders "No data available" because getEntityByFqn fell back to a
table fetch for a page FQN (#3).
- shard 5 ExplorePageRightPanel_KnowledgeCenter (22 failures) —
search result card links are "/explore/" (empty), same root cause
as (#2) inside getEntityLinkFromType default branch.
Compiles: mvn -pl openmetadata-service -q -DskipTests compile passes;
tsc --noEmit reports no new errors in the touched files.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Address remaining PR #27558 review feedback
Seven actionable fixes drawn from the still-open review threads; the
rest of the open threads in copilot's bot reviews are either already
addressed in earlier commits or stale against the current code and
are being resolved on the review UI alongside this commit.
- AssetRepository.getByFQN: the LOG.error message said "asset with id"
but was printing the FQN. Relabel to "asset with FQN" for accurate
troubleshooting (thread #42).
- KnowledgePageMapper.createToEntity: stop mutating the inbound
CreatePage by calling create.withRelatedEntities(...). Build the
effective list as a local variable and pass it to copy(...). Prevents
the Organization fallback from leaking into the caller's request
object, which is surprising when the request is re-used or logged
(thread #43).
- FolderIndex: default childrenCount to 0 when the entity hasn't yet
had its children recomputed (e.g. a freshly created folder). Prevents
the numeric field from being indexed as missing, which broke range
and sort queries that assume it is always present (thread #46).
- NoOpAssetService and InMemoryAssetService: override
generateDownloadURL to delegate to generateDownloadUrlWithExpiry,
matching S3/Azure. Without this, callers using the non-expiry API
got asset.getUrl() (often empty for these providers), yielding broken
download links (threads #39, #45).
- ObjectDeleteQueueService: register a JVM shutdown hook in the
singleton's initializer that calls stop(). The service already
implements Dropwizard Managed, but nothing currently wires it into
the application lifecycle, so non-daemon delete-worker threads were
at risk of keeping the JVM alive after ungraceful termination. The
hook is a belt-and-suspenders fallback to the Managed path
(threads #52, #53).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Add java-checkstyle skill for Claude + Codex agents
CI keeps surfacing "Java checkstyle failed — please run mvn spotless:apply"
comments on PRs (including this branch). CLAUDE.md and AGENTS.md already
mentioned the command, but a one-line prose note in the middle of each file
wasn't enough to make it a reliable habit.
This commit:
- Adds a dedicated invocable skill at .claude/skills/java-checkstyle/SKILL.md
(for the Claude Code harness) and a mirror at
.agents/skills/java-checkstyle/SKILL.md (for Codex-style agents). Both
describe the same procedure: when / why to run spotless, the `-pl <module>`
scoping option, the verify-only `spotless:check` form, the expected
diff shape, and the rule to never hand-edit formatting around a plugin
error.
- Promotes the existing one-liners in CLAUDE.md and AGENTS.md to explicit
"run before finishing any Java task" instructions, pointing at the skill so
agents have a reusable procedure to invoke rather than improvising.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Harden AttachmentResource upload/download against three regressions
Carried over from the latest AttachmentResource review. Three issues:
1. Content-Disposition header injection (security) — downloadAsset() built
the header by direct string interpolation of asset.getFileName(). A
filename containing double-quotes or CRLF could inject arbitrary HTTP
headers. ContextFileResource already has a sanitize + RFC-5987 encode
helper; rather than duplicate it, promote
ContextFileUploadSupport.sanitizeFileName / buildContentDisposition to
public, delete the duplicates from ContextFileResource (now delegators),
and reuse the shared helpers from AttachmentResource.
2. Unbounded upload buffering (performance / DoS) — createAssetFromUpload
read the full multipart body into a byte[] via IOUtils.toByteArray
before checking against MAX_FILE_SIZE. An attacker could send an
arbitrarily large body and exhaust heap before the validation ran.
Replace with ContextFileUploadSupport.bufferUpload(), which streams to
a bounded temp file and throws MaxFileSizeExceededException the moment
the configured limit is passed; translate that into the same
AttachmentException size-validation error the previous code raised.
Promoted BufferedUpload and MaxFileSizeExceededException to public so
the attachments package can consume them.
3. Startup NPE when objectStorage is null (bug) — initialize() called
config.getObjectStorage().getMaxFileSize() without a null guard, so a
deployment that doesn't configure object storage would NPE on server
start. Added the same guard ContextFileResource.initialize() already
uses, gave MAX_FILE_SIZE a safe 5 MiB default, and also null-guarded
the S3-configuration branch of the CDN URL lookup so a pure-Azure or
pure-NoOp setup doesn't fall off the end of the ternary.
Ran mvn spotless:apply — picks up formatting-only changes in
CollectionDAO.java and FolderIndex.java as a side effect of the shared
helper additions.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Add ui-checkstyle skill + fix residual import-order drift
CI's UI Checkstyle workflow has three per-area jobs (lint-src,
lint-playwright, lint-core-components) that reformat the files changed
in the PR and fail if the reformat produces a diff. CLAUDE.md and
AGENTS.md didn't previously document this flow, so re-running the fix
was a guessing game — the two lint-core-components and lint-playwright
failures on this branch came from stale import order left over from the
main→context_center merge.
This commit:
- Adds a dedicated invocable skill at .claude/skills/ui-checkstyle/SKILL.md
(Claude Code harness) and a mirror at .agents/skills/ui-checkstyle/SKILL.md
(Codex-style agents). Both describe the exact three-command sequence CI
runs — organize-imports-cli → eslint --fix → prettier --write — the
per-area file scoping, the `--check` dry-run mode, and the rule that
organize-imports must run BEFORE prettier (otherwise the indentation /
trailing-comma round-trip leaves a dirty diff).
- Promotes the existing one-liner in CLAUDE.md and AGENTS.md to an explicit
"run before finishing any UI task" instruction that points at the skill.
- Fixes two residual import-order drifts (KnowledgePagesHierarchy.tsx,
EntityUtilClassBase.ts) surfaced by running the skill's sequence locally.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Fix UI checkstyle on EntityUtilClassBase.ts
ESLint --fix inserted a blank line between the KNOWLEDGE_PAGE guard and the
fallback return in getEntityByFqn. Committing the formatted version.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Fix ContextFileIT.testFileAppearsInSearch flaky 500 from query_string parsing
The previous polling search used the namespaced unique name as a free-text
q= argument. The namespace prefix contains '-' which the ES 9.x query_string
parser treats as a NOT operator, producing a deterministic 500 across the
full 30s polling window even when the document was indexed.
Switch to the direct get-by-id endpoint (/v1/search/get/{index}/doc/{id}),
which performs a real-time ES GET with no query_string parsing and no
analyzer involvement — the most reliable signal that the document was
indexed. Bump the timeout to 60s and capture the response body on any
non-200 so future regressions surface the real ES error.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* Fix knowledge center icon
* update knowledge center to context center
Co-authored-by: Copilot <copilot@github.com>
* Revert "update knowledge center to context center"
This reverts commit f0cca5fd65.
* Fix UI checkstyle: sort tag*-related imports in SearchClassBase
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Fix Jest coverage failures in KnowledgeCenter Layout and right panel
KnowledgeCenterLayout was importing i18n directly from LocalUtil, but the
global setupTests mock for that module only exposes t/on. Switch to the
useTranslation() hook so it picks up the react-i18next mock that already
provides i18n.dir(), matching how LeftSidebar and RichTextEditor use the
direction.
EntityRightPanelClassBase.getKnowLedgeArticlesWidget now returns the
KnowledgePages component instead of null. Update the corresponding test
case to assert the new return value.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* Fix playwright tests and bugs
Co-authored-by: Copilot <copilot@github.com>
* Fix checkstyle
* Fix /knowledgeCenter/search/hierarchy 500 by removing _id sort
ES 9.x and OpenSearch 3.x reject sorts on the _id field by default
(indices.id_field_data.enabled is false), causing every call to
listPageHierarchy{,ForActivePage} to fail the search_phase_execution_exception
"all shards failed" we see in the screenshot. The _id sort was added
in 4a75852a7e as a tiebreaker for from/size pagination, but
fullyQualifiedName is already a keyword field with doc_values and is
unique per page (name is unique within a parent's children) — so no
tiebreaker is needed.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* Cascade hard-delete to descendant pages in search index
KnowledgeCenter pages are nested via FQN (parent.fqn -> parent.fqn.child),
not via a parent.id field on the child doc. The default deleteOrUpdateChildren
case for entity type "page" uses page.id field matching, which doesn't exist
on child page docs — so a recursive hard-delete on the parent removed the
parent from search but left every descendant orphaned in the index. Stale
docs only disappeared on a full reindex.
This logic was overridden in the collate fork's SearchRepositoryExt; it was
lost during the migration when the override class was removed. Fold the
override into the base SearchRepository as a Page-specific case that calls
deleteEntityByFQNPrefix, which deletes by fullyQualifiedName.keyword prefix
match — covering every descendant.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* Add page/folder/contextFile/securityService to SearchIndexingApp picker
The Search Indexing Application's "Entities" picker shows "No data" when
typing "Page" because the enum in src/utils/ApplicationSchemas/SearchIndexingApplication.json
does not include the Knowledge Center / Drive entity types added on this
branch. The collate fork carried these in SearchIndexingApplication-collate.json
(included page); folder, contextFile and securityService are new on this
branch and never made it into the picker enum during the migration.
Without them in the enum, users cannot select these entity types for
targeted reindex, even though every other reindex code path supports them.
src/jsons/applicationSchemas/* is generated by parseSchemas.js from
src/utils/ApplicationSchemas/* at build time and is gitignored, so only
the source schema is updated here.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* Restore live index settings on per-entity distributed-promote path
DefaultRecreateHandler exposes two finalization paths:
- finalizeReindex(...) — centralized end-of-job promotion. Calls
applyLiveServingSettings + maybeForceMerge
before the alias swap, reverting the bulk
overrides (refresh_interval=-1, replicas=0,
async translog) back to live values
(refresh=1s, replicas=1, durable translog).
- promoteEntityIndex(ctx, ok) — per-entity promotion. Used by the distributed
search-indexer's "promote as soon as all
partitions for an entity complete" callback
(DistributedSearchIndexExecutor.promoteEntityIndex).
Swaps the alias and cleans up old indices —
but never restored live settings.
When an entity finishes its partitions before the final reconciliation
(typically the smallest entities — e.g. knowledge `page` with ~11 rows),
its index is promoted via the per-entity path, the alias swap succeeds,
and the bulk-build overrides become the new live settings. refresh_interval
stays at -1 in production, so live writes after the reindex are buffered in
the translog and never reach searchable segments until a manual _refresh.
Externally this surfaces as "create an article, hierarchy is empty until I
re-trigger reindex" — exactly the user-reported bug.
Mirror the finalizeReindex sequence by calling applyLiveServingSettings
(and maybeForceMerge for parity) at the top of the promote block in
promoteEntityIndex, before the alias swap.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* Wire jobData into per-entity reindex promotion handler
DefaultRecreateHandler.applyLiveServingSettings reads from the handler's
jobData field (live + bulk index-settings overrides on the EventPublisherJob).
The per-entity distributed-promotion path in DistributedSearchIndexExecutor
created its own DefaultRecreateHandler instance and never called
withJobData(jobData) on it. With jobData=null, buildRevertJson returns null
and applyLiveServingSettings silently no-ops — meaning the previous fix
(b272de85f9) never actually re-applied live settings on the per-entity
promote path, even though the call was reached.
currentJob.getJobConfiguration() is the EventPublisherJob the strategy
created. Wire it into the new handler at construction time, mirroring the
withJobData call DistributedIndexingStrategy already makes on the strategy's
own handler instance.
With this change, the per-entity promote path now logs
"Applying live serving settings to staged index '...' for entity 'page':
{\"number_of_replicas\":1,\"refresh_interval\":\"1s\", ...}"
before the alias swap, and post-promotion `_settings` show
refresh_interval=1s instead of the stuck -1.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* Fix delete failure
* Fix java checkstyle
* Fix article deletion issue
* refactor(test): streamline Knowledge Center List setup and teardown processes
* Fix GlossaryTags
* Add missing pieces in knowledge articles
* Fix checkstyle
* Remove reviewer workflow spec
* remove unused util
* Fix the localization changes
* Fix unit tests
* deleted unused svg
* added missing svg
* improved ux of save button & autofocus on title
* lint fixes
* Update page index
* Make calculateFqnDepth static
* fixed the kc imports
* import fixes
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: Aniket Katkar <aniketkatkar97@gmail.com>
Co-authored-by: Copilot <copilot@github.com>
Co-authored-by: mohitdeuex <mohit.y@deuexsolutions.com>
Co-authored-by: Rohit0301 <rj03012002@gmail.com>
Co-authored-by: Rohit Jain <60229265+Rohit0301@users.noreply.github.com>
Co-authored-by: Harsh Vador <harsh.vador@somaiya.edu>