open-metadata/OpenMetadata 1.12.9-release

on GitHub

Changelog

OpenMetadata 1.12.9 is a maintenance release delivering new connector capabilities, performance improvements for workflow execution, and targeted bug fixes across search, ingestion, UI, and the OpenMetadata Python client.

✨ New Features

• Unity Catalog — incremental metadata extraction #28380: Unity Catalog connector now detects changed tables via information_schema.tables.last_altered and only re-ingests modified entities. Delete detection uses exact catalog matching (preventing wildcard misfires on catalog names containing underscores). Catalog names are validated against an allowlist before SQL interpolation.
• Task domains backfilled + stamped on approval #28402: Approval task threads now inherit the domain of the entity being approved, and a migration backfills domains on existing tasks.
• Workflow entity extended fields #28398: Workflow entities now support inputPorts, outputPorts, and glossaryTerms fields, bringing them to parity with pipeline and data-flow entities.
• MySQL — custom queryHistoryTable #28388: MySQL connector accepts a configurable queryHistoryTable for usage and lineage extraction, enabling use-cases where query history is stored in a non-default location.

🔒 Security

• Test-connection workflow authorization #28414: Test-connection workflow triggers now require proper authorization, closing an unauthenticated execution path.
• Snyk high/critical dependency patches in ingestion #28340: High and critical Snyk findings patched in ingestion dependencies and code paths.
• brace-expansion lockfile bump to 5.0.6 #28244: Resolves a ReDoS advisory in the transitive brace-expansion package.
• js-cookie bumped #28315: Frontend dependency bumped to address a Dependabot advisory.
• WebSocket Dependabot vulnerability in UI #28320: Dependency update for a reported WebSocket vulnerability in the UI bundle.

🎨 UI Changes

Improvements

• Data Quality — column selection dropdowns are searchable #28376: Column selectors in test case forms now include a search box, making it practical to navigate wide tables.

Fixes

• Notification links — plural alerts path and Query href #27918: Notification alert links now route to the correct plural /alerts path; Query entity hrefs in notifications are also corrected.
• Data Asset Header — permission fix #27967: Resolved a permission check issue in the Data Asset Header component that prevented certain actions for non-admin users.
• Bot name search on Bots page #27365: Fixed search not returning results when searching for bots by name on the Bots management page.
• Column bulk-ops filters use displayName #28390: Service, database, and schema filter dropdowns in the column bulk-operations flow now display displayName instead of raw FQN fragments.
• Bulk-asset operations enforce dryRun #28395: Tag, glossary, and data-product bulk operations now correctly honour the dryRun flag, preventing accidental mutations during preview runs.
• Table / dataModel — inline column.extension persisted #28392: Custom extension data attached inline to columns is now saved correctly through POST/PUT calls.
• ServicesPage tab accepts ServiceCategory enum values #28375: The :tab route parameter now accepts both the label string and the raw ServiceCategory enum value, fixing deep-links that used enum values.
• Unknown service category returns 404 #28372: Navigating to an unrecognised service category now returns a proper 404 page instead of an empty fallback.
• KPI widget date format #28370: Added missing space between day and month values in the KPI widget's X-axis date labels.
• Bug fix (#27433) #28266: Backported fix for entity display regression.

🔌 Connectors

Databases

• Databricks / Unity Catalog — valueless tags ingested #28294: Tags set without an explicit value (valueless) are now ingested correctly from Databricks and Unity Catalog sources.
• Snowflake — discovered databases logged with filter reasons #28385: During schema discovery the Snowflake connector now logs each discovered database alongside the reason when it is filtered out, improving debuggability.

Pipelines / Dashboards

• Bulk sink — charts and dataModels created before dashboards #28371: Fixed a bulk sink ordering bug where a Dashboard entity could be flushed before its referenced Charts or DashboardDataModels, causing HTTP 400 rejections from the server. Each topology stage now encodes its position so referenced entities are always created first.
• Power BI — sink buffer flushed before lineage resolution #28308: Ensures all Power BI entities are persisted before lineage resolution begins, preventing reference errors during lineage extraction.
• Power BI — TSQL dialect for Sql.Database M-query lineage #28380: The Power BI connector now parses Sql.Database M-query expressions using the TSQL dialect, fixing lineage extraction failures on SQL Server-backed datasets.

OpenLineage

• Resolve table identity from symlinks facet #28360: OpenLineage events that include a symlinks facet now resolve the true table identity from the symlink, fixing entity matching for Hive/Iceberg tables exposed under alternate names.
• Pipeline as node for single-sided lineage #28350: OpenLineage pipelines can now appear as a standalone node in the lineage graph when only one side (input or output) is present, instead of being dropped.

Ingestion Runtime

• Kubernetes client capped below 36.0.0 #28331: Pins the Kubernetes Python client to <36.0.0 to avoid a breaking API change introduced in that release.
• Ingestion-pipeline status preserved during queued-stage failures #28382: When a queued-stage task fails, the overall pipeline status remains accurate instead of being overwritten with a misleading value.

🛠 API (Backend)

Improvements

• Workflow field fetches scoped #28391: Wildcard field fetches in workflow-related queries replaced with scoped field requests, reducing payload size and improving performance.
• Data Insights enricher — per-step failure isolation #28379: Enricher step failures are now isolated per-step with individual error tracking, preventing one bad enricher from aborting the entire Data Insights run.

Fixes

• Search search_after with special characters in sort values #28386: search_after pagination no longer drops entities whose sort-field value contains special characters.
• Search — nested children flattened to avoid ES mapping depth limit #28387: Deeply nested children fields are now flattened before indexing to prevent Elasticsearch/OpenSearch from rejecting documents that exceed the default mapping depth limit.
• Table certification cascaded to child search documents #28229: A PATCH to a Table's certification tag is now propagated to all child search documents, keeping certification state consistent across the search index.
• Search-index reindex jobs no longer marked failed on clean completion #28381: Fixed a status-tracking bug that incorrectly recorded clean reindex jobs as failed in the job history.
• PDTS duplicate preemption and invalid index-mapping recovery #28373: Migration now preempts duplicate PDTS (Profiler Data Time Series) rows and recovers from invalid search index mappings rather than failing.
• MCP — search_metadata response capped #28383: The MCP search_metadata tool response is now capped in size to prevent LLM context overflow when the result set is large.
• Alert filter functions use strict literal matching #28393: Alert filter expressions now perform strict literal matching instead of substring matching, preventing false-positive alert triggers.
• Test case suite search membership preserved #28271: Test cases now retain their suite membership in the search index after updates, fixing a regression where suites lost members after re-ingestion.
• OMeta SSE transport switched to requests #28293: The OpenMetadata Python client now uses requests instead of httpx for SSE (Server-Sent Events) transport, resolving compatibility issues in certain deployment environments.
• OMeta — resilient transport with keepalive and retry #28389: The OMeta REST transport now uses keepalive connections and automatic retries on transient failures with a typed RestTransport abstraction.
• Shutdown logging captures full streamable log tail #28396: Synchronous shutdown now waits for in-flight streamable log writes to complete, preventing log truncation on graceful shutdown.