DQC Logo
|

v3.6

This release strengthens data improvement workflows, cross-table checks, operational notifications, and AI-assisted work across the DQC Platform. It also includes important reliability improvements for large tables, exports, OData access, worker processing, and air-gapped deployments.

undefined New Features

Multiple improvement workflows per ruleset

  • Rulesets can now be connected to multiple improvement workflows, making it easier to keep different remediation approaches side by side for the same quality checks.

  • The Improvements tab now has a dedicated list endpoint and safer data handling for the one-to-many relationship between rulesets and improvement workflows.

Operational notifications in the Notification Center

  • The Notification Center now also supports operational platform updates, including stale connector syncs that need attention.

  • This continues the move toward one central place for relevant platform events, instead of spreading important updates across different surfaces.

Usage and consumption reporting API

  • Added API support for tenant and cluster-admin usage reporting.

  • This creates a clearer basis for monitoring platform consumption across tenants and environments.

Volume rule weighted moving average baseline

  • Volume rules now support a weighted moving average baseline with configurable bounds per check.

  • This helps detect unusual volume changes more reliably in time-dependent datasets.

PDF chat visual feedback

  • PDF chat now gives clearer visual feedback while users work with documents.

  • Context file loading has also been made more robust, especially when files are used by assistants or indexers.

undefined Improvements

DQ Assistants answer navigation questions more reliably

  • Assistants now provide correct in-platform navigation paths when users ask where to find a feature or setting.

  • Navigation guidance was refreshed with updated context files, so answers better match the current product structure.

More reliable cross-table previews and executions

  • Cross-table execution now applies ruleset scope, filters, and limits to the primary table only, avoiding incorrect filtering of joined data.

  • Large-path cross-table previews now compute joined and failed counts more accurately, restoring correct chart values.

  • Large cross-table rules are more reliable on big tables through pushed-down counts and samples, bounded issue samples, and fail-soft handling to avoid worker out-of-memory failures.

  • Equals checks without join keys now support bidirectional set equality.

Ruleset and preview polish

  • Ruleset check tables are more compact and now show skeleton loading states while checks load.

  • The preview node message bar has been corrected.

  • Ruleset check scope row limits are now also respected in sample previews and related read paths.

  • Ruleset and company rule tag fields were updated, and tags now load before editing.

Faster overviews and exports

  • The overview experience has been cleaned up and made faster.

  • Issue export is more reliable for nested columns and now casts nested values before CSV export to avoid server errors.

  • Issue export description enrichment has been vectorized for better performance.

Connector and data access improvements

  • OData access is more robust: parallel read checks no longer serialize on the DuckDB write lock, cache survives concurrent access, and $skip remains aligned to $top.

  • The DuckDB Excel extension is now bundled into the base image for air-gapped environments.

  • Connector state handling was fixed so connectors no longer remain stuck in loading.

  • Breadcrumb data is now exposed through a hierarchy endpoint and used for improved breadcrumbs.

undefined Small improvements and bug fixes

  • Worker-side completion now saves result.parquet correctly and marks completed runs as completed after batch enrichments.

  • Batch data can be recreated on demand if temporary files were cleaned up before processing finished.

  • All rows in a failing batch are now marked as failed instead of incorrectly completed.

  • Python nodes no longer raise a type error for valid syntax in the affected case.

  • Bedrock LLM tool-result payload handling was fixed for malformed JSON content.

  • Profiling is more stable for inaccessible profiling data, UUID handling, and decimal correlation overflow cases.

  • Improvement workflow finalization no longer crashes when the improvements index is missing.

  • Rows now explicitly show 'No improvement found' instead of leaving the result empty.

  • The sticky toast that could cover the group-create button can now be dismissed.

  • AI monitoring was re-added to the Notification Center.

  • Dashboard notifications were removed to reduce noise.

undefined Breaking Changes

  • None in this release