Ataccama 17.1.0 Release Notes

Reverting a rule to a previous version correctly restores its expressions.

Test Rule returns correct results for rules with multiple input attributes. Previously, input values could be mapped to the wrong attributes after adding, removing, or renaming the attribute.

Date parameters in parameterized rules display the value you entered when testing or applying the rule, instead of a date one day earlier.

Rule suggestions list only single-input rules, preventing the monitoring project errors that occurred when multi-attribute suggestions were accepted.

Rules with missing mappings between rule inputs and catalog item attributes can no longer be published, preventing Missing input attributes for validation errors that were difficult to resolve in monitoring projects.

Filtering by attributes and applied rules on the catalog item Data Quality tab returns all matching results.

Rule Debug displays generated inputs for DateTime attributes.

When defining a rule condition, you can enter a value as an input.

Users with View metadata access level can use Test Rule on DQ rules again, without requiring Editing access level.

When you update a single rule and republish its catalog item in a monitoring project, the View changes screen now lists only that rule, instead of showing all rules of the catalog item as changed.

On the monitoring project Report tab, switching from a processing that uses attribute filters to one that doesn’t clears any applied filters, so the report and charts display correctly.

All results are displayed in the monitoring project report for every filter combination, including projects with a very large number of filter combinations.

Specific notifications in monitoring projects are no longer sent when data quality evaluation does not complete. Previously, a "Project run failed" email was incorrectly sent to recipients.

The Filter Attributes and Rules search bar on the monitoring project Configuration & Results tab accepts special characters.

Selecting an unsupported component type in an Embedded component step produces a non-blocking validation warning instead of failing the plan. Previously, selecting a component that contained a format-changing step, such as a Group aggregator, could fail the transformation or remediation plan, an issue that surfaced after upgrading to version 17.0.0 due to stricter validation. Removing a column that is not present in the input is now ignored rather than failing the plan.

Data quality rules implemented as components now produce correct results in monitoring projects when the component plan in ONE Desktop includes a record-reordering step such as Sort, Unique, or an aggregation. Previously, such a rule could corrupt the results of the rules evaluated after it.

When a rule implementation changes after it has been published, the monitoring project Configuration tab flags the affected DQ check as outdated.

The monitoring project Export tab displays correctly for projects that use post-processing on a virtual catalog item, instead of the tab layout breaking.

On the monitoring project Report tab, the overall data quality of a custom report section counts only the dimensions that contribute to overall quality.

When importing monitoring project configuration with an applied data slice to another monitoring project, the data slice reference is no longer carried over if a different catalog item is selected. This prevents monitoring project run failures with the "Data slice was not found" reason.

Monitoring project notifications no longer display an incorrect dash before DQ result percentages.

Monitoring projects whose catalog item is based on a local file no longer fail validation when their post-processing plan uses Spark processing (for example, a SparkWriter step). Validation still correctly fails when the plan references a data source that does not exist, while other validation issues are reported in the logs without stopping the run.

Validating a transformation plan shows all error messages for invalid ONE expressions, instead of only the first one.

Transformation plan validation and DQ rule input mapping now accept implicit type widening (INTEGER→LONG, INTEGER→FLOAT, LONG→FLOAT, DATE→DATETIME), so valid widening assignments no longer fail with errors like The type of the attribute 'X' must be LONG but is INTEGER after upgrading.

Transformation plans with a Database writer step let you select both the catalog and the schema when creating catalog items on Unity Catalog-enabled Databricks sources.

DQ firewall configuration no longer displays names of rules that the current user does not have access to. Previously, rule names were visible and the configuration banner misleadingly reported them as outdated or damaged.

The Create DQ Firewall option is not available for monitoring projects that have never been published.

Users with View metadata access level can debug rules in a DQ firewall and view the DQ firewall Documentation tab.

Restoring a previous version of a DQ firewall from the History tab works as expected.

Calling a DQ firewall no longer fails with a SQL error when one of its rules has no parameter values.

The DQ firewall Documentation tab displays correctly even when attribute names, rule names, or parameter values contain special characters.

Reconciliation projects no longer fail to publish with a validation error after a referenced catalog item is re-profiled or its structure changes.

Catalog items that have been selected appear at the top of the catalog item list on the Select sources screen, making them easier to find in large sources.

Tabs for filtering catalog items on the Mapping screen are clearly indicated as clickable.

Long attribute names no longer cause layout issues on the reconciliation project Results tab.

The anomaly detection state can no longer be edited from the detail page of a specific profiling run.

Anomalies filter on the Catalog Items screen returns correct results.

In Profile Inspector, the Display only anomalous items option now correctly filters the list to show only anomalous attributes.

ONE Data tables open correctly without TypeError error notifications.

When importing data to a ONE Data table from a catalog item, the selected stewardship is applied to the new table and permissions are no longer duplicated.

Cancelling an overwrite or upsert import to a ONE Data table no longer deletes the existing table. The cancelled import also preserves attribute definitions, the original column and row order, and the state of DQ checks (enabled or not).

Lookup-based detection rules are applied correctly during profiling, even when the rule expression uses a variable with the same name as a column in the profiled data.

Lookup item files are visible immediately after a review is approved, without requiring a page reload.

Mandatory term properties of type referenced object array and referenced object with the valueList:property trait now show the required field indicator (red asterisk) on term editing forms. Previously, the indicator was missing and the requirement was only surfaced as a validation error on publish.

When you select Add Term in the attribute detail sidebar, the dropdown with terms no longer incorrectly shows suggested terms as already applied; selecting a suggested term adds it to the attribute.

Redrawing the relationship graph for a term no longer creates duplicate relationship type entries in Global Settings > Graph Visualization.

Deleting a catalog item or attribute no longer removes valid term suggestions on similar attributes by incorrectly treating the deletion as negative feedback for the AI model.

The Stewardship widget on glossary term pages now reliably loads all governance roles on first page load. Previously, the widget would show only basic stewardship information in certain cases until the page was refreshed.

Rich text values written to business terms via the ONE Metadata Writer step in ONE Desktop display correctly in the ONE web application.

You can find and mention terms in comments.

Filtering by keyword values works correctly when the search query contains uppercase words such as AND, OR, or NOT.

Catalog search ranks catalog items whose names contain underscores correctly even when the search query omits them (for example, searching HR dept list matches HR_DEPT_LIST).

Searching by an indexed property that was inherited from a parent entity returns results, and the matching value is highlighted. For example, after marking the abbreviation property as searchable, you can find terms by their abbreviation and see the match highlighted in the results.

Search index event processing no longer accumulates partly failed events, which could lead to stale or incomplete search results.

Reindexing no longer leaves some catalog items missing from search results.

When you set numeric range filters on the Catalog items screen to a range that returns no results, you can now clear that individual filter, instead of having to reset all filters at once.

When expanding nested search results (such as attributes under catalog items), only the newly loaded items show a loading spinner, instead of all listed items briefly flashing spinners.

Long term names no longer overflow in the global search left sidebar.

Moving tasks on the Tasks Overview screen works as expected without freezing or requiring a page refresh.

Individual review tasks are automatically cleared once their changes are resolved through a combined review request, instead of lingering as pending tasks with no changes to review.

The Tasks Overview and home pages load correctly for users who have both the admin role and a custom governance role assigned. Previously, the page failed to load with an error for these users.

Sweep documentation flow no longer fails on Databricks data sources.

The Sweep documentation flow screen in Notification Center correctly states that items have been automatically deleted for scheduled runs, instead of manually deleted.

On the Sweep documentation flow Delete obsolete catalog objects screen, the Delete Selected option is inactive after a deletion, and it does not overlap the search bar when search returns no results.

When a scheduled sweep flow is started with missing or invalid configuration, an error notification with the specific reason is displayed, instead of failing silently.

Fixed an issue where scheduled Sweep documentation flow deleted obsolete catalog items only when an item had not been modified and had not been profiled within the threshold period. An item is now deleted when it has not been modified or has not been profiled within the threshold period.

OneLake data source connections now support Azure AD Workload Identity authentication. See OneLake.

Amazon Redshift connections support using AWS assumed-role (IAM) authentication.

Local filesystems now support mounting Parquet files.

Adding credentials to a connection shows a clear error message instead of a blank page when the DPE is disconnected or the connection type is not enabled in the DPE.

Editing credentials from the connection screen shows all fields, such as username and password, on first open.

Username and password credentials migrated from 15.4.x can be edited in the web application.

Databricks Spark pushdown jobs now work correctly when ADLS credentials are stored in Azure Key Vault.

Metadata import from ADLS Gen2 connections no longer fails when the connection uses path conditions and ACL settings.

Testing a Secret Management Service connection correctly reports failure for invalid credentials.

Advanced encryption is now available for secrets passed from Data Processing Module to hybrid Data Processing Engine; standard encryption is considered obsolete. We recommend upgrading at your earliest convenience, see Advanced encryption between DPM and DPE.

Fixed an issue where jobs could intermittently get stuck in the DPM queue and never be submitted to DPE, with the Detected old submitting job message appearing in the DPM Admin Console.

Job submission no longer fails with an "Engine not available" error when DPE is temporarily reconnecting.

Job submission no longer fails with JobInvalidArgumentException when a job with the same ID already exists on DPE.

DPM no longer throws a NullPointerException while handling a job that could not be submitted, for example when a hybrid DPE disconnects.

Improved hybrid DPE stability during high memory usage.

Fixed an issue where DPE could run out of memory and become unavailable with a There is no available data processing engine error after browsing several different sources.

In Data Observability, DQ evaluation no longer fails with DataSourceClientConfig not found error when the source connection is correctly configured.

DPE correctly creates the job folder when a custom TEMP_ROOT is specified in the configuration, preventing temporary file errors during profiling.

Profiling jobs with regex-based detection rules no longer get stuck on columns containing large text values. The maximum input length for regex evaluation can be limited using the MAX_REGEX_INPUT_LENGTH DPE property.

SAP metadata imports no longer fail with Columns must have unique orders when source tables have attributes with duplicate order values.

Snowflake DECFLOAT columns are now imported as a floating-point type instead of LONG, so profiling works correctly.

Profiling no longer fails on Trino columns of type TIMESTAMP(6) WITH TIME ZONE.

S3 sweep and profiling jobs no longer fail entirely when a password-protected file is encountered; such files are skipped and the rest of the job completes.

MMM jobs that never receive events from their DPM jobs fail with a clear timeout instead of remaining stuck in RUNNING indefinitely.

DPE now creates temporary directories on Hadoop worker nodes with more restrictive permissions.

Large profiling plans are stored via object storage instead of the database, and job data is cleaned up after completion, reducing database load and storage usage.

Faster DPM resource allocation queries, reducing database load in high-throughput environments.

Monitoring projects on catalog items from a local file system with Spark post-processing plans no longer incorrectly fail model validation.

Fixed an issue where long-running JDBC operations could intermittently fail with a NoClassDefFoundError, typically on Azure MS SQL connections using service principal authentication.

DPM now removes long-disconnected engines and their job records from the database, preventing the DPM console from slowing down as the engine table grows.

Databricks jobs on shared clusters no longer fail with "Plan file not found" errors caused by concurrent sessions sharing the same workspace path.

Fixed an issue where DPE slots were incorrectly reported as fully utilized, causing queued jobs to stall indefinitely.

Set the default logging level for hybrid DPEs to DEBUG.

Fixed a memory leak during profiling jobs that could lead to OutOfMemory errors on large environments.

IOMETE pushdown processing now uses the ArrowFlight JDBC driver instead of the deprecated Hive JDBC driver.

IOMETE pushdown processing supports exporting invalid records, however, the Export tab in monitoring projects doesn’t link directly to the export table. Instead, it shows the table identifier. See View exported records.

Attributes no longer show unrelated DQ dimensions after pushdown DQ evaluation.

When editing rule configuration, conditions with functions or expressions not supported by Snowflake pushdown are correctly marked. Rules with such conditions are excluded from evaluation in Snowflake pushdown.

Parameterized rules using float comparisons can be applied to data sources with pushdown enabled.

Data quality results for catalog items evaluated with pushdown correctly include all result data: data slices, partitions, term aggregations, post-processing results, and export results.

When export from a pushdown data source is not supported, the warning message is displayed in a tooltip on the monitoring project Export tab. The Snowflake-specific message for BigQuery sources has been removed.

Pushdown invalid record export includes additional columns and record identifiers even when invalid samples are turned off.

Invalid records exported with Databricks pushdown populate the RECORD_ID column from the configured primary key.

Snowflake pushdown profiling no longer fails with a SQL compilation error when frequency analysis is turned off.

On the Invalid Samples page, the catalog item name aligns correctly with the icon, even for catalog items with long names.

Source detail screen loads correctly when reopening a previously viewed source.

Job duration is displayed correctly in the web application, even for jobs running longer than 24 hours.

The source and location Overview tabs no longer display the Term occurrences section that showed internal IDs and produced an error when opened.

Editing a term or changing its stewardship no longer occasionally shows a TypeError instead of the expected notification.

The web application no longer displays irrelevant errors as recurring popups (such as those from the analytics library).

Catalog item attribute Profile & DQ insights tab opens the first time, instead of showing a render error until you open the attribute in a new tab.

In the thumbnail view of report catalog items, items marked for deletion show a For delete label on the tile, instead of only after opening the item in the sidebar.

Date filter inputs no longer change the entered date to an incorrect value in the Ataccama Cloud Portal.

Faster loading of user and role selection dropdowns in project notifications, observability settings, and DQ firewall configuration, especially in environments with many users.

Catalog item Data Quality tab loads significantly faster, especially on catalog items with a large number of attributes.

DQ evaluation on Spark no longer fails while generating the invalid record samples for DQ results.

Fixed long-running database transactions caused by slow OpenSearch responses, which could block other operations.

Turned off resource-intensive PostgreSQL custom metrics that could significantly slow down database performance.

In Ataccama Cloud environments, the PostgreSQL monitoring exporter correctly loads custom metric queries from the configuration, ensuring all configured database metrics are collected as expected.

After a database cluster failover, new and updated term suggestions continue to appear in the application automatically. Previously, term suggestions would stop updating until the service was manually restarted.

Recent changes to catalog items, terms, and other assets are reliably reflected in search, instead of search results going stale when the indexing service stalled on startup in large environments.

The application no longer becomes intermittently unavailable when catalog search indexing puts the search engine under heavy write load. Calls to the search engine now time out and recover instead of blocking indefinitely.

AI Term Suggestions use database space more efficiently, preventing database CPU overload in large environments.

Fixed long-running database transactions during data observability runs that could slow down other operations.

A change event that cannot be processed no longer blocks metadata updates across the platform. The affected event is skipped and logged so that subsequent changes continue to propagate.

Fixed 'user is null' messages flooding logs on some environments.

The Add to Slack option shows a clear error when Slack is not configured at the server level, instead of incorrectly indicating success.

The Data slices dialog no longer appears during SQL catalog item creation, where it served no purpose and could prevent the catalog item from being created.

Validation component templates include DQD_REC_ID on integration inputs, enabling DQ evaluation with complex component rule plans.

Data quality results exported to Atlan are rounded correctly, matching the values shown in ONE.

Fixed an upgrade issue where certain roles could end up with duplicate access levels, preventing modifications.

In Ataccama Cloud environments, JAVA_OPTS are correctly configured for DPE when custom trust certificates are enabled.

In Ataccama Cloud environments, new users are correctly prompted to set a password on first access instead of being redirected to the standard login page.

Keycloak no longer fails to start after upgrading from version 25 to 26 in self-managed deployments. This was caused by a missing client scope during realm re-initialization.

Upgraded third-party dependencies across multiple services (including Task Service, Workflow Service, PostgreSQL exporter, term suggestions, AI matching, and anomaly detection) for enhanced security and stability.

The Identify MDM API now evaluates key rules across all relationship nodes in the request, instead of stopping at the first node and missing matches on subsequent ones.

Fixed incorrect matching results when using multiple matching rules with filters in a key rule.

Fixed incorrect merge proposal creation when Rematch If Changed is enabled and a previously merged master is re-evaluated.

Merge proposal resolution no longer fails with a "Draft doesn’t exist" error when previous drafts in the same task were discarded.

Data load no longer fails with IllegalStateException: Queue full during party matching when a large number of matched pairs are produced.

Usernames are correctly displayed instead of appearing hashed.

The task properties section is again visible on the task details screen.

The Last run column on the Load Operations tab correctly updates when load operations are executed via the Run MDM Multi Load workflow task. Average duration is skipped as load durations within a multi-load are not directly comparable.

Selecting Clone config repository after a failed configuration apply correctly resets the configuration status to Idle.

The "MDM not responding" error is cleared once API calls succeed again, instead of remaining visible while the server is running.

The legacy Admin Center now reflects the latest server startup time after a stop/start performed in the new Admin Center.

Fixed database connection leaks that could occur after cancelling tasks repeatedly.

Restored the Remove option for crashed streaming event handlers in the Event Publishing view, where it was previously missing.

Crashed streaming event handler publishing shows the publishing error in the publishers list, instead of failing with no visible message.

MDM streaming event handler now publishes updates for every record change. Previously, only the first version of a record was written to the streaming output.

The streaming event handler correctly captures events only from configured entities, preventing out-of-memory (OOM) crash loops.

The MDM Server no longer runs out of memory when the streaming event handler encounters parallel write conflicts during large batch operations.

Action buttons in the MDM Web Application column properties dialog are no longer partially cut off when scrolling on some browsers.

Added confirmation popups after Delete task and Reassign task operations, and improved the notification messages shown after Resolve and discard and Resolve and publish actions.

When using the COMBO lookup type, it is now easier to distinguish between active and inactive lookup values in the MDM Web Application

Faster loading of large entity details in the MDM Web Application, reducing the likelihood of gateway timeouts when opening records with very large numbers of related rows.

When searching for records with empty values via the REST API, the filter is no longer ignored.

Snowflake JDBC connections in MDM Server no longer fail with a driver initialization error.

Kafka stream consumers no longer fail with Kafka stream source is not producing values of type string when receiving null (tombstone) messages.

MDM stream consumers no longer occasionally fail on startup with an "Authorizator is null" error due to a security initialization race condition.

Fixed intermittent Duplicate ID encountered failures during parallel batch processing when two transactions write to the same instance record and one of them also inserts a child record linked to the parent via CopyColumns. This is more likely with multiple parent-child relationships referencing the same entity.

MDM Server correctly logs online (SOAP/HTTP) service calls through the service-log logger when HTTP logging is enabled.

MDM Server stops during startup when the license check fails, instead of continuing initialization.

MDM runtime.properties files are loaded correctly on the first MDM Server startup in Ataccama Cloud environments.

MDM health checks now reflect Keycloak availability in the readiness probe, making authentication-provider issues easier to detect.

Added Google Cloud Storage (GCS) as a supported data source for MDM Server, enabled through additional drivers.

Added MDM persistence and NME configuration parameters as Prometheus metrics for monitoring.

Master records are no longer incorrectly marked as inactive after running a migration load.

MDM cleansing tasks no longer hang indefinitely under rare race conditions during plan reloads, which previously left tasks stuck and required a restart to recover.

The MDM REST API now returns preloaded related entities consistently across both instance endpoint styles (/instance/party/{id} and /instance/party?id=).

Restored the default value of nme.vldb.commitSize to 10000, re-enabling intermediate commits during the MDM committing phase. Previously, the default of -1 ran the entire phase as a single transaction.