What's new in Cloudera Runtime 7.1.9 SP2

Atlas

Impala lineage uses the Impala operation type

Impala lineage events include an explicit operationType field. Apache Atlas reads that value when it builds Impala process and column lineage so the lineage graph reflects the operation without relying only on parsing the full queryText. For queries that insert into a table or overwrite a table, the process segment in lineage qualified names uses INSERT or INSERT_OVERWRITE instead of QUERY or QUERY_WITH_CLAUSE.

For more information, see Impala lineage and Impala entities created in Atlas.

Faster Atlas entity detail for large relationship graphs

The first load of an entity detail page omits large embedded relationship payloads so the page opens faster. On the Relationships tab, you can switch between Table mode (a card per relationship type with counts, sort, optional deleted relationships, and paged loads as you scroll) and Graph mode. The relationship search API supports pagination, optional sorting, and more reliable counts for very large graphs. Together, these changes address cases such as Apache Ozone buckets or wide Hive tables where the detail view previously failed to show Properties, Classifications, Relationships, or Audits because the response was too large.

For more information, see Apache Atlas dashboard tour.

Atlas Java client uses Commons Configuration2 and Commons Lang3

Custom applications that link against the atlas-client-v2 JAR and call AtlasClientV2 constructors that take org.apache.commons.configuration.Configuration (Commons Configuration 1.x) must migrate to org.apache.commons.configuration2.Configuration, commons-configuration2, and recompile and redeploy against the Cloudera Runtime 7.1.9 SP2 client stack. The constructor surface is a bytecode-level signature change, so old compiled callers against new JARs can throw NoSuchMethodError at runtime if they are not rebuilt. The same change set removes commons-lang2 in favor of Apache Commons Lang3. Align imports and dependencies with Lang3 when you refresh your Atlas client integration.

This upgrade removes vulnerable Commons Configuration 1.x from the Atlas client path and addresses CVE-2025-46392.

Generic ignore patterns

Configuration snippets for both Apache Atlas server side and hook side metadata ignore patterns are introduced. You can use these limit the incoming amount of metadata notifications and messages to reduce data clutter and resource consumption.

For more information, see Using generic ignore patterns.

Impala

New TCMalloc metrics

You can now monitor memory allocation more effectively by using new numeric properties from TCMalloc on the metrics page.

The following metrics are now available:

• tcmalloc.current-total-thread-cache-bytes: The number of bytes used across all thread caches.
• tcmalloc.central-cache-free-bytes: The number of free bytes in the central cache that are assigned to size classes.
• tcmalloc.transfer-cache-free-bytes: The number of free bytes waiting to be transferred between the central cache and a thread cache.
• tcmalloc.thread-cache-free-bytes: The number of free bytes currently in thread caches.

Apache Jira: IMPALA-14874

Iceberg

Query optimization for MIN, MAX, and COUNT DISTINCT on Iceberg tables

Impala now optimizes MIN(key_column), MAX(key_column), or COUNT(DISTINCT key_column) queries on partition columns of Iceberg tables by leveraging partition-level metadata.

For Iceberg tables, partition statistics are retrieved from Iceberg metadata through the Iceberg API. This optimization applies to Iceberg v1 and v2 tables and is effective when partition transforms use identity.

Apache Jira: IMPALA-11986

Navigator Encrypt

Ozone

A new replicatedSizeOfFiles field has been added to the NSSummary object in Recon

A new replicatedSizeOfFiles field is now available in the NSSummary object within Recon. This field ensures that disk usage calculations automatically account for replication overhead, providing an accurate reflection of the actual physical storage consumed. By optimizing these backend calculations, this enhancement significantly improves the responsiveness and performance of the Recon UI, especially when managing large-scale clusters.

Added safeguards for RocksDB iterators during volume failures

Added validation safeguards to prevent active RocksDB iterators from accessing closed database instances during volume failure events. This enhancement eliminates a potential JVM crash scenario that occurred when background processes, such as container scanners, attempted to read from a natively closed database during automated volume failure handling.

Support for deleting empty QUASI_CLOSED containers

The QUASI_CLOSED containers are now deleted when all available replicas are completely empty, regardless of their lifecycle state. To protect against data loss from lagging nodes, if a previously unseen or offline replica later reconnects with a higher Block Commit Sequence ID (BCSID), the container is automatically restored to its original state. This enhancement enables clusters to safely reclaim metadata and storage space from empty and QUASI_CLOSED containers while guaranteeing that the system can still recover and re-replicate data if a newer replica emerges.

New Ozone Debug CLI for Container-to-Key Mapping

Introduced a new Ozone debug CLI tool for Container-to-Key mapping. This allows users to input a specific list of container IDs and retrieve a comprehensive mapping of all residing keys. The tool supports multiple key variants, including File System Optimized (FSO), Object Store (OBS), Open, and Multipart Upload (MPU).

New capacity-based pipeline selection policy

A new pipeline choosing policy, CapacityPipelineChoosePolicy, has been introduced to improve write performance in unevenly distributed clusters. This policy randomly selects pipelines with relatively lower utilization, making it ideal for environments where older nodes are nearing capacity and the balancer is slow to redistribute data. To enable this feature, set the hdds.scm.pipeline.choose.policy.impl property to org.apache.hadoop.hdds.scm.pipeline.choose.algorithms.CapacityPipelineChoosePolicy.

Dynamic CLI configuration for Container Balancer

The Container Balancer now supports comprehensive runtime configuration directly through the CLI, eliminating the requirement for a service restart to apply tuning adjustments. Administrators can now immediately launch the balancer with custom values specified via the CLI, including node inclusion/exclusion filters, iteration intervals (balancing.iteration.interval), data movement timeouts (move.timeout and move.replication.timeout), and network topology awareness (move.networkTopology.enable).

Ranger

Ranger KMS

Zeppelin