February 27, 2026 - Hotfix

Review the fixed issues in this release of the Cloudera Data Warehouse service on cloud.

DWX-22470: Fluentd out-of-memory errors when shipping audit logs from the coordinator pod

Previously, in Cloudera Data Warehouse, the audit log Fluentd sidecar in the coordinator pod could encounter out-of-memory (OOM) errors. This occurred because the Fluentd configuration used a chunk_limit_size of 256 MB, which matched the container memory limit of 256 MB. Under heavy audit log loads, buffering and flushing data caused Fluentd’s memory usage to exceed the container limit, leading to OOM kills and interruptions in audit log shipping.

This issue is now resolved by optimizing the Fluentd configuration to reduce memory usage and improve reliability under heavy audit log loads.

DWX-22700, DWX-22330: Multiple configuration updates during a single update operation

Previously, the Cloudera Data Warehouse user interface had trouble saving multiple configuration settings for a Virtual Warehouse when multiple values were updated at the same time using a single update operation. Some keys would get excluded, causing incomplete or inconsistent configurations.

This issue is resolved by ensuring that all modified configuration flags are saved correctly in a single update operation.

DWX-22347: Database Catalog initialization failures due to short informer cache sync timeout in Cloudera Data Warehouse

Previously, in Cloudera Data Warehouse environments, Database Catalog initialization could fail with errors such as:

error getting metastore-sys-init-job job for '<catalog-name>': timed out waiting for caches to sync Error Code : 9999

This issue occurred in private AKS/EKS clusters and other environments with higher control-plane or CCM round-trip latency. The default informer cache sync timeout of 10 seconds was insufficient for the Kubernetes informers used by Cloudera Data Warehouse control pane components. As a result, cache synchronization could not complete within the allotted time, causing initialization jobs to fail and leaving the Database Catalog in an error state.

This issue is now resolved by increasing the informer cache sync timeout.

DWX-22235: Impala profile upload fails on Azure due to missing home directory for diagnostic tool user

Previously, in Cloudera Data Warehouse on Azure, Impala profile uploads failed when using the diagnostic tools image. This was caused by a configuration issue where the Azure CLI, used during the upload process, required a writable directory for its configuration but none was available.

This issue is now resolved.

DWX-22253: Cluster autoscaler scale-down-delay-after-add causing unstable scaling behavior in cloud deployments

Previously, in Cloudera Data Warehouse on AWS and Azure, the scale-down-delay-after-add parameter was set to 1 minute, causing the cluster autoscaler to scale down too quickly after scaling up. This led to a continuous loop of scaling up and down, resulting in unstable cluster capacity and inefficient resource usage.

This issue has been resolved by increasing the scale-down-delay-after-add parameter to 5 minutes, allowing the system to stabilize after scaling up and improving cluster stability.

DWX-22606: Insufficient ephemeral storage for Impala coordinator heap dumps and logs

Previously, Impala coordinator pods in Cloudera Data Warehouse requested only 24 GB of ephemeral storage, which was insufficient to handle heap dumps, query profiles, and logs. This could lead to disk pressure or pod eviction on coordinator nodes.

This issue is now resolved by increasing the requested ephemeral storage for Impala coordinator pods, ensuring stable operation without disk-related issues.

CDPD-97080: Memory leak in global admissiond for cancelled queued queries

Previously, a memory leak occurred in the global admissiond when queries in the admission queue were cancelled due to backpressure. The system identified the cancellation but did not remove the query from the admission state map.

To resolve this memory leak, this fix introduces an asynchronous cleanup mechanism. Cancelled queued queries are now added to a queue for a background process to safely clear them from the admission state map.

Apache Jira: IMPALA-14605

CDPD-45130: Truncating excessive length queries to prevent database indexing errors

Previously, large or complex SQL statements, such as lengthy INSERT queries, were indexed by the Query Processor. This resulted in increased load times for the Job Browser. You can now configure query truncation by using the hue.query-processor.query.max-length property in the Query Processor configuration under dasConf section. By default, no truncation is performed to ensure backward compatibility.

This release of the Cloudera Data Warehouse service on Cloudera on cloud has the following behavior changes: