All Products
Search

This Product

    Hologres:V4.0.31 release note

    Document Center

    Hologres:V4.0.31 release note

    Last Updated:May 15, 2026

    This topic describes the improvements and bug fixes in Hologres V4.0.31.

    Improvements

    • Add replica count limit check for compute groups: A limit check has been added for the hg_table_group_load_to_warehouse path. This prevents abnormal shards when the replica count exceeds the number of workers.

    • Optimize Paimon cache policy: Adjusted the default page size of the Paimon manifest cache and disabled the split cache by default. This reduces unnecessary caching overhead.

    Bug fixes

    Severity

    Title

    Description

    Affected versions

    Workaround

    P2

    Fixed a query engine coredump triggered by an invalid string array input.

    An invalid string array input triggered an exception in the Arrow BufferBuilder::Append path, causing a query engine worker coredump.

    V2.0 and later

    Upgrade to V4.0.31 or later.

    P2

    Fixed an issue where persisting a snapshot during a query could cause MemTable data loss.

    Persisting a snapshot while querying a target table could lead to data loss in the MemTable. This version disables that persistence path.

    V4.0 and later

    P2

    Fixed an issue where the Paimon reader would hang when reading a fully deleted Avro file.

    The Paimon reader could hang when reading an Avro data file marked as fully deleted.

    V4.0 and later

    P2

    Fixed an issue where getting the row count from an empty ORC file caused high CPU utilization.

    When getting the row count from an empty ORC file with zero stripes, an infinite loop could occur. This caused the holo worker to experience high CPU utilization, which led to queries hanging or failing.

    V3.2 and later

    Stop reading the affected table.

    P2

    Fixed a null pointer dereference in SPI_getvalue for empty results.

    SPI_getvalue may cause a null pointer dereference when the query result is empty.

    V4.0 and later

    Upgrade to V4.0.31 or later.

    P2

    Fixed an issue where a transaction abort would hang after a Change Data scan error.

    During concurrent reads of multiple ORC files for Change Data, an underlying exception failed to wake other waiting readers. This caused the transaction to hang during the abort process after a refresh error occurred.

    V3.1 and later

    Terminate the worker that is stuck in the aborting state.

    P0

    Fixed a use-after-free error caused by a lifecycle defect in ForeignTableExecutorAccessor.

    A defect in shared_ptr lifecycle management in the ForeignTableExecutorAccessor::GetNext path triggers use-after-free, which causes the holo worker to coredump.

    V3.1 and later

    Upgrade to V4.0.31 or later.

    P2

    Fixed a memory leak in the Fixed FE when consuming binlogs.

    Memory was not released promptly after a connection used by the Fixed FE to consume binlogs was disconnected. Frequent reconnections led to a continuous increase in memory usage, which could eventually cause a holo worker OOM.

    V2.0 and later

    P2

    Fixed a dangling pointer issue in partition pruning caused by CachedPlan changes.

    When a prepared statement reused a CachedPlan, the partition pruning node was updated with an invalid pointer. Subsequent executions could dereference this dangling pointer, causing an FE coredump.

    V4.0 and later