This topic describes major updates and bug fixes in Blink 3.6.5.

Major updates

  • The CPU utilization of DataHub source tables is reduced.
  • The partition join feature of dimension tables is improved, and data can be loaded to partitions in dimension tables for which the cache parameter is set to ALL. This version supports parallel asynchronous optimization for multiple dimension tables. To enable the partition join feature for dimension tables that support parallel asynchronous optimization, you must set the partitionedJoin parameter to true in the WITH clause.
  • The startupMode parameter is added to Log Service source tables. The startupMode parameter has the following valid values:
    • TIMESTAMP: indicates that data of a shard is consumed from the specified point of time.
    • Earliest: indicates that data of a shard is consumed from the earliest data in the shard.
    • Latest: indicates that data of a shard is consumed from the latest data in the shard.
    • Group_Offsets: indicates that data of a shard is consumed from a checkpoint that is stored in a specific consumer group.
    Note The TIMESTAMP, Earliest, Latest, and Group_Offsets values take effect only when no checkpoint exists in the state data.
  • The memory consumed to store vertex topology information is reduced for jobs that involve a large amount of data.
  • The timeFieldType parameter in Oracle source tables supports multiple time formats. This parameter has the following valid values:
    • TO_DATE: the DATE type.
    • TIMESTAMP: the TIMESTAMP type.
    • VARCHAR: the DATETIME string type.
    • NUMBER: the NUMERIC type.
    • The OracleSourceQueryCondition interface is supported and a class name is configured for the interface.

Major bug fixes

  • Fixes the bug that data errors occur in MaxCompute dimension tables if a TaskManager uses multiple threads and the partition join feature is enabled.
  • Fixes the bug that an error message is reported when the DDL field type of the sink node is different from the type of the inserted data.
  • Fixes the bug that Realtime Compute for Apache Flink does not consume data in Log Service source tables.