Rebalance Mapping#

Shuffles data to produce a specified amount of Spark partitions with an approximately equal number of records. This is useful for balancing the processing load in clusters when the data is highly skewed.

Example#

mappings:
  balanced_dataset:
    kind: rebalance
    input: unbalanced_dataset
    partitions: 32

Fields#

  • kind (mandatory) (string): rebalance

  • broadcast (optional) (type: boolean) (default: false): Hint for broadcasting the result of this mapping for map-side joins.

  • cache (optional) (type: string) (default: NONE): Cache mode for the results of this mapping. Supported values are

    • NONE - Disables caching of teh results of this mapping

    • DISK_ONLY - Caches the results on disk

    • MEMORY_ONLY - Caches the results in memory. If not enough memory is available, records will be uncached.

    • MEMORY_ONLY_SER - Caches the results in memory in a serialized format. If not enough memory is available, records will be uncached.

    • MEMORY_AND_DISK - Caches the results first in memory and then spills to disk.

    • MEMORY_AND_DISK_SER - Caches the results first in memory in a serialized format and then spills to disk.

  • input (mandatory) (string): The name of the input mapping

  • partitions (mandatory) (integer): The number of output partitions

  • filter (optional) (type: string) (default: empty): An optional SQL filter expression that is applied before the rebalance operation itself.

Outputs#

  • main - the only output of the mapping

Description#