maxframe.liteframe.LiteFrame.rebalance#

LiteFrame.rebalance(axis=0, factor=None, num_partitions=None, rough=False)#

Make data more balanced across entire cluster.

LiteFrame is only ever chunked along the row axis (axis 0); axis is accepted for API parity with DataFrame.rebalance but should remain 0.

Parameters:

  • axis (int) – The axis to rebalance.

  • factor (float) – Specified so the number of chunks after balance along the selected axis per plane is chunk_shape[axis] * factor.

  • num_partitions (int) – Specified so the number of chunks are at most num_partitions.

  • rough (bool, default False) –

    Hint to rebalance without first resolving unknown row shapes.

    When True, rebalance may partition the row axis using a per-chunk size estimate (rough_size_ratio) instead of waiting for real (resolved) shapes. This only has an effect when the row shape is unknown and the source chunks carry such an estimate; currently that estimate is produced only by SIZE-mode read_odps_table reads. In every other case – known shapes, or an unknown row axis with no estimate – rough=True silently falls back to the normal path (resolving real shapes then rechunking). The estimate is approximate, so the output is balanced by estimated share, not exactly equal in size.

Returns:

Result of LiteFrame after rebalanced.

Return type:

LiteFrame