Closed antiguru closed 8 months ago
With https://github.com/TimelyDataflow/timely-dataflow/pull/530, we could shrink the Columnation-based variant, too.
Not needed anymore with ord_neu
. In addition, this only releases virtual memory, so it might need more memory than before because it needs to copy into a new allocation.
This eagerly releases memory after retracting large amount of data, without having to wait for another operation to happen.