With the 2-pass strategy, we can write arbitrary row group sizes while using a fixed amount of memory, with probably 100-200 MiB of overhead for the parquet file processing, depending on how large the metadata is for the scratch file. without the 2 pass strategy, the amount of memory is proportional to the size of the row group.
3 comments
[ 3.3 ms ] story [ 18.7 ms ] thread