df1 - large dataset
df2 = df1.sample(tiny_fraction)
df1 is written to disk as a parquet with snappy compression (~75GB)
df2 is written to disk as a parquet with snappy compression (~90GB)
df3 = read df2's saved parquet
# set sql context to use snappy compression
df3.partition(1).save() -> (~100MB)
df4 = read df1's saved parquet
df4.sample(tiny_fraction).partition(1).save() -> (~100MB)
The file size of the subsampled dataframe (without repartitioning) was larger than the original dataset Does anyone have any ideas as to why this could happen?