df.max()/min() on 1 column df leads to "could not broadcast input array from shape (6,) into shape (5,)" what from parquet loaded with ray

[Liquidmasl]

I am still trying to reliably save and load from parquet, but running into new problems.

It seams most of my problems are windows related, as on Linux the experience is a lot less painful.

While atempting to get managable .parquet chunks, I used a partition column.
But modin.read_parquet does not support partition columns and is defaulting to pandas, which exploses my RAM.
ray.from_parquet works though, and with .to_modin() I get a modin dataframe again, that looks fine.

but when I do

df['z'].max()
I get

could not broadcast input array from shape (6,) into shape (5,)

those 2 ints are not always the same though. They depend on the file I load, and I think on the set partitions.
But I cant seam to figure out what they mean.

I tried repartitioning but that did not help.

Any hint whats going on here?

To make my code run on windows as well, it would be great if I could use this workaround. On Linux it seams the modin load and save to parquet methods work a lot better

[Liquidmasl]

probably the same issue as here: #7383