Pushdown Filter Type BLOOM_FILTER is not currently supported in PyArrow Scans

[jakkes]

What happens?

Hi,

I am running into this error message and am unsure what to do.

_duckdb.NotImplementedException: Not implemented Error: Pushdown Filter Type BLOOM_FILTER is not currently supported in PyArrow Scans

It seems I can either disable pushdown filtering through joins, or materialize the query before doing my join. I would rather not do either, since it is a lot of data that I am processing.

Reading the source files with a non-pyarrow reader is unfortunately not an option at this time.

To Reproduce

Unfortunately, I don't really have a reproducer, but I think the maintainers will know what the issue is anyways from the error message. It seems to me that Pyarrow do support bloom filter, but the pushdown is just not implemented?

OS:

Ubuntu 22

DuckDB Package Version:

1.5.2

Python Version:

3.11

Full Name:

Jakob Stigenberg

Affiliation:

Qubos Systematic

What is the latest build you tested with? If possible, we recommend testing with the latest nightly build.

I have tested with a stable release

Did you include all relevant data sets for reproducing the issue?

No - I cannot share the data sets because they are confidential

Did you include all code required to reproduce the issue?

• Yes, I have

Did you include all relevant configuration to reproduce the issue?

• Yes, I have

[evertlammerts]

Hey @jakkes, thanks for the report!

There does seem to be a bug in pyarrow_filter_pushdown.cpp, we don't support bloom filter pushdown in pyarrow but instead of just refusing to push down, we throw an exception.

But I can't seem to trigger this path. Bloom filter table filters are only generated when CanUseBloomFilter is true, which depends on cardinality stats being available. But we d not expose those for arrow right now. Me and Claude tried hard to hit the path but for now, no dice.

Please share some more details on what you're doing:

1. What SQL are you executing (number of joins, whether there are CTEs / GROUP BYs / nested subqueries upstream of the failing join)? More details are better.
2. How do you load the pyarrow data into DuckDB? con.register(table), con.register(dataset), con.from_arrow(...), a pyarrow.dataset.Scanner, a RecordBatchReader, or reading parquet via pyarrow.dataset.dataset(...).
3. What's the approximate row count on each side of the failing join (orders of magnitude is fine)?
4. Is the pyarrow source is backed by a parquet file?