Error when Z Ordering a larger dataset

[MrPowers]

Environment

Delta-rs version: 0.10.0

Binding: Python

Environment:

• Cloud provider: Localhost
• OS: Macbook
• Other: Macbook M1 with 64 GB of RAM

---

Bug

What happened: Z Order command worked on 5 GB h2o groupby dataset (1e8), but errors out of 50 GB dataset (1e9)

What you expected to happen: I expected the Z Ordering to work

How to reproduce it: This notebook shows the computations working well on the 1e8 dataset, but erroring out on the 1e9 dataset.

More details: I'm Z Ordering on a single column. Here's the error message:

thread 'tokio-runtime-worker' panicked at 'overflow', /Users/runner/.cargo/registry/src/github.com-1ecc6299db9ec823/arrow-select-39.0.0/src/interleave.rs:172:56
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
---------------------------------------------------------------------------
DeltaError                                Traceback (most recent call last)
File <timed eval>:1

File ~/opt/miniconda3/envs/deltalake-0100/lib/python3.9/site-packages/deltalake/table.py:697, in TableOptimizer.z_order(self, columns, partition_filters, target_size, max_concurrent_tasks)
    675 def z_order(
    676     self,
    677     columns: Iterable[str],
   (...)
    680     max_concurrent_tasks: Optional[int] = None,
    681 ) -> Dict[str, Any]:
    682     """
    683     Reorders the data using a Z-order curve to improve data skipping.
    684 
   (...)
    695     :return: the metrics from optimize
    696     """
--> 697     metrics = self.table._table.z_order_optimize(
    698         list(columns), partition_filters, target_size, max_concurrent_tasks
    699     )
    700     self.table.update_incremental()
    701     return json.loads(metrics)

DeltaError: Generic error: task 334 panicked

[wjones127]

Got it. That's likely from this line:

https://github.com/apache/arrow-rs/blob/700bd334ad8be53455f5dd80023b6c8c237559a7/arrow-select/src/interleave.rs#L172

Which indicates that we have a binary column that is too large for StringArray. Maybe we can use large types for this? Or else interleave the results in chunks.

[MrPowers]

Here's a few rows of the data:

	id1	id2	id3	id4	id5	id6	v1	v2	v3
0	id016	id046	id0000109363	88	13	146094	4	6	18.837686
1	id039	id087	id0000466766	14	30	111330	4	14	46.797328
2	id047	id098	id0000307804	85	23	187639	3	5	47.577311
3	id043	id017	id0000344864	87	76	256509	2	5	80.462924
4	id054	id027	id0000433679	99	67	32736	1	7	15.796662

Here are the dtypes:

id1     object
id2     object
id3     object
id4      int64
id5      int64
id6      int64
v1       int64
v2       int64
v3     float64
dtype: object

Here's the code I ran: dt.optimize.z_order(["id1"])

Should have included these details in the initial bug report.