Skip to content

Derive per-chunk memory reservation from scan statistics #22846

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
rapids-bot merged 3 commits into from
Jun 11, 2026
Merged

Derive per-chunk memory reservation from scan statistics #22846

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
66eb035
Use statistics-based chunk size estimate for memory reservation in sc…
Matt711 Jun 10, 2026
af18f48
Add it to IOPartitionPlan
Matt711 Jun 10, 2026
a14bcee
Merge branch 'main' into imp/polars/better-chunk-size-estimate
Matt711 Jun 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion python/cudf_polars/cudf_polars/streaming/actor_graph/io.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -606,7 +606,9 @@ def _(
rec.state["ir_context"],
ch_out,
num_producers=num_producers,
estimated_chunk_bytes=executor.target_partition_size,
estimated_chunk_bytes=(
plan.estimated_chunk_bytes or executor.target_partition_size
),
)
]
return nodes, channels
Expand Down
13 changes: 11 additions & 2 deletions python/cudf_polars/cudf_polars/streaming/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -166,12 +166,21 @@ class IOPartitionPlan:
- SINGLE_READ: `factor` is the total number of files.
"""

__slots__ = ("factor", "flavor")
__slots__ = ("estimated_chunk_bytes", "factor", "flavor")
factor: int
flavor: IOPartitionFlavor
estimated_chunk_bytes: int | None
"""Estimated decompressed bytes per chunk (Scan nodes only)."""

def __init__(self, factor: int, flavor: IOPartitionFlavor) -> None:
def __init__(
self,
factor: int,
flavor: IOPartitionFlavor,
*,
estimated_chunk_bytes: int | None = None,
) -> None:
if flavor == IOPartitionFlavor.SINGLE_FILE and factor != 1: # pragma: no cover
raise ValueError(f"Expected factor == 1 for {flavor}, got: {factor}")
self.factor = factor
self.flavor = flavor
self.estimated_chunk_bytes = estimated_chunk_bytes
8 changes: 6 additions & 2 deletions python/cudf_polars/cudf_polars/streaming/io.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,15 +90,19 @@ def scan_partition_plan(
if (file_size := sum(column_sizes)) > 0:
if file_size > blocksize:
# Split large files
factor = math.ceil(file_size / blocksize)
return IOPartitionPlan(
math.ceil(file_size / blocksize),
factor,
IOPartitionFlavor.SPLIT_FILES,
estimated_chunk_bytes=file_size // factor,
)
else:
# Fuse small files
factor = max(blocksize // int(file_size), 1)
return IOPartitionPlan(
max(blocksize // int(file_size), 1),
factor,
IOPartitionFlavor.FUSED_FILES,
estimated_chunk_bytes=file_size * factor,
)

# TODO: Use file sizes for csv and json
Expand Down
Loading