Repository navigation
[Python][C++] Crash when performing join on a table larger than INT32_MAX bytes #36879
Closed as duplicate
Description
Describe the bug, including details regarding any error messages, version, and platform.
pyarrow version: 12.0.1
Minimal Repro:
import pyarrow as pa
ROW_COUNT = 15_000
query = pa.Table.from_pydict({
"key": [bytes.fromhex('ABC123'), bytes.fromhex('DEF431')]
})
t = pa.Table.from_pydict(
{
"blob_2": [x.to_bytes(4, 'little') * 25_000 for x in range(ROW_COUNT)],
"blob_1": [x.to_bytes(4, 'little') * 25_000 for x in range(ROW_COUNT)],
"key": [x.to_bytes(4, 'little') for x in range(ROW_COUNT)]
},
schema=pa.schema(
[
("blob_2", pa.large_binary()),
("blob_1", pa.large_binary()),
("key", pa.binary()),
]
)
)
print(f"Table size: {t.nbytes} bytes (bigger than INT_MAX: {t.nbytes > (1 << 31)})")
for x in range(10):
print(f"Attempt {x}")
_ = query.join(t, 'key', join_type="inner")Executing this results in a segfault:
terminate called after throwing an instance of 'std::length_error'
what(): vector::_M_default_append
Note that flipping the join (e.g. _ = t.join(q, 'key', join_type="inner")) no longer segfaults.
Component(s)
C++
Activity
[-]Crash when performing `join` on a table larger than INT32_MAX bytes[/-][+][Python][C++] Crash when performing `join` on a table larger than INT32_MAX bytes[/+]idailylife commentedon Jan 23, 2024
Is this (#33151) related?
zanmato1984 commentedon Oct 22, 2026
This is the same Basic Hash Join /
RowEncoderoverflow as #54885, but in the build-side payload rather than the keys.The
large_binarycolumns select Basic Join. Their combined encoded size exceeds its signed 32-bit offset range, even though each column alone is below 2 GiB. Reversing the inputs makes the small table the build side and avoids that accumulation.Closed as a duplicate of #54885; the underlying bug remains unresolved.
large_stringkeys near 2gb #54885