Skip to content

Commit e0f344a

Browse files
rustyconoverclaude
andcommitted
Cap the no-substream_id test at four subprocess workers
It fanned out to a worker per core — 48 processes on a 48-core box. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
1 parent 3def8b1 commit e0f344a

1 file changed

Lines changed: 4 additions & 2 deletions

File tree

‎tests/table_in_out/generator/test_substream_partial_sum_function.py‎

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -91,8 +91,10 @@ def test_a_client_sending_no_substream_id_still_sums_over_subprocess(self, monke
9191
batches = [
9292
pa.RecordBatch.from_pydict({"n": list(range(i * 100, (i + 1) * 100))}, schema=schema) for i in range(20)
9393
]
94-
# A plain subprocess worker whatever transport the suite runs on.
95-
with Client("vgi-fixture-worker") as client:
94+
# A plain subprocess worker whatever transport the suite runs on; four
95+
# connections are enough to share an execution, where a worker per core
96+
# would be dozens of processes.
97+
with Client("vgi-fixture-worker", worker_limit=4) as client:
9698
output = list(
9799
client.table_in_out_function(
98100
function_name="substream_partial_sum",

0 commit comments

Comments
 (0)