Yicong Huang created SPARK-59755:
------------------------------------
Summary: Add ASV benchmarks for Arrow incremental aggregate and
window UDF eval types
Key: SPARK-59755
URL: https://issues.apache.org/jira/browse/SPARK-59755
Project: Spark
Issue Type: Sub-task
Components: PySpark
Affects Versions: 5.0.0
Reporter: Yicong Huang
Add ASV microbenchmarks to python/benchmarks/bench_eval_type.py for the three
Arrow-native incremental Aggregator eval types, which have no worker
microbenchmark today: SQL_GROUPED_AGG_ARROW_INCREMENTAL_PARTIAL_UDF (255),
SQL_GROUPED_AGG_ARROW_INCREMENTAL_FINAL_UDF (256) and
SQL_WINDOW_AGG_ARROW_INCREMENTAL_UDF (257). Each gets a _*BenchMixin driving
worker_main over the mock wire protocol plus Time/Peakmem subclasses, following
the existing pattern; the UDF payload is a cloudpickled incremental Aggregator
(zero/reduce/merge/finish + bufferSchema). Scenarios: PARTIAL few keys vs many
keys hitting the maxRecordsPerBatch cap flush; FINAL many groups of struct
partial buffers including null buffer rows; WINDOW unbounded, running
(running-buffer extension) and sliding (refold) frames. Benchmark-only, no
product change; enables before/after measurement of the SPARK-59727 migration.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
