From d1512e2166ed15005f168577e9d9609793fc6596 Mon Sep 17 00:00:00 2001 From: Adrian Vogelsgesang Date: Sat, 15 Aug 2026 20:25:57 +0000 Subject: [PATCH] Disable sampling for Hyper Parquet benchmarks Disable creation of Parquet samples so the benchmark measures query execution rather than optimizer sampling. --- hyper-parquet-partitioned/start | 7 ++++++- hyper-parquet-single/start | 7 ++++++- 2 files changed, 12 insertions(+), 2 deletions(-) diff --git a/hyper-parquet-partitioned/start b/hyper-parquet-partitioned/start index f13ebd21d7..f6a0ef1f14 100755 --- a/hyper-parquet-partitioned/start +++ b/hyper-parquet-partitioned/start @@ -40,7 +40,12 @@ signal.signal(signal.SIGINT, _terminate) with HyperProcess( telemetry=Telemetry.DO_NOT_SEND_USAGE_DATA_TO_TABLEAU, - parameters={"experimental_persisted_external_tables": "on"}, + parameters={ + "experimental_persisted_external_tables": "on", + # External-table samples are useful for complex queries but meaningless + # for ClickBench's rather simple queries + "external_sampling": "false", + }, ) as hyper: # Keep-alive: hold hits_parquet.hyper attached for the server lifetime so # its external-table catalog (and the plan-cache entries depending on it) diff --git a/hyper-parquet-single/start b/hyper-parquet-single/start index f13ebd21d7..7e21d9f284 100755 --- a/hyper-parquet-single/start +++ b/hyper-parquet-single/start @@ -40,7 +40,12 @@ signal.signal(signal.SIGINT, _terminate) with HyperProcess( telemetry=Telemetry.DO_NOT_SEND_USAGE_DATA_TO_TABLEAU, - parameters={"experimental_persisted_external_tables": "on"}, + parameters={ + "experimental_persisted_external_tables": "on", + # External-table samples are not persisted, so every cold restart would + # otherwise resample the wide Parquet table before executing the query. + "external_sampling": "false", + }, ) as hyper: # Keep-alive: hold hits_parquet.hyper attached for the server lifetime so # its external-table catalog (and the plan-cache entries depending on it)