clickbench-bench

Overview ¶

clickbench-bench runs the ClickBench 43-query suite against an embedded wadjet DB over local hits parquet parts, following the official methodology (https://github.com/ClickHouse/ClickBench):

• per query: drop the OS page cache, then run 3 tries; the first try is the cold run, tries 2-3 are warm
• emit the official results JSON (system/date/machine/tags/load_time/ data_size/result), one [cold, warm, warm] triple per query

Data layout: --data-dir points at a directory of hits_*.parquet parts (athena_partitioned). With --s3-bucket set, parts are first downloaded there from S3 (the download is NOT part of load_time; load_time covers registration, plus ANALYZE when --analyze is set — wadjet queries parquet in place, there is no import step).

Usage (local smoke):

clickbench-bench --data-dir ~/hits-parts --out results.json

Usage (EC2 official run):

clickbench-bench --s3-bucket wadjet-bench-clickbench-use2 \
  --s3-region us-east-2 --s3-prefix tables/hits/ \
  --data-dir /data/hits --machine "c6a.4xlarge, 500gb gp2" \
  --out results.json