feat: Added prefer_round_robin for clickbench_partitioned

Author: Weijun-H

benchmarks/bench.sh

@@ -481,7 +481,7 @@ run_clickbench_partitioned() {
     RESULTS_FILE="${RESULTS_DIR}/clickbench_partitioned.json"
     echo "RESULTS_FILE: ${RESULTS_FILE}"
     echo "Running clickbench (partitioned, 100 files) benchmark..."
-    $CARGO_COMMAND --bin dfbench -- clickbench  --iterations 5 --path "${DATA_DIR}/hits_partitioned" --queries-path "${SCRIPT_DIR}/queries/clickbench/queries.sql" -o "${RESULTS_FILE}"
+    $CARGO_COMMAND --bin dfbench -- clickbench  --iterations 5 --path "${DATA_DIR}/hits_partitioned" --prefer_round_robin "${PREFER_ROUND_ROBIN}" --queries-path "${SCRIPT_DIR}/queries/clickbench/queries.sql" -o "${RESULTS_FILE}"
 }
 
 # Runs the clickbench "extended" benchmark with a single large parquet file

benchmarks/src/clickbench.rs

@@ -27,6 +27,9 @@ use datafusion_common::exec_datafusion_err;
 use datafusion_common::instant::Instant;
 use structopt::StructOpt;
 
+// hack to avoid `default_value is meaningless for bool` errors
+type BoolDefaultTrue = bool;
+
 /// Run the clickbench benchmark
 ///
 /// The ClickBench[1] benchmarks are widely cited in the industry and
@@ -68,6 +71,11 @@ pub struct RunOpt {
     /// If present, write results json here
     #[structopt(parse(from_os_str), short = "o", long = "output")]
     output_path: Option<PathBuf>,
+
+    /// If true then round robin repartitioning is used, if false then on demand repartitioning
+    /// True by default.
+    #[structopt(short = "r", long = "prefer_round_robin", default_value = "true")]
+    prefer_round_robin: BoolDefaultTrue,
 }
 
 struct AllQueries {