Skip to content

Commit 1e8ecd2

Browse files
committed
chore: remove benchmark from this PR (tracked in apache#21582)
1 parent b1784ab commit 1e8ecd2

6 files changed

Lines changed: 1 addition & 168 deletions

File tree

‎benchmarks/bench.sh‎

Lines changed: 0 additions & 130 deletions
Original file line numberDiff line numberDiff line change
@@ -109,9 +109,6 @@ clickbench_extended: ClickBench \"inspired\" queries against a single parquet
109109
# Sort Pushdown Benchmarks
110110
sort_pushdown: Sort pushdown baseline (no WITH ORDER) on TPC-H data (SF=1)
111111
sort_pushdown_sorted: Sort pushdown with WITH ORDER — tests sort elimination on non-overlapping files
112-
sort_pushdown_inexact: Sort pushdown Inexact path (--sorted DESC) — tests reverse scan + RG reorder
113-
sort_pushdown_inexact_unsorted: Sort pushdown Inexact path (no WITH ORDER) — tests Unsupported path + RG reorder
114-
sort_pushdown_inexact_overlap: Sort pushdown Inexact path — partially overlapping RGs (streaming data scenario)
115112
116113
# Sorted Data Benchmarks (ORDER BY Optimization)
117114
clickbench_sorted: ClickBench queries on pre-sorted data using prefer_existing_sort (tests sort elimination optimization)
@@ -319,9 +316,6 @@ main() {
319316
sort_pushdown|sort_pushdown_sorted)
320317
data_sort_pushdown
321318
;;
322-
sort_pushdown_inexact|sort_pushdown_inexact_unsorted|sort_pushdown_inexact_overlap)
323-
data_sort_pushdown_inexact
324-
;;
325319
sort_tpch)
326320
# same data as for tpch
327321
data_tpch "1" "parquet"
@@ -528,15 +522,6 @@ main() {
528522
sort_pushdown_sorted)
529523
run_sort_pushdown_sorted
530524
;;
531-
sort_pushdown_inexact)
532-
run_sort_pushdown_inexact
533-
;;
534-
sort_pushdown_inexact_unsorted)
535-
run_sort_pushdown_inexact_unsorted
536-
;;
537-
sort_pushdown_inexact_overlap)
538-
run_sort_pushdown_inexact_overlap
539-
;;
540525
sort_tpch)
541526
run_sort_tpch "1"
542527
;;
@@ -1152,121 +1137,6 @@ run_sort_pushdown_sorted() {
11521137
debug_run $CARGO_COMMAND --bin dfbench -- sort-pushdown --sorted --iterations 5 --path "${SORT_PUSHDOWN_DIR}" --queries-path "${SCRIPT_DIR}/queries/sort_pushdown" -o "${RESULTS_FILE}" ${QUERY_ARG} ${LATENCY_ARG}
11531138
}
11541139

1155-
# Generates data for sort pushdown Inexact benchmark.
1156-
#
1157-
# Produces a single large lineitem parquet file where row groups have
1158-
# NON-OVERLAPPING but OUT-OF-ORDER l_orderkey ranges (each RG internally
1159-
# sorted, RGs shuffled). This simulates append-heavy workloads where data
1160-
# is written in batches at different times.
1161-
data_sort_pushdown_inexact() {
1162-
INEXACT_DIR="${DATA_DIR}/sort_pushdown_inexact/lineitem"
1163-
if [ -d "${INEXACT_DIR}" ] && [ "$(ls -A ${INEXACT_DIR}/*.parquet 2>/dev/null)" ]; then
1164-
echo "Sort pushdown Inexact data already exists at ${INEXACT_DIR}"
1165-
return
1166-
fi
1167-
1168-
echo "Generating sort pushdown Inexact benchmark data (single file, shuffled RGs)..."
1169-
1170-
# Re-use the sort_pushdown data as the source (generate if missing)
1171-
data_sort_pushdown
1172-
1173-
mkdir -p "${INEXACT_DIR}"
1174-
SRC_DIR="${DATA_DIR}/sort_pushdown/lineitem"
1175-
1176-
# Use datafusion-cli to bucket rows into 64 groups by a deterministic
1177-
# scrambler, then sort within each bucket by orderkey. This produces
1178-
# ~64 RG-sized segments where each has a tight orderkey range but the
1179-
# segments appear in scrambled (non-sorted) order in the file.
1180-
(cd "${SCRIPT_DIR}/.." && cargo run --release -p datafusion-cli -- -c "
1181-
CREATE EXTERNAL TABLE src
1182-
STORED AS PARQUET
1183-
LOCATION '${SRC_DIR}';
1184-
1185-
COPY (
1186-
SELECT * FROM src
1187-
ORDER BY
1188-
(l_orderkey * 1664525 + 1013904223) % 64,
1189-
l_orderkey
1190-
)
1191-
TO '${INEXACT_DIR}/shuffled.parquet'
1192-
STORED AS PARQUET
1193-
OPTIONS ('format.max_row_group_size' '100000');
1194-
")
1195-
1196-
echo "Sort pushdown Inexact shuffled data generated at ${INEXACT_DIR}"
1197-
ls -la "${INEXACT_DIR}"
1198-
1199-
# Also generate a file with partially overlapping row groups.
1200-
# Simulates streaming data with network delays: each chunk is mostly
1201-
# in order but has a small overlap with the next chunk (±5% of the
1202-
# chunk range). This is the pattern described by @adriangb — data
1203-
# arriving with timestamps that are generally increasing but with
1204-
# network-induced jitter causing small overlaps between row groups.
1205-
OVERLAP_DIR="${DATA_DIR}/sort_pushdown_inexact_overlap/lineitem"
1206-
if [ -d "${OVERLAP_DIR}" ] && [ "$(ls -A ${OVERLAP_DIR}/*.parquet 2>/dev/null)" ]; then
1207-
echo "Sort pushdown Inexact overlap data already exists at ${OVERLAP_DIR}"
1208-
return
1209-
fi
1210-
1211-
echo "Generating sort pushdown Inexact overlap data (partially overlapping RGs)..."
1212-
mkdir -p "${OVERLAP_DIR}"
1213-
1214-
(cd "${SCRIPT_DIR}/.." && cargo run --release -p datafusion-cli -- -c "
1215-
CREATE EXTERNAL TABLE src
1216-
STORED AS PARQUET
1217-
LOCATION '${SRC_DIR}';
1218-
1219-
-- Add jitter to l_orderkey: shift each row by a random-ish offset
1220-
-- proportional to its position. This creates overlap between adjacent
1221-
-- row groups while preserving the general ascending trend.
1222-
-- Formula: l_orderkey + (l_orderkey * 7 % 5000) - 2500
1223-
-- This adds ±2500 jitter, creating ~5K overlap between adjacent 100K-row RGs.
1224-
COPY (
1225-
SELECT * FROM src
1226-
ORDER BY l_orderkey + (l_orderkey * 7 % 5000) - 2500
1227-
)
1228-
TO '${OVERLAP_DIR}/overlapping.parquet'
1229-
STORED AS PARQUET
1230-
OPTIONS ('format.max_row_group_size' '100000');
1231-
")
1232-
1233-
echo "Sort pushdown Inexact overlap data generated at ${OVERLAP_DIR}"
1234-
ls -la "${OVERLAP_DIR}"
1235-
}
1236-
1237-
# Runs the sort pushdown Inexact benchmark (tests RG reorder by statistics).
1238-
# Enables pushdown_filters so TopK's dynamic filter is pushed to the parquet
1239-
# reader for late materialization (only needed for Inexact path).
1240-
run_sort_pushdown_inexact() {
1241-
INEXACT_DIR="${DATA_DIR}/sort_pushdown_inexact"
1242-
RESULTS_FILE="${RESULTS_DIR}/sort_pushdown_inexact.json"
1243-
echo "Running sort pushdown Inexact benchmark (--sorted, DESC, reverse scan path)..."
1244-
DATAFUSION_EXECUTION_PARQUET_PUSHDOWN_FILTERS=true \
1245-
debug_run $CARGO_COMMAND --bin dfbench -- sort-pushdown --sorted --iterations 5 --path "${INEXACT_DIR}" --queries-path "${SCRIPT_DIR}/queries/sort_pushdown_inexact" -o "${RESULTS_FILE}" ${QUERY_ARG} ${LATENCY_ARG}
1246-
}
1247-
1248-
# Runs the sort pushdown Inexact benchmark WITHOUT declared ordering.
1249-
# Tests the Unsupported path in try_pushdown_sort where RG reorder by
1250-
# statistics can still help TopK queries without any file ordering guarantee.
1251-
run_sort_pushdown_inexact_unsorted() {
1252-
INEXACT_DIR="${DATA_DIR}/sort_pushdown_inexact"
1253-
RESULTS_FILE="${RESULTS_DIR}/sort_pushdown_inexact_unsorted.json"
1254-
echo "Running sort pushdown Inexact benchmark (no WITH ORDER, Unsupported path)..."
1255-
DATAFUSION_EXECUTION_PARQUET_PUSHDOWN_FILTERS=true \
1256-
debug_run $CARGO_COMMAND --bin dfbench -- sort-pushdown --iterations 5 --path "${INEXACT_DIR}" --queries-path "${SCRIPT_DIR}/queries/sort_pushdown_inexact_unsorted" -o "${RESULTS_FILE}" ${QUERY_ARG} ${LATENCY_ARG}
1257-
}
1258-
1259-
# Runs the sort pushdown benchmark with partially overlapping RGs.
1260-
# Simulates streaming data with network jitter — RGs are mostly in order
1261-
# but have small overlaps (±2500 orderkey jitter between adjacent RGs).
1262-
run_sort_pushdown_inexact_overlap() {
1263-
OVERLAP_DIR="${DATA_DIR}/sort_pushdown_inexact_overlap"
1264-
RESULTS_FILE="${RESULTS_DIR}/sort_pushdown_inexact_overlap.json"
1265-
echo "Running sort pushdown Inexact benchmark (overlapping RGs, streaming data pattern)..."
1266-
DATAFUSION_EXECUTION_PARQUET_PUSHDOWN_FILTERS=true \
1267-
debug_run $CARGO_COMMAND --bin dfbench -- sort-pushdown --sorted --iterations 5 --path "${OVERLAP_DIR}" --queries-path "${SCRIPT_DIR}/queries/sort_pushdown_inexact_overlap" -o "${RESULTS_FILE}" ${QUERY_ARG} ${LATENCY_ARG}
1268-
}
1269-
12701140
# Runs the sort integration benchmark
12711141
run_sort_tpch() {
12721142
SCALE_FACTOR=$1

‎benchmarks/queries/sort_pushdown_inexact/q1.sql‎

Lines changed: 0 additions & 8 deletions
This file was deleted.

‎benchmarks/queries/sort_pushdown_inexact/q2.sql‎

Lines changed: 0 additions & 7 deletions
This file was deleted.

‎benchmarks/queries/sort_pushdown_inexact/q3.sql‎

Lines changed: 0 additions & 8 deletions
This file was deleted.

‎benchmarks/queries/sort_pushdown_inexact/q4.sql‎

Lines changed: 0 additions & 7 deletions
This file was deleted.

‎benchmarks/src/sort_pushdown.rs‎

Lines changed: 1 addition & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -159,14 +159,7 @@ impl RunOpt {
159159
async fn benchmark_query(&self, query_id: usize) -> Result<Vec<QueryResult>> {
160160
let sql = self.load_query(query_id)?;
161161

162-
let mut config = self.common.config()?;
163-
// Enable parquet filter pushdown + late materialization. This is
164-
// essential for the Inexact sort pushdown path: TopK's dynamic
165-
// filter is pushed to the parquet reader, so only sort-column
166-
// rows pass the filter's Decode non-sort columns are skipped for
167-
// rows that don't pass the filter — this is where RG reorder's
168-
// tight-threshold-first strategy pays off for wide-row queries.
169-
config.options_mut().execution.parquet.pushdown_filters = true;
162+
let config = self.common.config()?;
170163
let rt = self.common.build_runtime()?;
171164
let state = SessionStateBuilder::new()
172165
.with_config(config)

0 commit comments

Comments
 (0)