Skip to content

Apache DataFusion SQL Query Engine

Sql benchmark sync updated queries (#24583)

Bruce Ritchiec6999963455d
diff --git a/benchmarks/sql_benchmarks/clickbench_extended/benchmarks/q13.benchmark b/benchmarks/sql_benchmarks/clickbench_extended/benchmarks/q13.benchmark
new file mode 100644
index 0000000000..66f832f3cb
--- /dev/null
+++ b/benchmarks/sql_benchmarks/clickbench_extended/benchmarks/q13.benchmark
@@ -0,0 +1,19 @@
+name Q13
+group clickbench_extended
+subgroup ${CLICKBENCH_TYPE:-single}
+
+init sql_benchmarks/clickbench/init/set_config.sql
+
+load sql_benchmarks/clickbench/init/load-${CLICKBENCH_TYPE:-single}.sql
+
+assert I
+SELECT COUNT(*) > 0 from hits;
+----
+true
+
+run
+SELECT SUM("CounterID") AS counter_id_sum
+FROM hits
+WHERE "URL" < 'zzzz';
+
+result sql_benchmarks/clickbench_extended/results/q13.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q13.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q13.benchmark
new file mode 100644
index 0000000000..20382d65ba
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q13.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q13
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (ROW_NUMBER top-2 per partition)
+SELECT id2, largest2_v2 FROM (
+ SELECT id2, v2 AS largest2_v2,
+ ROW_NUMBER() OVER (PARTITION BY id2 ORDER BY v2 DESC) AS order_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE order_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q13.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q14.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q14.benchmark
new file mode 100644
index 0000000000..0617791988
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q14.benchmark
@@ -0,0 +1,26 @@
+subgroup window
+
+name Q14
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N partition cardinality sweep (id3 % N gives N distinct partitions).
+-- These exercise PartitionedTopKExec across cardinalities to validate it stays
+-- competitive with the SortExec+Filter baseline as partition count grows.
+-- Window Top-N: 100 partitions
+SELECT pk, largest2_v2 FROM (
+ SELECT id3 % 100 AS pk, v2 AS largest2_v2,
+ ROW_NUMBER() OVER (PARTITION BY id3 % 100 ORDER BY v2 DESC) AS order_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE order_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q14.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q15.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q15.benchmark
new file mode 100644
index 0000000000..f030b249db
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q15.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q15
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N: 1,000 partitions
+SELECT pk, largest2_v2 FROM (
+ SELECT id3 % 1000 AS pk, v2 AS largest2_v2,
+ ROW_NUMBER() OVER (PARTITION BY id3 % 1000 ORDER BY v2 DESC) AS order_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE order_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q15.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q16.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q16.benchmark
new file mode 100644
index 0000000000..3356f4f463
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q16.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q16
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N: 10,000 partitions
+SELECT pk, largest2_v2 FROM (
+ SELECT id3 % 10000 AS pk, v2 AS largest2_v2,
+ ROW_NUMBER() OVER (PARTITION BY id3 % 10000 ORDER BY v2 DESC) AS order_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE order_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q16.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q17.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q17.benchmark
new file mode 100644
index 0000000000..3228204558
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q17.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q17
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N: 100,000 partitions
+SELECT pk, largest2_v2 FROM (
+ SELECT id3 % 100000 AS pk, v2 AS largest2_v2,
+ ROW_NUMBER() OVER (PARTITION BY id3 % 100000 ORDER BY v2 DESC) AS order_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE order_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q17.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q18.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q18.benchmark
new file mode 100644
index 0000000000..666b4a6810
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q18.benchmark
@@ -0,0 +1,27 @@
+subgroup window
+
+name Q18
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (RANK top-2 per partition, ~100 partitions)
+-- The RANK queries below mirror the ROW_NUMBER cardinality sweep
+-- above and add heavy-ties variants. RANK semantics retain boundary
+-- ties (`WHERE rk <= K` may keep more than K rows per partition), so
+-- this exercises PartitionedTopKRank's ties-Vec path.
+SELECT pk, largest_v2 FROM (
+ SELECT (id3 % 100) AS pk, v2 AS largest_v2,
+ RANK() OVER (PARTITION BY (id3 % 100) ORDER BY v2 DESC) AS rk_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE rk_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q18.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q19.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q19.benchmark
new file mode 100644
index 0000000000..2ddf98d5de
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q19.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q19
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (RANK top-2 per partition, ~1K partitions)
+SELECT pkey, largest_v2 FROM (
+ SELECT (id3 % 1000) AS pkey, v2 AS largest_v2,
+ RANK() OVER (PARTITION BY (id3 % 1000) ORDER BY v2 DESC) AS rk_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE rk_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q19.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q20.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q20.benchmark
new file mode 100644
index 0000000000..ca31c84334
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q20.benchmark
@@ -0,0 +1,25 @@
+subgroup window
+
+name Q20
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (RANK top-2 per partition, ~1K partitions, heavy ties)
+-- v2 % 10 forces 10 distinct OBY values, so most rows tie at the boundary
+-- and exercise PartitionedTopKRank's ties-Vec path.
+SELECT pkey, largest_v2 FROM (
+ SELECT (id3 % 1000) AS pkey, v2 AS largest_v2,
+ RANK() OVER (PARTITION BY (id3 % 1000) ORDER BY (v2 % 10) DESC) AS rk_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE rk_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q20.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q21.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q21.benchmark
new file mode 100644
index 0000000000..b26fbc264e
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q21.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q21
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (RANK top-2 per partition, ~10K partitions, low ties)
+SELECT id2, largest_v2 FROM (
+ SELECT id2, v2 AS largest_v2,
+ RANK() OVER (PARTITION BY id2 ORDER BY v2 DESC) AS rk_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE rk_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q21.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q22.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q22.benchmark
new file mode 100644
index 0000000000..cbe2dbe558
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q22.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q22
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (RANK top-2 per partition, ~10K partitions, heavy ties)
+SELECT id2, largest_v2 FROM (
+ SELECT id2, v2 AS largest_v2,
+ RANK() OVER (PARTITION BY id2 ORDER BY (v2 % 10) DESC) AS rk_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE rk_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q22.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q23.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q23.benchmark
new file mode 100644
index 0000000000..b86b2f742c
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q23.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q23
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (RANK top-2 per partition, ~100K partitions)
+SELECT pk, largest_v2 FROM (
+ SELECT (id3 % 100000) AS pk, v2 AS largest_v2,
+ RANK() OVER (PARTITION BY (id3 % 100000) ORDER BY v2 DESC) AS rk_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE rk_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q23.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q24.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q24.benchmark
new file mode 100644
index 0000000000..0241fb6f15
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q24.benchmark
@@ -0,0 +1,28 @@
+subgroup window
+
+name Q24
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (DENSE_RANK top-2 per partition, ~100 partitions)
+-- The DENSE_RANK queries below mirror the RANK cardinality sweep above.
+-- DENSE_RANK semantics keep every row whose ORDER BY value is among the
+-- K distinct-greatest values in the partition, so total kept per partition
+-- is unbounded in rows-per-distinct-value — exercises PartitionedTopKDenseRank's
+-- HashMap-of-groups path.
+SELECT pk, largest_v2 FROM (
+ SELECT (id3 % 100) AS pk, v2 AS largest_v2,
+ DENSE_RANK() OVER (PARTITION BY (id3 % 100) ORDER BY v2 DESC) AS dr_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE dr_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q24.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q25.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q25.benchmark
new file mode 100644
index 0000000000..0110f94892
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q25.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q25
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (DENSE_RANK top-2 per partition, ~1K partitions)
+SELECT pkey, largest_v2 FROM (
+ SELECT (id3 % 1000) AS pkey, v2 AS largest_v2,
+ DENSE_RANK() OVER (PARTITION BY (id3 % 1000) ORDER BY v2 DESC) AS dr_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE dr_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q25.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q26.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q26.benchmark
new file mode 100644
index 0000000000..d5080c92b7
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q26.benchmark
@@ -0,0 +1,26 @@
+subgroup window
+
+name Q26
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (DENSE_RANK top-2 per partition, ~1K partitions, heavy ties)
+-- v2 % 10 forces 10 distinct OBY values; most rows share the top-2 distinct
+-- values so appends dominate — exercises the "Case A" append-to-existing-Vec
+-- fast path in PartitionedTopKDenseRank.
+SELECT pkey, largest_v2 FROM (
+ SELECT (id3 % 1000) AS pkey, v2 AS largest_v2,
+ DENSE_RANK() OVER (PARTITION BY (id3 % 1000) ORDER BY (v2 % 10) DESC) AS dr_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE dr_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q26.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q27.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q27.benchmark
new file mode 100644
index 0000000000..72eaa36b3e
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q27.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q27
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (DENSE_RANK top-2 per partition, ~10K partitions, low ties)
+SELECT id2, largest_v2 FROM (
+ SELECT id2, v2 AS largest_v2,
+ DENSE_RANK() OVER (PARTITION BY id2 ORDER BY v2 DESC) AS dr_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE dr_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q27.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q28.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q28.benchmark
new file mode 100644
index 0000000000..d71e994e15
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q28.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q28
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (DENSE_RANK top-2 per partition, ~10K partitions, heavy ties)
+SELECT id2, largest_v2 FROM (
+ SELECT id2, v2 AS largest_v2,
+ DENSE_RANK() OVER (PARTITION BY id2 ORDER BY (v2 % 10) DESC) AS dr_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE dr_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q28.csv
diff --git a/benchmarks/sql_benchmarks/h2o/benchmarks/window/q29.benchmark b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q29.benchmark
new file mode 100644
index 0000000000..e6eade7e14
--- /dev/null
+++ b/benchmarks/sql_benchmarks/h2o/benchmarks/window/q29.benchmark
@@ -0,0 +1,23 @@
+subgroup window
+
+name Q29
+group h2o
+
+echo Loading ${H2O_BENCH_SIZE:-small} window ${H2O_FILE_TYPE:-csv} h2o data
+
+load sql_benchmarks/h2o/init/load_window_${H2O_BENCH_SIZE:-small}_${H2O_FILE_TYPE:-csv}.sql
+
+assert I
+SELECT COUNT(*) > 0 FROM x
+----
+true
+
+run
+-- Window Top-N (DENSE_RANK top-2 per partition, ~100K partitions)
+SELECT pk, largest_v2 FROM (
+ SELECT (id3 % 100000) AS pk, v2 AS largest_v2,
+ DENSE_RANK() OVER (PARTITION BY (id3 % 100000) ORDER BY v2 DESC) AS dr_v2
+ FROM x WHERE v2 IS NOT NULL
+) sub_query WHERE dr_v2 <= 2;
+
+result sql_benchmarks/h2o/results/window/${H2O_BENCH_SIZE:-small}/q29.csv
diff --git a/benchmarks/sql_benchmarks/smj/benchmarks/q24.benchmark b/benchmarks/sql_benchmarks/smj/benchmarks/q24.benchmark
index 893eb1fb78..44ad0c6774 100644
--- a/benchmarks/sql_benchmarks/smj/benchmarks/q24.benchmark
+++ b/benchmarks/sql_benchmarks/smj/benchmarks/q24.benchmark
@@ -5,6 +5,7 @@ init
set datafusion.optimizer.prefer_hash_join=false;
expect_plan SortMergeJoinExec
+expect_plan LeftMark
run
-- Q24: LEFT MARK 1M x 10M | 1:10 | 1%
diff --git a/benchmarks/sql_benchmarks/smj/benchmarks/q25.benchmark b/benchmarks/sql_benchmarks/smj/benchmarks/q25.benchmark
new file mode 100644
index 0000000000..a1ec8dcf76
--- /dev/null
+++ b/benchmarks/sql_benchmarks/smj/benchmarks/q25.benchmark
@@ -0,0 +1,33 @@
+name Q25
+group smj
+
+init
+set datafusion.optimizer.prefer_hash_join=false;
+
+expect_plan SortMergeJoinExec
+expect_plan LeftMark
+
+run
+-- Q25: LEFT MARK 1M x 10M | 1:10 | 50%
+WITH t1_sorted AS (
+ SELECT value % 100000 as key, value as data
+FROM range(1000000)
+ORDER BY key, data
+ ),
+ t2_sorted AS (
+SELECT value % 100000 as key, value as data
+FROM range(10000000)
+ORDER BY key, data
+ )
+SELECT t1_sorted.key, t1_sorted.data
+FROM t1_sorted
+WHERE t1_sorted.data < 0
+ OR EXISTS (
+ SELECT 1 FROM t2_sorted
+ WHERE t2_sorted.key = t1_sorted.key
+ AND t2_sorted.data <> t1_sorted.data
+ AND t2_sorted.data % 2 = 0
+)
+
+cleanup
+reset datafusion.optimizer.prefer_hash_join;
diff --git a/benchmarks/sql_benchmarks/smj/benchmarks/q26.benchmark b/benchmarks/sql_benchmarks/smj/benchmarks/q26.benchmark
new file mode 100644
index 0000000000..b164817d2b
--- /dev/null
+++ b/benchmarks/sql_benchmarks/smj/benchmarks/q26.benchmark
@@ -0,0 +1,33 @@
+name Q26
+group smj
+
+init
+set datafusion.optimizer.prefer_hash_join=false;
+
+expect_plan SortMergeJoinExec
+expect_plan LeftMark
+
+run
+-- Q26: LEFT MARK 1M x 10M | 1:10 | 90%
+WITH t1_sorted AS (
+ SELECT value % 100000 as key, value as data
+FROM range(1000000)
+ORDER BY key, data
+ ),
+ t2_sorted AS (
+SELECT value % 100000 as key, value as data
+FROM range(10000000)
+ORDER BY key, data
+ )
+SELECT t1_sorted.key, t1_sorted.data
+FROM t1_sorted
+WHERE t1_sorted.data < 0
+ OR EXISTS (
+ SELECT 1 FROM t2_sorted
+ WHERE t2_sorted.key = t1_sorted.key
+ AND t2_sorted.data <> t1_sorted.data
+ AND t2_sorted.data % 10 <> 0
+)
+
+cleanup
+reset datafusion.optimizer.prefer_hash_join;