Optimize single-row prolly mutations

[timsehn]

Summary

• add conservative fast paths for single-row prolly delete/insert mutations when leaf chunk boundaries and separator keys are preserved
• rebuild only the edited leaf plus ancestor spine instead of falling through to streaming merge
• add invariant coverage for delete/reinsert on a multi-level integer-key table
• fix non-int sysbench type benchmarks so sbtest_types uses each suite's actual key shape: composite, TEXT, or BLOB

Benchmark target

Latest merged PR #1034 sysbench comment highest multiplier was in the compositepk suite, file-backed autocommit, types_delete_insert_ac at 5.05x. That exposed single-row delete/reinsert costs, and the benchmark now uses the composite PK shape it is supposed to measure.

Local targeted 100k-row timing before the benchmark schema correction, Doltlite only:

• baseline median: 257,222 us
• this branch median: 145,565 us

Tests

• make -j8 sqlite3.o libdoltlite.a
• make -j8 doltlite
• make -j8 invariant_test corruption_test && ./invariant_test && ./corruption_test
• make -j8 c-tests && test/run_c_tests.sh .
• targeted 100k-row types_delete_insert_ac baseline/new timing with test/sysbench_timer.c helper
• BENCH_ROWS=200 BENCH_RUNS=1 BENCH_MAX_MULTIPLIER=1000 BENCH_AVG_MAX_MULTIPLIER=1000 ./test/sysbench_compare_compositepk.sh
• BENCH_ROWS=200 BENCH_RUNS=1 BENCH_MAX_MULTIPLIER=1000 BENCH_AVG_MAX_MULTIPLIER=1000 ./test/sysbench_compare_textpk.sh
• BENCH_ROWS=200 BENCH_RUNS=1 BENCH_MAX_MULTIPLIER=1000 BENCH_AVG_MAX_MULTIPLIER=1000 ./test/sysbench_compare_blobpk.sh

[github-actions]

Sysbench-Style Benchmark (BLOB PK): Doltlite vs SQLite

Companion to the classic Sysbench-Style Benchmark. Every workload here
runs against tables with a 16-byte big-endian BLOB PRIMARY KEY.
Individual ratios gated at 6×; section averages gated at 5×.

In-Memory

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	30,303	44,738	1.48
oltp_range_select	12,557	19,875	1.58
oltp_sum_range	11,838	19,124	1.62
oltp_order_range	2,904	3,877	1.34
oltp_distinct_range	3,902	4,968	1.27
oltp_index_scan	4,437	7,669	1.73
select_random_points	17,991	27,742	1.54
select_random_ranges	4,025	6,480	1.61
covering_index_scan	4,348	5,441	1.25
groupby_scan	32,415	39,401	1.22
index_join	6,601	12,560	1.90
index_join_scan	3,897	7,464	1.92
types_table_scan	1,061,709	1,564,274	1.47
table_scan	1,214,036	1,747,479	1.44
oltp_read_only	125,625	170,264	1.36
Average			1.51

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	240,874	339,113	1.41
oltp_insert	19,953	37,853	1.90
oltp_update_index	67,422	139,571	2.07
oltp_update_non_index	47,613	89,403	1.88
oltp_delete_insert	47,776	105,845	2.22
oltp_write_only	27,207	62,878	2.31
types_delete_insert	32,067	59,113	1.84
oltp_read_write	82,540	161,417	1.96
Average			1.95

File-Backed

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	106,407	120,009	1.13
oltp_range_select	21,660	48,761	2.25
oltp_sum_range	21,099	48,234	2.29
oltp_order_range	3,909	7,825	2.00
oltp_distinct_range	4,908	8,958	1.83
oltp_index_scan	12,860	20,979	1.63
select_random_points	33,368	82,214	2.46
select_random_ranges	11,482	17,523	1.53
covering_index_scan	11,665	13,115	1.12
groupby_scan	35,058	61,274	1.75
index_join	13,068	32,789	2.51
index_join_scan	6,406	23,001	3.59
types_table_scan	1,235,877	3,550,057	2.87
table_scan	1,460,501	4,241,584	2.90
oltp_read_only	239,142	358,762	1.50
Average			2.09

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	251,294	364,040	1.45
oltp_insert	36,415	57,491	1.58
oltp_update_index	169,795	303,632	1.79
oltp_update_non_index	105,461	184,892	1.75
oltp_delete_insert	116,431	211,033	1.81
oltp_write_only	72,211	133,377	1.85
types_delete_insert	63,547	115,512	1.82
oltp_read_write	156,128	339,214	2.17
Average			1.78

File-Backed (autocommit)

Each statement runs as its own transaction — exposes per-commit
fixed costs that the wrapped-in-BEGIN/COMMIT tests amortize away.
SQLite uses WAL mode with synchronous=FULL in this section so
the comparison uses SQLite's durable WAL autocommit path.

Reads

Reads have no commit cost; these are the same SQL files as the
File-Backed Reads section, included here for symmetry and to
catch any per-statement overhead doltlite pays on the read path.

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	60,951	119,200	1.96
oltp_range_select	17,284	48,622	2.81
oltp_sum_range	16,879	48,673	2.88
oltp_order_range	3,415	7,857	2.30
oltp_distinct_range	4,506	8,984	1.99
oltp_index_scan	8,502	21,104	2.48
select_random_points	28,537	82,518	2.89
select_random_ranges	7,202	17,677	2.45
covering_index_scan	7,495	13,078	1.74
groupby_scan	34,791	61,636	1.77
index_join	11,005	32,545	2.96
index_join_scan	6,065	23,123	3.81
types_table_scan	1,239,354	3,552,837	2.87
table_scan	1,475,800	4,244,559	2.88
oltp_read_only	172,894	356,859	2.06
Average			2.52

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert_ac	25,648	81,407	3.17
oltp_insert_ac	26,658	103,507	3.88
oltp_update_index_ac	28,778	115,560	4.02
oltp_update_non_index_ac	24,424	96,179	3.94
oltp_delete_insert_ac	25,329	107,446	4.24
oltp_write_only_ac	25,367	110,130	4.34
types_delete_insert_ac	23,494	101,801	4.33
oltp_read_write_ac	32,751	126,058	3.85
Average			3.97

100000 rows, median of 5 invocations per test, workload-only timing via host monotonic clock when available.

Performance Ceiling Check (6x individual, 5x average)

All tests within ceilings.

[github-actions]

Sysbench-Style Benchmark (composite PK): Doltlite vs SQLite

Companion to the classic Sysbench-Style Benchmark. Every workload here
runs against tables with a 2-column INTEGER PRIMARY KEY(a, b) WITHOUT ROWID.
Individual ratios gated at 6×; section averages gated at 5×.

In-Memory

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	34,567	45,159	1.31
oltp_range_select	21,634	26,032	1.20
oltp_sum_range	19,493	24,586	1.26
oltp_order_range	3,868	4,445	1.15
oltp_distinct_range	4,973	5,600	1.13
oltp_index_scan	4,841	7,558	1.56
select_random_points	28,520	38,394	1.35
select_random_ranges	7,789	9,548	1.23
covering_index_scan	4,350	5,075	1.17
groupby_scan	40,631	45,976	1.13
index_join	8,133	12,575	1.55
index_join_scan	4,289	7,301	1.70
types_table_scan	1,231,850	1,523,209	1.24
table_scan	1,288,917	1,724,256	1.34
oltp_read_only	152,131	186,643	1.23
Average			1.30

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	243,447	318,062	1.31
oltp_insert	19,104	33,559	1.76
oltp_update_index	68,552	125,676	1.83
oltp_update_non_index	50,612	84,717	1.67
oltp_delete_insert	49,933	96,561	1.93
oltp_write_only	27,190	59,898	2.20
types_delete_insert	32,454	58,576	1.80
oltp_read_write	99,565	165,861	1.67
Average			1.77

File-Backed

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	125,465	126,519	1.01
oltp_range_select	31,007	53,883	1.74
oltp_sum_range	29,790	52,426	1.76
oltp_order_range	4,796	8,214	1.71
oltp_distinct_range	5,868	9,220	1.57
oltp_index_scan	14,322	20,943	1.46
select_random_points	45,057	96,534	2.14
select_random_ranges	16,607	21,333	1.28
covering_index_scan	12,468	11,180	0.90
groupby_scan	43,260	66,855	1.55
index_join	13,552	27,298	2.01
index_join_scan	5,421	17,976	3.32
types_table_scan	1,331,767	3,401,565	2.55
table_scan	1,622,575	4,241,182	2.61
oltp_read_only	289,897	377,485	1.30
Average			1.79

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	253,224	345,155	1.36
oltp_insert	25,080	50,359	2.01
oltp_update_index	170,297	283,933	1.67
oltp_update_non_index	105,153	185,849	1.77
oltp_delete_insert	106,449	200,695	1.89
oltp_write_only	64,347	129,461	2.01
types_delete_insert	61,477	109,515	1.78
oltp_read_write	165,846	348,309	2.10
Average			1.82

File-Backed (autocommit)

Each statement runs as its own transaction — exposes per-commit
fixed costs that the wrapped-in-BEGIN/COMMIT tests amortize away.
SQLite uses WAL mode with synchronous=FULL in this section so
the comparison uses SQLite's durable WAL autocommit path.

Reads

Reads have no commit cost; these are the same SQL files as the
File-Backed Reads section, included here for symmetry and to
catch any per-statement overhead doltlite pays on the read path.

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	70,792	130,086	1.84
oltp_range_select	25,998	55,887	2.15
oltp_sum_range	24,577	54,203	2.21
oltp_order_range	4,374	8,244	1.88
oltp_distinct_range	5,452	9,372	1.72
oltp_index_scan	9,122	21,777	2.39
select_random_points	40,810	97,606	2.39
select_random_ranges	11,287	21,193	1.88
covering_index_scan	7,504	11,030	1.47
groupby_scan	42,577	66,778	1.57
index_join	10,802	27,253	2.52
index_join_scan	4,959	17,885	3.61
types_table_scan	1,306,061	3,381,265	2.59
table_scan	1,558,204	4,231,301	2.72
oltp_read_only	213,055	380,100	1.78
Average			2.18

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert_ac	15,317	56,969	3.72
oltp_insert_ac	18,334	77,267	4.21
oltp_update_index_ac	21,487	89,051	4.14
oltp_update_non_index_ac	17,604	75,329	4.28
oltp_delete_insert_ac	18,543	83,884	4.52
oltp_write_only_ac	18,492	85,416	4.62
types_delete_insert_ac	15,574	77,487	4.98
oltp_read_write_ac	26,174	110,395	4.22
Average			4.34

100000 rows, median of 5 invocations per test, workload-only timing via host monotonic clock when available.

Performance Ceiling Check (6x individual, 5x average)

All tests within ceilings.

[github-actions]

Sysbench-Style Benchmark (TEXT PK): Doltlite vs SQLite

Companion to the classic Sysbench-Style Benchmark. Every workload here
runs against tables with a 32-char hex TEXT PRIMARY KEY (UUID-shaped).
Individual ratios gated at 6×; section averages gated at 5×.

In-Memory

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	29,369	45,458	1.55
oltp_range_select	13,031	20,391	1.56
oltp_sum_range	11,810	19,638	1.66
oltp_order_range	3,015	3,918	1.30
oltp_distinct_range	4,094	4,999	1.22
oltp_index_scan	4,517	7,824	1.73
select_random_points	17,663	29,392	1.66
select_random_ranges	4,024	6,492	1.61
covering_index_scan	4,383	5,201	1.19
groupby_scan	32,604	39,127	1.20
index_join	6,709	12,437	1.85
index_join_scan	4,584	7,796	1.70
types_table_scan	1,081,628	1,608,248	1.49
table_scan	1,299,840	1,750,462	1.35
oltp_read_only	116,590	169,890	1.46
Average			1.50

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	228,627	335,867	1.47
oltp_insert	21,191	38,726	1.83
oltp_update_index	69,087	143,469	2.08
oltp_update_non_index	47,684	91,451	1.92
oltp_delete_insert	49,345	107,718	2.18
oltp_write_only	27,657	63,530	2.30
types_delete_insert	31,557	60,035	1.90
oltp_read_write	82,658	162,599	1.97
Average			1.96

File-Backed

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	104,667	122,185	1.17
oltp_range_select	22,273	51,801	2.33
oltp_sum_range	21,382	51,731	2.42
oltp_order_range	3,865	8,104	2.10
oltp_distinct_range	4,895	9,151	1.87
oltp_index_scan	12,886	21,225	1.65
select_random_points	33,016	83,932	2.54
select_random_ranges	11,453	17,664	1.54
covering_index_scan	12,287	13,172	1.07
groupby_scan	35,327	64,942	1.84
index_join	14,068	34,541	2.46
index_join_scan	7,625	24,992	3.28
types_table_scan	1,308,794	3,906,242	2.98
table_scan	1,557,068	4,617,665	2.97
oltp_read_only	243,273	381,725	1.57
Average			2.12

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	240,360	367,738	1.53
oltp_insert	47,694	59,472	1.25
oltp_update_index	187,332	319,416	1.71
oltp_update_non_index	137,497	199,190	1.45
oltp_delete_insert	130,240	228,012	1.75
oltp_write_only	128,539	141,116	1.10
types_delete_insert	68,390	124,040	1.81
oltp_read_write	184,135	366,256	1.99
Average			1.57

File-Backed (autocommit)

Each statement runs as its own transaction — exposes per-commit
fixed costs that the wrapped-in-BEGIN/COMMIT tests amortize away.
SQLite uses WAL mode with synchronous=FULL in this section so
the comparison uses SQLite's durable WAL autocommit path.

Reads

Reads have no commit cost; these are the same SQL files as the
File-Backed Reads section, included here for symmetry and to
catch any per-statement overhead doltlite pays on the read path.

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	61,216	126,235	2.06
oltp_range_select	18,203	53,527	2.94
oltp_sum_range	17,010	53,031	3.12
oltp_order_range	3,498	8,168	2.34
oltp_distinct_range	4,553	9,138	2.01
oltp_index_scan	8,744	21,170	2.42
select_random_points	28,214	84,513	3.00
select_random_ranges	7,079	17,734	2.51
covering_index_scan	7,979	13,275	1.66
groupby_scan	34,636	64,278	1.86
index_join	11,889	34,594	2.91
index_join_scan	7,277	24,536	3.37
types_table_scan	1,309,750	3,855,156	2.94
table_scan	1,502,430	4,525,660	3.01
oltp_read_only	174,773	374,230	2.14
Average			2.55

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert_ac	23,776	78,654	3.31
oltp_insert_ac	26,303	90,179	3.43
oltp_update_index_ac	26,832	110,646	4.12
oltp_update_non_index_ac	22,082	96,284	4.36
oltp_delete_insert_ac	25,206	103,540	4.11
oltp_write_only_ac	24,652	99,864	4.05
types_delete_insert_ac	21,354	87,943	4.12
oltp_read_write_ac	31,446	121,335	3.86
Average			3.92

100000 rows, median of 5 invocations per test, workload-only timing via host monotonic clock when available.

Performance Ceiling Check (6x individual, 5x average)

All tests within ceilings.

[github-actions]

Sysbench-Style Benchmark: Doltlite vs SQLite

In-Memory

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	23,797	36,175	1.52
oltp_range_select	9,684	14,295	1.48
oltp_sum_range	9,117	14,268	1.56
oltp_order_range	2,516	3,204	1.27
oltp_distinct_range	3,556	4,272	1.20
oltp_index_scan	3,667	6,623	1.81
select_random_points	9,270	17,311	1.87
select_random_ranges	2,696	5,197	1.93
covering_index_scan	4,113	4,596	1.12
groupby_scan	30,740	34,432	1.12
index_join	5,966	9,334	1.56
index_join_scan	3,489	5,708	1.64
types_table_scan	1,135,043	1,434,103	1.26
table_scan	1,234,842	1,542,049	1.25
oltp_read_only	112,941	145,683	1.29
Average			1.46

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	178,869	246,451	1.38
oltp_insert	15,939	28,356	1.78
oltp_update_index	50,450	101,641	2.01
oltp_update_non_index	33,549	67,572	2.01
oltp_delete_insert	43,461	77,839	1.79
oltp_write_only	21,222	48,592	2.29
types_delete_insert	24,186	44,249	1.83
oltp_read_write	65,089	129,977	2.00
Average			1.89

File-Backed

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	98,074	103,406	1.05
oltp_range_select	18,935	35,225	1.86
oltp_sum_range	18,703	35,088	1.88
oltp_order_range	3,513	6,224	1.77
oltp_distinct_range	4,563	7,276	1.59
oltp_index_scan	11,925	17,369	1.46
select_random_points	23,576	62,635	2.66
select_random_ranges	10,606	14,961	1.41
covering_index_scan	11,326	8,113	0.72
groupby_scan	33,429	51,710	1.55
index_join	10,466	18,825	1.80
index_join_scan	4,727	13,510	2.86
types_table_scan	1,264,624	2,760,426	2.18
table_scan	1,427,462	3,333,273	2.34
oltp_read_only	219,534	282,985	1.29
Average			1.76

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	187,880	263,182	1.40
oltp_insert	22,004	37,838	1.72
oltp_update_index	125,051	227,782	1.82
oltp_update_non_index	76,904	145,567	1.89
oltp_delete_insert	95,070	164,779	1.73
oltp_write_only	58,274	107,012	1.84
types_delete_insert	52,162	89,362	1.71
oltp_read_write	122,323	262,594	2.15
Average			1.78

File-Backed (autocommit)

Each statement runs as its own transaction — exposes per-commit
fixed costs that the wrapped-in-BEGIN/COMMIT tests amortize away.
SQLite uses WAL mode with synchronous=FULL in this section so
the comparison uses SQLite's durable WAL autocommit path.

Reads

Reads have no commit cost; these are the same SQL files as the
File-Backed Reads section, included here for symmetry and to
catch any per-statement overhead doltlite pays on the read path.

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	53,763	103,791	1.93
oltp_range_select	14,392	36,107	2.51
oltp_sum_range	14,417	35,953	2.49
oltp_order_range	3,305	6,394	1.93
oltp_distinct_range	4,343	7,589	1.75
oltp_index_scan	7,749	17,554	2.27
select_random_points	19,243	62,627	3.25
select_random_ranges	6,202	14,885	2.40
covering_index_scan	7,020	7,930	1.13
groupby_scan	32,675	50,489	1.55
index_join	8,009	17,916	2.24
index_join_scan	3,951	12,623	3.19
types_table_scan	1,232,537	2,818,399	2.29
table_scan	1,437,551	3,467,889	2.41
oltp_read_only	159,797	286,671	1.79
Average			2.21

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert_ac	21,928	74,876	3.41
oltp_insert_ac	25,570	98,223	3.84
oltp_update_index_ac	25,720	111,430	4.33
oltp_update_non_index_ac	23,537	92,706	3.94
oltp_delete_insert_ac	24,217	105,058	4.34
oltp_write_only_ac	24,012	99,301	4.14
types_delete_insert_ac	22,492	85,821	3.82
oltp_read_write_ac	31,792	112,827	3.55
Average			3.92

100000 rows, median of 5 invocations per test, workload-only timing via host monotonic clock when available.

Performance Ceiling Check (6x individual, 5x average)

All tests within ceilings.