Don't do alignment computations during BFLATN ser/de

[gefjon]

Description of Changes

AlgebraicTypeLayout and friends already include full layout information, including properly-aligned offsets for ProductTypeElementLayouts. As such, there's no need to do any alignment computation during serialize_value or write_value.

Instead, while traversing a ProductTypeLayout, we can use each element's offset to update the curr_offset.

API and ABI breaking changes

N/a

Expected complexity level and risk

3 - minor change, but in close contact with some unsafe code.

Testing

• table tests pass, including proptests.
• Wait for full CI to pass, incl. core tests and smoketests.

[gefjon]

benchmarks please

[github-actions]

Criterion benchmark results

Criterion benchmark report

YOU SHOULD PROBABLY IGNORE THESE RESULTS.

Criterion is a wall time based benchmarking system that is extremely noisy when run on CI. We collect these results for longitudinal analysis, but they are not reliable for comparing individual PRs.

Go look at the callgrind report instead.

empty

db	on disk	new latency	old latency	new throughput	old throughput
sqlite	💿	-	412.2±1.39ns	-	-
sqlite	🧠	-	406.1±1.22ns	-	-
stdb_raw	💿	707.1±1.14ns	713.9±1.10ns	-	-
stdb_raw	🧠	684.0±0.88ns	680.0±0.74ns	-	-

insert_1

db	on disk	schema	indices	preload	new latency	old latency	new throughput	old throughput

insert_bulk

db	on disk	schema	indices	preload	count	new latency	old latency	new throughput	old throughput
sqlite	💿	u32_u64_str	btree_each_column	2048	256	-	530.0±49.57µs	-	1886 tx/sec
sqlite	💿	u32_u64_str	unique_0	2048	256	-	134.7±0.36µs	-	7.2 Ktx/sec
sqlite	💿	u32_u64_u64	btree_each_column	2048	256	-	421.7±0.73µs	-	2.3 Ktx/sec
sqlite	💿	u32_u64_u64	unique_0	2048	256	-	121.1±0.73µs	-	8.1 Ktx/sec
sqlite	🧠	u32_u64_str	btree_each_column	2048	256	-	443.9±0.67µs	-	2.2 Ktx/sec
sqlite	🧠	u32_u64_str	unique_0	2048	256	-	120.7±0.78µs	-	8.1 Ktx/sec
sqlite	🧠	u32_u64_u64	btree_each_column	2048	256	-	366.8±0.45µs	-	2.7 Ktx/sec
sqlite	🧠	u32_u64_u64	unique_0	2048	256	-	103.0±0.76µs	-	9.5 Ktx/sec
stdb_raw	💿	u32_u64_str	btree_each_column	2048	256	709.7±0.61µs	725.6±1.60µs	1409 tx/sec	1378 tx/sec
stdb_raw	💿	u32_u64_str	unique_0	2048	256	617.2±0.56µs	625.7±0.57µs	1620 tx/sec	1598 tx/sec
stdb_raw	💿	u32_u64_u64	btree_each_column	2048	256	417.7±0.68µs	435.4±0.93µs	2.3 Ktx/sec	2.2 Ktx/sec
stdb_raw	💿	u32_u64_u64	unique_0	2048	256	376.2±0.52µs	393.9±0.68µs	2.6 Ktx/sec	2.5 Ktx/sec
stdb_raw	🧠	u32_u64_str	btree_each_column	2048	256	487.6±0.44µs	500.8±0.22µs	2.0 Ktx/sec	1996 tx/sec
stdb_raw	🧠	u32_u64_str	unique_0	2048	256	398.2±1.19µs	407.1±0.65µs	2.5 Ktx/sec	2.4 Ktx/sec
stdb_raw	🧠	u32_u64_u64	btree_each_column	2048	256	315.8±0.57µs	328.6±0.42µs	3.1 Ktx/sec	3.0 Ktx/sec
stdb_raw	🧠	u32_u64_u64	unique_0	2048	256	278.2±0.38µs	295.9±0.99µs	3.5 Ktx/sec	3.3 Ktx/sec

iterate

db	on disk	schema	indices	new latency	old latency	new throughput	old throughput
sqlite	💿	u32_u64_str	unique_0	-	20.4±0.18µs	-	47.9 Ktx/sec
sqlite	💿	u32_u64_u64	unique_0	-	19.4±0.16µs	-	50.3 Ktx/sec
sqlite	🧠	u32_u64_str	unique_0	-	18.7±0.15µs	-	52.3 Ktx/sec
sqlite	🧠	u32_u64_u64	unique_0	-	18.1±0.26µs	-	53.8 Ktx/sec
stdb_raw	💿	u32_u64_str	unique_0	18.7±0.00µs	18.7±0.00µs	52.3 Ktx/sec	52.3 Ktx/sec
stdb_raw	💿	u32_u64_u64	unique_0	15.9±0.00µs	15.8±0.00µs	61.6 Ktx/sec	61.6 Ktx/sec
stdb_raw	🧠	u32_u64_str	unique_0	18.7±0.00µs	18.6±0.00µs	52.3 Ktx/sec	52.4 Ktx/sec
stdb_raw	🧠	u32_u64_u64	unique_0	15.8±0.00µs	15.8±0.00µs	61.7 Ktx/sec	61.8 Ktx/sec

find_unique

db	on disk	key type	preload	new latency	old latency	new throughput	old throughput

filter

db	on disk	key type	index strategy	load	count	new latency	old latency	new throughput	old throughput
sqlite	💿	string	index	2048	256	-	68.3±0.28µs	-	14.3 Ktx/sec
sqlite	💿	u64	index	2048	256	-	64.0±0.19µs	-	15.2 Ktx/sec
sqlite	🧠	string	index	2048	256	-	65.5±0.17µs	-	14.9 Ktx/sec
sqlite	🧠	u64	index	2048	256	-	59.7±0.29µs	-	16.4 Ktx/sec
stdb_raw	💿	string	index	2048	256	5.6±0.00µs	5.6±0.00µs	175.5 Ktx/sec	175.5 Ktx/sec
stdb_raw	💿	u64	index	2048	256	5.5±0.00µs	5.5±0.00µs	178.1 Ktx/sec	178.2 Ktx/sec
stdb_raw	🧠	string	index	2048	256	5.5±0.00µs	5.5±0.01µs	176.6 Ktx/sec	176.3 Ktx/sec
stdb_raw	🧠	u64	index	2048	256	5.4±0.00µs	5.5±0.00µs	179.3 Ktx/sec	179.0 Ktx/sec

serialize

schema	format	count	new latency	old latency	new throughput	old throughput
u32_u64_str	bsatn	100	2.4±0.00µs	2.4±0.00µs	39.6 Mtx/sec	39.0 Mtx/sec
u32_u64_str	json	100	5.2±0.04µs	5.0±0.04µs	18.2 Mtx/sec	19.3 Mtx/sec
u32_u64_str	product_value	100	623.8±1.08ns	651.3±0.91ns	152.9 Mtx/sec	146.4 Mtx/sec
u32_u64_u64	bsatn	100	1690.0±28.78ns	1722.3±37.03ns	56.4 Mtx/sec	55.4 Mtx/sec
u32_u64_u64	json	100	3.4±0.07µs	3.4±0.09µs	28.4 Mtx/sec	27.9 Mtx/sec
u32_u64_u64	product_value	100	598.0±0.40ns	603.2±0.27ns	159.5 Mtx/sec	158.1 Mtx/sec

stdb_module_large_arguments

arg size	new latency	old latency	new throughput	old throughput
64KiB	78.7±4.33µs	96.2±6.65µs	-	-

stdb_module_print_bulk

line count	new latency	old latency	new throughput	old throughput
1	39.5±8.01µs	43.2±5.37µs	-	-
100	346.6±5.52µs	359.8±7.97µs	-	-
1000	2.9±0.15ms	3.1±0.10ms	-	-

remaining

name	new latency	old latency	new throughput	old throughput
sqlite/💿/update_bulk/u32_u64_str/unique_0/load=2048/count=256	-	45.9±0.29µs	-	21.3 Ktx/sec
sqlite/💿/update_bulk/u32_u64_u64/unique_0/load=2048/count=256	-	40.6±0.19µs	-	24.1 Ktx/sec
sqlite/🧠/update_bulk/u32_u64_str/unique_0/load=2048/count=256	-	38.3±0.31µs	-	25.5 Ktx/sec
sqlite/🧠/update_bulk/u32_u64_u64/unique_0/load=2048/count=256	-	34.8±0.29µs	-	28.0 Ktx/sec
stdb_module/💿/update_bulk/u32_u64_str/unique_0/load=2048/count=256	3.1±0.04ms	3.1±0.01ms	319 tx/sec	321 tx/sec
stdb_module/💿/update_bulk/u32_u64_u64/unique_0/load=2048/count=256	2.0±0.01ms	2.2±0.02ms	490 tx/sec	462 tx/sec
stdb_raw/💿/update_bulk/u32_u64_str/unique_0/load=2048/count=256	1108.9±0.59µs	1121.6±1.34µs	901 tx/sec	891 tx/sec
stdb_raw/💿/update_bulk/u32_u64_u64/unique_0/load=2048/count=256	744.5±0.66µs	759.1±0.62µs	1343 tx/sec	1317 tx/sec
stdb_raw/🧠/update_bulk/u32_u64_str/unique_0/load=2048/count=256	784.8±0.52µs	795.9±0.53µs	1274 tx/sec	1256 tx/sec
stdb_raw/🧠/update_bulk/u32_u64_u64/unique_0/load=2048/count=256	552.6±0.40µs	564.2±0.98µs	1809 tx/sec	1772 tx/sec

[github-actions]

Callgrind benchmark results

Callgrind Benchmark Report

These benchmarks were run using callgrind,
an instruction-level profiler. They allow comparisons between sqlite (sqlite), SpacetimeDB running through a module (stdb_module), and the underlying SpacetimeDB data storage engine (stdb_raw). Callgrind emulates a CPU to collect the below estimates.

Measurement changes larger than five percent are in bold.

In-memory benchmarks

callgrind: empty transaction

db	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
stdb_raw	6154	6171	-0.28%	6986	6943	0.62%
sqlite	5568	5558	0.18%	6100	6072	0.46%

callgrind: filter

db	schema	indices	count	preload	_column	data_type	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
stdb_raw	u32_u64_str	no_index	64	128	1	u64	84053	91094	-7.73%	84997	92180	-7.79%
stdb_raw	u32_u64_str	no_index	64	128	2	string	126745	132378	-4.26%	128009	133838	-4.36%
stdb_raw	u32_u64_str	btree_each_column	64	128	2	string	29442	29443	-0.00%	30638	30779	-0.46%
stdb_raw	u32_u64_str	btree_each_column	64	128	1	u64	28430	28401	0.10%	29398	29499	-0.34%
sqlite	u32_u64_str	no_index	64	128	1	u64	118033	118033	0.00%	119267	119439	-0.14%
sqlite	u32_u64_str	no_index	64	128	2	string	138612	138612	0.00%	140118	140264	-0.10%
sqlite	u32_u64_str	btree_each_column	64	128	2	string	128475	128475	0.00%	130187	130249	-0.05%
sqlite	u32_u64_str	btree_each_column	64	128	1	u64	125350	125350	0.00%	126906	126834	0.06%

callgrind: insert bulk

db	schema	indices	count	preload	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
stdb_raw	u32_u64_str	unique_0	64	128	1464239	1484619	-1.37%	1492735	1513867	-1.40%
stdb_raw	u32_u64_str	btree_each_column	64	128	1605629	1631435	-1.58%	1646437	1679011	-1.94%
sqlite	u32_u64_str	unique_0	64	128	396372	396342	0.01%	414614	420392	-1.37%
sqlite	u32_u64_str	btree_each_column	64	128	981664	981664	0.00%	1019010	1027218	-0.80%

callgrind: iterate

db	schema	indices	count	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
stdb_raw	u32_u64_str	unique_0	1024	179195	179196	-0.00%	179627	179696	-0.04%
stdb_raw	u32_u64_str	unique_0	64	18805	18806	-0.01%	19125	19182	-0.30%
sqlite	u32_u64_str	unique_0	1024	1044679	1044669	0.00%	1047857	1047835	0.00%
sqlite	u32_u64_str	unique_0	64	74741	74741	0.00%	75917	75977	-0.08%

callgrind: serialize_product_value

count	format	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
64	json	49521	49521	0.00%	52207	52143	0.12%
64	bsatn	26627	26627	0.00%	28905	28905	0.00%
16	json	12643	12643	0.00%	14581	14513	0.47%
16	bsatn	8357	8357	0.00%	9683	9717	-0.35%

callgrind: update bulk

db	schema	indices	count	preload	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
stdb_raw	u32_u64_str	unique_0	1024	1024	46256448	46821607	-1.21%	47592230	48311273	-1.49%
stdb_raw	u32_u64_str	unique_0	64	128	2709207	2744552	-1.29%	2800241	2845820	-1.60%
sqlite	u32_u64_str	unique_0	1024	1024	1801965	1801965	0.00%	1811333	1811231	0.01%
sqlite	u32_u64_str	unique_0	64	128	128479	128479	0.00%	131427	131369	0.04%

On-disk benchmarks

callgrind: empty transaction

db	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
stdb_raw	6396	6405	-0.14%	7252	7177	1.05%
sqlite	5600	5600	0.00%	6128	6144	-0.26%

callgrind: filter

db	schema	indices	count	preload	_column	data_type	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
stdb_raw	u32_u64_str	no_index	64	128	1	u64	84295	91336	-7.71%	85259	92494	-7.82%
stdb_raw	u32_u64_str	no_index	64	128	2	string	126987	132620	-4.25%	128495	134280	-4.31%
stdb_raw	u32_u64_str	btree_each_column	64	128	1	u64	28642	28643	-0.00%	29630	29733	-0.35%
stdb_raw	u32_u64_str	btree_each_column	64	128	2	string	29684	29685	-0.00%	30872	30969	-0.31%
sqlite	u32_u64_str	no_index	64	128	1	u64	119954	119954	0.00%	121532	121660	-0.11%
sqlite	u32_u64_str	no_index	64	128	2	string	140533	140533	0.00%	142399	142493	-0.07%
sqlite	u32_u64_str	btree_each_column	64	128	2	string	130525	130525	0.00%	132587	132545	0.03%
sqlite	u32_u64_str	btree_each_column	64	128	1	u64	127446	127446	0.00%	129360	129284	0.06%

callgrind: insert bulk

db	schema	indices	count	preload	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
stdb_raw	u32_u64_str	unique_0	64	128	2274331	2294563	-0.88%	2304955	2325197	-0.87%
stdb_raw	u32_u64_str	btree_each_column	64	128	2424892	2448052	-0.95%	2466430	2493708	-1.09%
sqlite	u32_u64_str	unique_0	64	128	413848	413848	0.00%	431508	437264	-1.32%
sqlite	u32_u64_str	btree_each_column	64	128	1019854	1019854	0.00%	1058018	1066370	-0.78%

callgrind: iterate

db	schema	indices	count	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
stdb_raw	u32_u64_str	unique_0	1024	179437	179438	-0.00%	179889	179998	-0.06%
stdb_raw	u32_u64_str	unique_0	64	19047	19048	-0.01%	19363	19480	-0.60%
sqlite	u32_u64_str	unique_0	1024	1047737	1047737	0.00%	1051503	1051523	-0.00%
sqlite	u32_u64_str	unique_0	64	76507	76507	0.00%	77857	77857	0.00%

callgrind: serialize_product_value

count	format	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
64	json	49521	49521	0.00%	52207	52143	0.12%
64	bsatn	26627	26627	0.00%	28905	28905	0.00%
16	json	12643	12643	0.00%	14581	14513	0.47%
16	bsatn	8357	8357	0.00%	9683	9717	-0.35%

callgrind: update bulk

db	schema	indices	count	preload	total reads + writes	old total reads + writes	Δrw	estimated cycles	old estimated cycles	Δcycles
stdb_raw	u32_u64_str	unique_0	1024	1024	71503262	72071065	-0.79%	73169820	73908125	-1.00%
stdb_raw	u32_u64_str	unique_0	64	128	4147562	4183139	-0.85%	4248708	4294189	-1.06%
sqlite	u32_u64_str	unique_0	1024	1024	1809726	1809726	0.00%	1818670	1818596	0.00%
sqlite	u32_u64_str	unique_0	64	128	132627	132627	0.00%	135811	135701	0.08%

[joshua-spacetime]

full-scan               time:   [59.269 ms 59.444 ms 59.639 ms]
                        change: [-28.386% -28.129% -27.845%] (p = 0.00 < 0.05)
                        Performance has improved.

full-join               time:   [210.61 µs 210.80 µs 211.01 µs]
                        change: [-13.686% -13.497% -13.312%] (p = 0.00 < 0.05)
                        Performance has improved.

[Centril]

I'll review this on Monday :)

[Centril]

🚀