You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Release performance: 0.186.2 (candidate) vs 0.186.2 (baseline)
Same runner: AMD EPYC 9V45 96-Core Processor, 4 CPUs; load average 3.95/3.58/2.08 before and 1.50/2.07/2.04 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
Command
Wall (s)
Wall ratio
CPU (s)
CPU ratio
Peak RSS (MiB)
RSS ratio
Result
compile (no cache)
3.80 → 5.30
1.395x
8.31 → 11.88
1.430x
923 → 1010
1.094x
❌ regressed
compile (warm cache)
0.18 → 0.19
1.056x
0.12 → 0.11
0.917x
71 → 71
1.005x
✅ ok
plan --json
5.49 → 5.55
1.011x
7.27 → 7.33
1.008x
1004 → 1023
1.019x
✅ ok
lineage hub downstream
0.19 → 0.17
0.895x
0.26 → 0.25
0.962x
157 → 162
1.032x
✅ ok
lineage rollup upstream
0.19 → 0.17
0.895x
0.26 → 0.25
0.962x
156 → 171
1.091x
✅ ok
lineage column trace
1.87 → 1.88
1.005x
2.57 → 2.62
1.019x
920 → 916
0.997x
✅ ok
dag --json
2.52 → 2.51
0.996x
3.21 → 3.17
0.988x
972 → 976
1.003x
✅ ok
scope --json
0.19 → 0.19
1.000x
0.18 → 0.17
0.944x
118 → 124
1.048x
✅ ok
compile (one-model edit)
0.81 → 0.83
1.025x
0.69 → 0.70
1.014x
359 → 358
0.997x
✅ ok
dense compile (warm cache)
0.31 → 0.30
0.968x
0.18 → 0.19
1.056x
94 → 94
1.001x
✅ ok
dense compile (one-model edit)
1.25 → 1.20
0.960x
1.06 → 1.05
0.991x
434 → 440
1.014x
✅ ok
dense compile (no cache)
6.45 → 6.49
1.006x
15.87 → 15.77
0.994x
1368 → 1418
1.036x
✅ ok
build (empty warehouse)
30.14 → 30.14
1.000x
30.25 → 30.24
1.000x
1732 → 1746
1.008x
✅ ok
Result: failed. compile (no cache): wall 1.395x exceeds 1.15x; compile (no cache): CPU 1.430x exceeds 1.15x.
Release performance: 0.186.2 (candidate) vs 0.186.2 (baseline)
Same runner: AMD EPYC 7763 64-Core Processor, 4 CPUs; load average 3.69/3.79/2.67 before and 1.24/1.62/2.14 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
Command
Wall (s)
Wall ratio
CPU (s)
CPU ratio
Peak RSS (MiB)
RSS ratio
Result
compile (no cache)
6.28 → 8.58
1.366x
15.12 → 21.66
1.433x
918 → 1005
1.095x
❌ regressed
compile (warm cache)
0.17 → 0.17
1.000x
0.16 → 0.16
1.000x
71 → 71
0.999x
✅ ok
plan --json
8.60 → 8.74
1.016x
12.11 → 12.20
1.007x
1001 → 1015
1.014x
✅ ok
lineage hub downstream
0.25 → 0.25
1.000x
0.37 → 0.37
1.000x
155 → 162
1.044x
✅ ok
lineage rollup upstream
0.25 → 0.25
1.000x
0.37 → 0.37
1.000x
156 → 166
1.067x
✅ ok
lineage column trace
2.90 → 2.90
1.000x
4.27 → 4.28
1.002x
912 → 919
1.008x
✅ ok
dag --json
3.65 → 3.64
0.997x
5.20 → 5.16
0.992x
985 → 998
1.014x
✅ ok
scope --json
0.27 → 0.27
1.000x
0.26 → 0.26
1.000x
111 → 120
1.081x
✅ ok
compile (one-model edit)
0.91 → 0.93
1.022x
0.99 → 1.01
1.020x
365 → 370
1.014x
✅ ok
dense compile (warm cache)
0.26 → 0.26
1.000x
0.26 → 0.25
0.962x
94 → 95
1.002x
✅ ok
dense compile (one-model edit)
1.30 → 1.30
1.000x
1.56 → 1.52
0.974x
438 → 429
0.979x
✅ ok
dense compile (no cache)
10.28 → 10.31
1.003x
28.02 → 27.92
0.996x
1371 → 1383
1.009x
✅ ok
build (empty warehouse)
50.75 → 50.58
0.997x
53.06 → 52.86
0.996x
1729 → 1735
1.003x
✅ ok
Result: failed. compile (no cache): wall 1.366x exceeds 1.15x; compile (no cache): CPU 1.433x exceeds 1.15x.
Release performance: 0.186.2 (candidate) vs 0.186.2 (baseline)
Same runner: AMD EPYC 9V74 80-Core Processor, 4 CPUs; load average 4.10/3.86/2.65 before and 1.13/1.52/2.05 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
Command
Wall (s)
Wall ratio
CPU (s)
CPU ratio
Peak RSS (MiB)
RSS ratio
Result
compile (no cache)
6.40 → 8.58
1.341x
14.97 → 21.18
1.415x
918 → 1012
1.103x
❌ regressed
compile (warm cache)
0.17 → 0.17
1.000x
0.17 → 0.17
1.000x
70 → 71
1.004x
✅ ok
plan --json
8.79 → 8.89
1.011x
12.13 → 12.17
1.003x
1008 → 1015
1.006x
✅ ok
lineage hub downstream
0.26 → 0.26
1.000x
0.38 → 0.38
1.000x
155 → 163
1.051x
✅ ok
lineage rollup upstream
0.26 → 0.26
1.000x
0.38 → 0.38
1.000x
162 → 162
1.000x
✅ ok
lineage column trace
2.97 → 3.00
1.010x
4.28 → 4.32
1.009x
920 → 912
0.992x
✅ ok
dag --json
3.76 → 3.78
1.005x
5.25 → 5.25
1.000x
977 → 977
1.000x
✅ ok
scope --json
0.30 → 0.30
1.000x
0.28 → 0.28
1.000x
109 → 121
1.105x
✅ ok
compile (one-model edit)
0.94 → 0.97
1.032x
1.04 → 1.04
1.000x
366 → 385
1.053x
✅ ok
dense compile (warm cache)
0.26 → 0.26
1.000x
0.27 → 0.26
0.963x
94 → 94
0.999x
✅ ok
dense compile (one-model edit)
1.36 → 1.34
0.985x
1.59 → 1.56
0.981x
462 → 444
0.961x
✅ ok
dense compile (no cache)
10.31 → 10.30
0.999x
27.49 → 27.47
0.999x
1345 → 1406
1.045x
✅ ok
build (empty warehouse)
49.45 → 49.36
0.998x
51.84 → 51.88
1.001x
1735 → 1737
1.001x
✅ ok
Result: failed. compile (no cache): wall 1.341x exceeds 1.15x; compile (no cache): CPU 1.415x exceeds 1.15x.
Release performance: 0.186.2 (candidate) vs 0.186.2 (baseline)
Same runner: INTEL(R) XEON(R) PLATINUM 8573C, 4 CPUs; load average 4.71/4.14/2.84 before and 1.34/1.57/2.07 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
Command
Wall (s)
Wall ratio
CPU (s)
CPU ratio
Peak RSS (MiB)
RSS ratio
Result
compile (no cache)
5.76 → 7.44
1.292x
13.54 → 18.62
1.375x
911 → 981
1.077x
❌ regressed
compile (warm cache)
0.22 → 0.22
1.000x
0.13 → 0.13
1.000x
71 → 71
1.001x
✅ ok
plan --json
8.19 → 8.29
1.012x
11.08 → 11.17
1.008x
1059 → 1017
0.960x
✅ ok
lineage hub downstream
0.25 → 0.25
1.000x
0.35 → 0.35
1.000x
155 → 166
1.071x
✅ ok
lineage rollup upstream
0.25 → 0.26
1.040x
0.36 → 0.36
1.000x
154 → 171
1.108x
✅ ok
lineage column trace
2.70 → 2.67
0.989x
3.92 → 3.91
0.997x
898 → 918
1.022x
✅ ok
dag --json
3.51 → 3.52
1.003x
4.74 → 4.73
0.998x
946 → 949
1.004x
✅ ok
scope --json
0.26 → 0.25
0.962x
0.25 → 0.25
1.000x
108 → 120
1.115x
✅ ok
compile (one-model edit)
1.04 → 1.08
1.038x
0.95 → 0.95
1.000x
362 → 384
1.061x
✅ ok
dense compile (warm cache)
0.31 → 0.28
0.903x
0.19 → 0.19
1.000x
94 → 94
1.000x
✅ ok
dense compile (one-model edit)
1.48 → 1.46
0.986x
1.41 → 1.39
0.986x
420 → 444
1.058x
✅ ok
dense compile (no cache)
8.92 → 8.86
0.993x
23.52 → 23.97
1.019x
1344 → 1404
1.044x
✅ ok
build (empty warehouse)
55.25 → 55.15
0.998x
55.13 → 55.28
1.003x
1736 → 1735
0.999x
✅ ok
Result: failed. compile (no cache): wall 1.292x exceeds 1.15x; compile (no cache): CPU 1.375x exceeds 1.15x.
Release performance: 0.186.2 (candidate) vs 0.186.2 (baseline)
Same runner: AMD EPYC 9V74 80-Core Processor, 4 CPUs; load average 4.33/4.01/2.69 before and 1.28/1.57/2.09 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
Command
Wall (s)
Wall ratio
CPU (s)
CPU ratio
Peak RSS (MiB)
RSS ratio
Result
compile (no cache)
6.60 → 8.93
1.353x
15.31 → 21.52
1.406x
909 → 961
1.057x
❌ regressed
compile (warm cache)
0.24 → 0.23
0.958x
0.17 → 0.17
1.000x
70 → 71
1.003x
✅ ok
plan --json
9.26 → 9.58
1.035x
12.48 → 12.87
1.031x
1012 → 1020
1.008x
✅ ok
lineage hub downstream
0.28 → 0.28
1.000x
0.40 → 0.40
1.000x
154 → 161
1.046x
✅ ok
lineage rollup upstream
0.28 → 0.26
0.929x
0.39 → 0.39
1.000x
153 → 165
1.081x
✅ ok
lineage column trace
3.14 → 3.19
1.016x
4.45 → 4.44
0.998x
902 → 920
1.020x
✅ ok
dag --json
4.03 → 4.07
1.010x
5.34 → 5.28
0.989x
942 → 956
1.015x
✅ ok
scope --json
0.30 → 0.28
0.933x
0.27 → 0.26
0.963x
109 → 120
1.103x
✅ ok
compile (one-model edit)
1.11 → 1.12
1.009x
1.01 → 1.02
1.010x
369 → 374
1.014x
✅ ok
dense compile (warm cache)
0.46 → 0.50
1.087x
0.26 → 0.26
1.000x
94 → 94
0.999x
✅ ok
dense compile (one-model edit)
1.63 → 1.57
0.963x
1.51 → 1.50
0.993x
422 → 423
1.004x
✅ ok
dense compile (no cache)
9.91 → 9.68
0.977x
26.22 → 26.07
0.994x
1342 → 1400
1.043x
✅ ok
build (empty warehouse)
66.80 → 63.20
0.946x
52.74 → 52.69
0.999x
1735 → 1734
0.999x
✅ ok
Result: failed. compile (no cache): wall 1.353x exceeds 1.15x; compile (no cache): CPU 1.406x exceeds 1.15x.
Release performance: 0.187.0 (candidate) vs 0.186.2 (baseline)
Same runner: AMD EPYC 9V45 96-Core Processor, 4 CPUs; load average 3.94/3.49/2.03 before and 1.23/1.74/1.86 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
Command
Wall (s)
Wall ratio
CPU (s)
CPU ratio
Peak RSS (MiB)
RSS ratio
Result
compile (no cache)
4.22 → 5.78
1.370x
9.15 → 13.58
1.484x
915 → 952
1.041x
❌ regressed
compile (warm cache)
0.20 → 0.18
0.900x
0.12 → 0.12
1.000x
71 → 71
1.001x
✅ ok
plan --json
5.98 → 6.09
1.018x
7.77 → 7.98
1.027x
1007 → 1024
1.018x
✅ ok
lineage hub downstream
0.20 → 0.19
0.950x
0.27 → 0.26
0.963x
155 → 161
1.040x
✅ ok
lineage rollup upstream
0.20 → 0.20
1.000x
0.27 → 0.26
0.963x
153 → 170
1.109x
✅ ok
lineage column trace
2.01 → 1.97
0.980x
2.74 → 2.69
0.982x
920 → 923
1.004x
✅ ok
dag --json
2.60 → 2.62
1.008x
3.24 → 3.32
1.025x
956 → 968
1.012x
✅ ok
scope --json
0.19 → 0.19
1.000x
0.17 → 0.18
1.059x
117 → 124
1.057x
✅ ok
compile (one-model edit)
0.85 → 0.87
1.024x
0.70 → 0.71
1.014x
356 → 356
1.001x
✅ ok
dense compile (warm cache)
0.28 → 0.28
1.000x
0.20 → 0.19
0.950x
94 → 94
1.001x
✅ ok
dense compile (one-model edit)
1.24 → 1.27
1.024x
1.10 → 1.12
1.018x
436 → 432
0.991x
✅ ok
dense compile (no cache)
6.63 → 6.37
0.961x
16.89 → 16.47
0.975x
1395 → 1410
1.011x
✅ ok
build (empty warehouse)
32.74 → 31.70
0.968x
32.13 → 31.90
0.993x
1736 → 1738
1.001x
✅ ok
Result: failed. compile (no cache): wall 1.370x exceeds 1.15x; compile (no cache): CPU 1.484x exceeds 1.15x.
Release performance: 0.207.0 (candidate) vs 0.207.0 (baseline)
Same runner: AMD EPYC 9V45 96-Core Processor, 4 CPUs; load average 1.13/2.63/3.27 before and 1.26/1.80/2.61 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.207.0), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
Command
Wall (s)
Wall ratio
CPU (s)
CPU ratio
Peak RSS (MiB)
RSS ratio
Result
compile (no cache)
5.60 → 6.12
1.093x
12.79 → 15.65
1.224x
887 → 876
0.988x
❌ regressed
compile (warm cache)
0.21 → 0.20
0.952x
0.13 → 0.13
1.000x
78 → 79
1.006x
✅ ok
plan --json
6.60 → 6.84
1.036x
8.52 → 8.89
1.043x
944 → 883
0.935x
✅ ok
lineage hub downstream
0.21 → 0.21
1.000x
0.29 → 0.29
1.000x
91 → 101
1.104x
✅ ok
lineage rollup upstream
0.21 → 0.21
1.000x
0.29 → 0.29
1.000x
91 → 101
1.115x
✅ ok
lineage column trace
2.50 → 2.31
0.924x
3.28 → 3.11
0.948x
839 → 805
0.959x
✅ ok
dag --json
3.25 → 3.01
0.926x
4.00 → 3.78
0.945x
876 → 822
0.938x
✅ ok
scope --json
0.23 → 0.23
1.000x
0.23 → 0.22
0.957x
87 → 94
1.076x
✅ ok
compile (one-model edit)
1.05 → 0.99
0.943x
1.00 → 0.93
0.930x
334 → 309
0.927x
✅ ok
dense compile (warm cache)
0.29 → 0.29
1.000x
0.20 → 0.20
1.000x
102 → 102
0.998x
✅ ok
dense compile (one-model edit)
1.57 → 1.51
0.962x
1.52 → 1.45
0.954x
378 → 392
1.037x
✅ ok
dense compile (no cache)
7.00 → 7.04
1.006x
17.12 → 19.24
1.124x
1282 → 1263
0.985x
✅ ok
build (empty warehouse)
34.20 → 33.34
0.975x
33.56 → 33.50
0.998x
1651 → 1634
0.990x
✅ ok
Result: failed. compile (no cache): CPU 1.224x exceeds 1.15x.
Run https://github.com/sqlstack-io/sqlbuild/actions/runs/42825566391 (main at c642221, after #1397) flagged compile (no cache) CPU at 1.224x against 0.207.0 (12.79 s -> 15.65 s; wall 1.093x, peak RSS 0.988x). dense compile (no cache) CPU is 1.124x, within its limit, and every other command is at or below 1.0x.
This is the known cost of moving model analysis onto SQLQuack's native parsers in #1397. It was accepted by the maintainers as a temporary allowance when the cutover was approved, and the analysis-sharing change in the same PR already reduced it (same-machine A/B on the 3,000-model compiler-only case: cold CPU 17.46 s -> 15.62 s). A follow-up to bring cold no-cache CPU back within 1.10x of the previous release is tracked; the comparison limits themselves are unchanged.
Release performance: 0.207.0 (candidate) vs 0.207.0 (baseline)
Same runner: AMD EPYC 9V74 80-Core Processor, 4 CPUs; load average 0.98/2.28/3.14 before and 1.14/1.50/2.29 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.207.0), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
Command
Wall (s)
Wall ratio
CPU (s)
CPU ratio
Peak RSS (MiB)
RSS ratio
Result
compile (no cache)
7.24 → 7.69
1.062x
17.32 → 20.18
1.165x
891 → 868
0.975x
❌ regressed
compile (warm cache)
0.20 → 0.22
1.100x
0.14 → 0.14
1.000x
78 → 78
0.996x
✅ ok
plan --json
7.57 → 7.30
0.964x
10.22 → 9.97
0.976x
944 → 880
0.932x
✅ ok
lineage hub downstream
0.22 → 0.22
1.000x
0.32 → 0.32
1.000x
92 → 101
1.100x
✅ ok
lineage rollup upstream
0.22 → 0.22
1.000x
0.31 → 0.32
1.032x
92 → 101
1.096x
✅ ok
lineage column trace
2.71 → 2.54
0.937x
3.78 → 3.61
0.955x
840 → 804
0.957x
✅ ok
dag --json
3.51 → 3.28
0.934x
4.58 → 4.38
0.956x
879 → 824
0.937x
✅ ok
scope --json
0.24 → 0.24
1.000x
0.23 → 0.23
1.000x
88 → 94
1.071x
✅ ok
compile (one-model edit)
1.13 → 1.10
0.973x
1.06 → 1.01
0.953x
329 → 312
0.949x
✅ ok
dense compile (warm cache)
0.30 → 0.33
1.100x
0.21 → 0.21
1.000x
101 → 100
0.996x
✅ ok
dense compile (one-model edit)
1.59 → 1.61
1.013x
1.57 → 1.56
0.994x
375 → 384
1.025x
✅ ok
dense compile (no cache)
8.93 → 8.86
0.992x
22.87 → 25.92
1.133x
1276 → 1243
0.974x
✅ ok
build (empty warehouse)
44.14 → 43.88
0.994x
43.49 → 43.34
0.997x
1649 → 1626
0.986x
✅ ok
Result: failed. compile (no cache): CPU 1.165x exceeds 1.15x.
Closing: the last flagged regression (compile without cache, CPU 1.22x at 0.207.0) is the known cost of moving model analysis onto SQLQuack's native parsers, explained above, and later release comparisons up to 0.211.0 passed. The release performance workflow opens a new issue if a regression comes back.
Activity
github-actions commentedon Nov 1, 2026
Release performance regressed at 45f102e on main.
Comparison run: https://github.com/sqlstack-io/sqlbuild/actions/runs/41327968958
Release performance: 0.186.2 (candidate) vs 0.186.2 (baseline)
Same runner: AMD EPYC 7763 64-Core Processor, 4 CPUs; load average 3.69/3.79/2.67 before and 1.24/1.62/2.14 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
compile (no cache)compile (warm cache)plan --jsonlineage hub downstreamlineage rollup upstreamlineage column tracedag --jsonscope --jsoncompile (one-model edit)dense compile (warm cache)dense compile (one-model edit)dense compile (no cache)build (empty warehouse)Result: failed. compile (no cache): wall 1.366x exceeds 1.15x; compile (no cache): CPU 1.433x exceeds 1.15x.
github-actions commentedon Nov 1, 2026
Release performance regressed at 69bf510 on main.
Comparison run: https://github.com/sqlstack-io/sqlbuild/actions/runs/41328954763
Release performance: 0.186.2 (candidate) vs 0.186.2 (baseline)
Same runner: AMD EPYC 9V74 80-Core Processor, 4 CPUs; load average 4.10/3.86/2.65 before and 1.13/1.52/2.05 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
compile (no cache)compile (warm cache)plan --jsonlineage hub downstreamlineage rollup upstreamlineage column tracedag --jsonscope --jsoncompile (one-model edit)dense compile (warm cache)dense compile (one-model edit)dense compile (no cache)build (empty warehouse)Result: failed. compile (no cache): wall 1.341x exceeds 1.15x; compile (no cache): CPU 1.415x exceeds 1.15x.
github-actions commentedon Nov 1, 2026
Release performance regressed at b66b83f on main.
Comparison run: https://github.com/sqlstack-io/sqlbuild/actions/runs/41338371912
Release performance: 0.186.2 (candidate) vs 0.186.2 (baseline)
Same runner: INTEL(R) XEON(R) PLATINUM 8573C, 4 CPUs; load average 4.71/4.14/2.84 before and 1.34/1.57/2.07 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
compile (no cache)compile (warm cache)plan --jsonlineage hub downstreamlineage rollup upstreamlineage column tracedag --jsonscope --jsoncompile (one-model edit)dense compile (warm cache)dense compile (one-model edit)dense compile (no cache)build (empty warehouse)Result: failed. compile (no cache): wall 1.292x exceeds 1.15x; compile (no cache): CPU 1.375x exceeds 1.15x.
github-actions commentedon Nov 2, 2026
Release performance regressed at 3b77897 on main.
Comparison run: https://github.com/sqlstack-io/sqlbuild/actions/runs/41375594941
Release performance: 0.186.2 (candidate) vs 0.186.2 (baseline)
Same runner: AMD EPYC 9V74 80-Core Processor, 4 CPUs; load average 4.33/4.01/2.69 before and 1.28/1.57/2.09 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
compile (no cache)compile (warm cache)plan --jsonlineage hub downstreamlineage rollup upstreamlineage column tracedag --jsonscope --jsoncompile (one-model edit)dense compile (warm cache)dense compile (one-model edit)dense compile (no cache)build (empty warehouse)Result: failed. compile (no cache): wall 1.353x exceeds 1.15x; compile (no cache): CPU 1.406x exceeds 1.15x.
github-actions commentedon Nov 2, 2026
Release performance regressed at 50382ae on main.
Comparison run: https://github.com/sqlstack-io/sqlbuild/actions/runs/41378766538
Release performance: 0.187.0 (candidate) vs 0.186.2 (baseline)
Same runner: AMD EPYC 9V45 96-Core Processor, 4 CPUs; load average 3.94/3.49/2.03 before and 1.23/1.74/1.86 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.186.2), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
compile (no cache)compile (warm cache)plan --jsonlineage hub downstreamlineage rollup upstreamlineage column tracedag --jsonscope --jsoncompile (one-model edit)dense compile (warm cache)dense compile (one-model edit)dense compile (no cache)build (empty warehouse)Result: failed. compile (no cache): wall 1.370x exceeds 1.15x; compile (no cache): CPU 1.484x exceeds 1.15x.
github-actions commentedon Nov 10, 2026
Release performance regressed at c642221 on main.
Comparison run: https://github.com/sqlstack-io/sqlbuild/actions/runs/42825566391
Release performance: 0.207.0 (candidate) vs 0.207.0 (baseline)
Same runner: AMD EPYC 9V45 96-Core Processor, 4 CPUs; load average 1.13/2.63/3.27 before and 1.26/1.80/2.61 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.207.0), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
compile (no cache)compile (warm cache)plan --jsonlineage hub downstreamlineage rollup upstreamlineage column tracedag --jsonscope --jsoncompile (one-model edit)dense compile (warm cache)dense compile (one-model edit)dense compile (no cache)build (empty warehouse)Result: failed. compile (no cache): CPU 1.224x exceeds 1.15x.
kvlonge commentedon Nov 10, 2026
Run https://github.com/sqlstack-io/sqlbuild/actions/runs/42825566391 (main at c642221, after #1397) flagged
compile (no cache)CPU at 1.224x against 0.207.0 (12.79 s -> 15.65 s; wall 1.093x, peak RSS 0.988x).dense compile (no cache)CPU is 1.124x, within its limit, and every other command is at or below 1.0x.This is the known cost of moving model analysis onto SQLQuack's native parsers in #1397. It was accepted by the maintainers as a temporary allowance when the cutover was approved, and the analysis-sharing change in the same PR already reduced it (same-machine A/B on the 3,000-model compiler-only case: cold CPU 17.46 s -> 15.62 s). A follow-up to bring cold no-cache CPU back within 1.10x of the previous release is tracked; the comparison limits themselves are unchanged.
github-actions commentedon Nov 10, 2026
Release performance regressed at 5b4cbbf on main.
Comparison run: https://github.com/sqlstack-io/sqlbuild/actions/runs/42874254488
Release performance: 0.207.0 (candidate) vs 0.207.0 (baseline)
Same runner: AMD EPYC 9V74 80-Core Processor, 4 CPUs; load average 0.98/2.28/3.14 before and 1.14/1.50/2.29 after. 5 interleaved runs per version; wall and CPU are medians, peak RSS is the worst run.
Benchmark projects are generated per side: the baseline runs projects from its own generator (v0.207.0), the candidate from the candidate's.
Limits: wall +25% (ignored under 0.5 s), CPU +25% (ignored under 0.5 s), peak RSS +25% (ignored under 32 MiB). Cold, warm and one-model edit compiles use +15% for wall and CPU.
compile (no cache)compile (warm cache)plan --jsonlineage hub downstreamlineage rollup upstreamlineage column tracedag --jsonscope --jsoncompile (one-model edit)dense compile (warm cache)dense compile (one-model edit)dense compile (no cache)build (empty warehouse)Result: failed. compile (no cache): CPU 1.165x exceeds 1.15x.
kvlonge commentedon Nov 11, 2026
Closing: the last flagged regression (compile without cache, CPU 1.22x at 0.207.0) is the known cost of moving model analysis onto SQLQuack's native parsers, explained above, and later release comparisons up to 0.211.0 passed. The release performance workflow opens a new issue if a regression comes back.