test(compile): tighten the compile wall, memory and delta budgets to CI measurements

Each absolute budget is now about 1.2x the slowest of two CI runs of this
branch, and never below 1.15x the slowest recent CI run on the same code (runs
based on 0.158.2 or later). That covers the fresh-process, cache, varied-cache,
dense cold, warm and edit, and delta edit-loop wall budgets, the delta CPU
ratios and the peak RSS limits. A limit already below that floor is
unchanged. The fallback CPU limit, the 4 GiB cgroup and the same-runner ratio
limits are unchanged.

Author: kvlonge

tests/e2e/src/sqlbuild/cli/commands/main/compile/test_dense_compile_performance.py

@@ -53,28 +53,28 @@
         DenseCompileGuardTestCase(
             "dense_models_1000_all_rules",
             1000,
-            7.5,
+            6.0,
             800 * _MIB,
             "de23b14d87e954a65c2da91918fce062d158419edd63cc8cf3dc938b3e47cd44",
         ),
         DenseCompileGuardTestCase(
             "dense_models_3000_all_rules",
             3000,
-            16.5,
+            12.0,
             3 * _GIB // 2,
             "196fca3e87989eb3e89e443b7ca93bb094198538d5bde32b9d6fcca58ffc8eed",
         ),
         DenseCompileGuardTestCase(
             "dense_models_5000_all_rules",
             5000,
-            26.0,
+            18.0,
             9 * _GIB // 4,
             "3477a472fa76d6236ebd0c9758d79c18f67f1bf55e9bbec05b002ba3744cfd22",
         ),
         DenseCompileGuardTestCase(
             "dense_models_10000_all_rules",
             10000,
-            51.0,
+            35.0,
             15 * _GIB // 4,
             "f2a574af6e8068a0634fa3c078b168dff9a475ada5d71502482ad2926c90e1c3",
         ),
@@ -154,8 +154,8 @@ def test_given_dense_project_when_compiling_cold_then_preserves_rules_semantics_
             "dense_models_3000_warm_and_one_edit",
             3000,
             1521,
-            8.0,
-            8.5,
+            6.0,
+            6.0,
             3 * _GIB // 2,
             "196fca3e87989eb3e89e443b7ca93bb094198538d5bde32b9d6fcca58ffc8eed",
             "1afe754dd8c66bc36e38a40c66e9183ecbaf365fd8d87a784069cd615c3ce64a",
@@ -165,8 +165,8 @@ def test_given_dense_project_when_compiling_cold_then_preserves_rules_semantics_
             "dense_models_5000_warm_and_one_edit",
             5000,
             2521,
-            13.0,
-            13.5,
+            9.0,
+            9.5,
             9 * _GIB // 4,
             "3477a472fa76d6236ebd0c9758d79c18f67f1bf55e9bbec05b002ba3744cfd22",
             "dfbada4c59a3ab8ec7733784890d74c7b64e9b0adcb03364baaabef5bdba5574",
@@ -240,9 +240,9 @@ def test_given_dense_project_when_compiling_warm_and_after_one_edit_then_matches
             model_count=3000,
             edited_model_index=1521,
             full_pairs=4,
-            expected_max_full_seconds=15.0,
-            expected_max_delta_seconds=4.5,
-            expected_max_delta_cpu_ratio=0.375,
+            expected_max_full_seconds=7.0,
+            expected_max_delta_seconds=2.0,
+            expected_max_delta_cpu_ratio=0.30,
             expected_max_fallback_cpu_ratio=1.05,
             expected_max_rss_bytes=3 * _GIB // 2,
         ),
@@ -251,9 +251,9 @@ def test_given_dense_project_when_compiling_warm_and_after_one_edit_then_matches
             model_count=10000,
             edited_model_index=5021,
             full_pairs=3,
-            expected_max_full_seconds=60.0,
-            expected_max_delta_seconds=9.0,
-            expected_max_delta_cpu_ratio=0.30,
+            expected_max_full_seconds=28.0,
+            expected_max_delta_seconds=8.0,
+            expected_max_delta_cpu_ratio=0.275,
             expected_max_fallback_cpu_ratio=1.05,
             expected_max_rss_bytes=15 * _GIB // 4,
         ),

tests/e2e/src/sqlbuild/cli/commands/main/compile/test_fresh_process_compile_cache_performance.py

@@ -50,14 +50,14 @@
             macro_count=500,
             test_count=2_945,
             audit_count=5_056,
-            expected_cold_max_wall_seconds=10.5,
-            expected_warm_max_wall_seconds=6.0,
-            expected_edit_max_wall_seconds=6.0,
+            expected_cold_max_wall_seconds=9.0,
+            expected_warm_max_wall_seconds=4.5,
+            expected_edit_max_wall_seconds=4.5,
             expected_max_warm_to_cold_ratio=_MAX_WARM_TO_COLD_RATIO,
             expected_max_edit_to_cold_ratio=_MAX_EDIT_TO_COLD_RATIO,
             expected_max_cache_write_cpu_overhead_ratio=_MAX_CACHE_WRITE_CPU_OVERHEAD_RATIO,
             expected_max_cache_write_wall_overhead_ratio=_MAX_CACHE_WRITE_WALL_OVERHEAD_RATIO,
-            expected_max_rss_bytes=2 * _GIB,
+            expected_max_rss_bytes=1152 * _MIB,
             expected_max_cache_bytes=96 * _MIB,
             expected_cold_fingerprint=(
                 "1914b788abcb452422eaf6c3a2c240ff390bab9530217285a5ca93334db7b6c3"
@@ -84,14 +84,14 @@
             macro_count=834,
             test_count=4_908,
             audit_count=8_427,
-            expected_cold_max_wall_seconds=18.0,
-            expected_warm_max_wall_seconds=10.5,
-            expected_edit_max_wall_seconds=10.5,
+            expected_cold_max_wall_seconds=14.5,
+            expected_warm_max_wall_seconds=7.0,
+            expected_edit_max_wall_seconds=7.0,
             expected_max_warm_to_cold_ratio=_MAX_WARM_TO_COLD_RATIO,
             expected_max_edit_to_cold_ratio=_MAX_EDIT_TO_COLD_RATIO,
             expected_max_cache_write_cpu_overhead_ratio=_MAX_CACHE_WRITE_CPU_OVERHEAD_RATIO,
             expected_max_cache_write_wall_overhead_ratio=_MAX_CACHE_WRITE_WALL_OVERHEAD_RATIO,
-            expected_max_rss_bytes=2 * _GIB,
+            expected_max_rss_bytes=1664 * _MIB,
             expected_max_cache_bytes=160 * _MIB,
             expected_cold_fingerprint=(
                 "2f1e5dfd29025c25e0dffc6e74f4c2be020a5ef133ecd081f44e0c55c24858b4"
@@ -118,15 +118,15 @@
             macro_count=1_667,
             test_count=9_816,
             audit_count=16_855,
-            expected_cold_max_wall_seconds=34.5,
-            expected_warm_max_wall_seconds=19.5,
-            expected_edit_max_wall_seconds=19.5,
+            expected_cold_max_wall_seconds=28.0,
+            expected_warm_max_wall_seconds=14.0,
+            expected_edit_max_wall_seconds=13.5,
             expected_max_warm_to_cold_ratio=_MAX_WARM_TO_COLD_RATIO,
             expected_max_edit_to_cold_ratio=_MAX_EDIT_TO_COLD_RATIO,
             expected_max_cache_write_cpu_overhead_ratio=_MAX_LARGE_CACHE_WRITE_CPU_OVERHEAD_RATIO,
             expected_max_cache_write_wall_overhead_ratio=_MAX_CACHE_WRITE_WALL_OVERHEAD_RATIO,
             # Temporary memory exception (2026-10-10); remove after conversion optimisation.
-            expected_max_rss_bytes=3 * _GIB,
+            expected_max_rss_bytes=3008 * _MIB,
             expected_max_cache_bytes=320 * _MIB,
             expected_cold_fingerprint=(
                 "ecdd7f2cf961fab739aa806b6d90283916cffad2007e1e2f3ba90a466db6cb75"

tests/e2e/src/sqlbuild/cli/commands/main/compile/test_fresh_process_compile_performance.py

@@ -19,7 +19,8 @@
 )
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
-_GIB: int = 1024 * 1024 * 1024
+_MIB: int = 1024 * 1024
+_GIB: int = 1024 * _MIB
 
 
 @pytest.mark.performance
@@ -36,8 +37,8 @@
             macro_count=37,
             test_count=2_945,
             audit_count=5_056,
-            expected_max_wall_seconds=9.5,
-            expected_max_rss_bytes=int(1.5 * _GIB),
+            expected_max_wall_seconds=8.0,
+            expected_max_rss_bytes=1088 * _MIB,
             expected_semantic_fingerprint=(
                 "d935821df54043fabc3a6911c1926578178c9b744f2a589cbef99a10d5207543"
             ),
@@ -51,8 +52,8 @@
             macro_count=61,
             test_count=4_908,
             audit_count=8_427,
-            expected_max_wall_seconds=16.0,
-            expected_max_rss_bytes=int(1.75 * _GIB),
+            expected_max_wall_seconds=12.5,
+            expected_max_rss_bytes=1664 * _MIB,
             expected_semantic_fingerprint=(
                 "874cbfd473846c6a335ec2f79cb6b7b0b77e8e8d58652815fad1e9d59d423c46"
             ),
@@ -66,7 +67,7 @@
             macro_count=123,
             test_count=9_816,
             audit_count=16_855,
-            expected_max_wall_seconds=31.0,
+            expected_max_wall_seconds=25.5,
             expected_max_rss_bytes=int(2.5 * _GIB),
             expected_semantic_fingerprint=(
                 "5ccdabeaabfdfba1ecbae80259dd3c6eb801d683df140b7fb7a249b9555308c6"

tests/e2e/src/sqlbuild/cli/commands/main/compile/test_varied_compile_cache_performance.py

@@ -31,10 +31,10 @@
         VariedCompileCacheTestCase(
             "models_1000_varied_shared_graph",
             1000,
-            24.0,
-            10.0,
-            13.0,
-            2 * 1024**3,
+            9.0,
+            2.5,
+            2.5,
+            832 * 1024**2,
             950,
             900,
             40,