udo/uce
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Skip wasm bodies on serialized module hits

Browse Source
This commit is contained in:
udo
2026-07-16 20:31:25 +00:00
parent 11025ab8b3
commit bbfbd63b53
5 changed files with 191 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
site/demo/index.uce
+1 -1
View File
@@ -105,7 +105,7 @@ RENDER(Request& context)
print("Component phases: path ", (u64)perf["component_path_us"].to_u64(), " / artifact ", (u64)perf["component_artifact_us"].to_u64(), " / load ", (u64)perf["component_load_us"].to_u64(), " / link ", (u64)perf["component_link_us"].to_u64(), " us\n");
print("Unit loads: ", (u64)perf["unit_load_count"].to_u64(), " / module ", (u64)perf["unit_module_us"].to_u64(), " / allocate ", (u64)perf["unit_allocate_us"].to_u64(), " / imports ", (u64)perf["unit_import_us"].to_u64(), " / instantiate ", (u64)perf["unit_instantiate_us"].to_u64(), " / initialize ", (u64)perf["unit_initialize_us"].to_u64(), " us\n");
print("Unit module cache: ", (u64)perf["unit_module_cache_hit_count"].to_u64(), " worker hits / ", (u64)perf["unit_module_cache_miss_count"].to_u64(), " misses / ", (u64)perf["unit_module_serialized_cache_hit_count"].to_u64(), " serialized hits / ", (u64)perf["unit_module_compile_count"].to_u64(), " wasm compiles\n");
print("Unit module phases: lookup ", (u64)perf["unit_module_lookup_us"].to_u64(), " / read ", (u64)perf["unit_module_read_us"].to_u64(), " / parse ", (u64)perf["unit_module_parse_us"].to_u64(), " / deserialize/compile ", (u64)perf["unit_module_compile_us"].to_u64(), " / classify ", (u64)perf["unit_module_classify_us"].to_u64(), " us\n");
print("Unit module phases: lookup ", (u64)perf["unit_module_lookup_us"].to_u64(), " / read ", (u64)perf["unit_module_read_us"].to_u64(), " us / ", (u64)perf["unit_module_read_bytes"].to_u64(), " bytes / parse ", (u64)perf["unit_module_parse_us"].to_u64(), " / deserialize/compile ", (u64)perf["unit_module_compile_us"].to_u64(), " / classify ", (u64)perf["unit_module_classify_us"].to_u64(), " us\n");
print("Unit symbol resolves: ", (u64)perf["unit_symbol_resolve_count"].to_u64(), " / ", (u64)perf["unit_symbol_resolve_us"].to_u64(), " us\n");
print("Output buffer size: ", context.ob->str().length(), "\n");
?></pre>
site/doc/pages/request_perf.txt
+1 -1
View File
@@ -15,7 +15,7 @@ Component resolution is divided into `component_path_us`, `component_artifact_us
Successful first loads within the request are counted by `unit_load_count` and divided into `unit_module_us`, `unit_allocate_us`, `unit_import_us`, `unit_instantiate_us`, and `unit_initialize_us`. These cover compiled-module lookup, guest memory/table allocation, import construction, Wasmtime instantiation, and relocations/constructors/request binding. Repeated handlers from an already loaded unit are excluded.
`unit_module_cache_hit_count` and `unit_module_cache_miss_count` divide module loads by the worker's compiled-module cache. A miss is further identified by `unit_module_serialized_cache_hit_count` when Wasmtime deserializes the current `.cwasm`; `unit_module_compile_count` means it fell back to compiling the `.wasm`. `unit_module_lookup_us`, `unit_module_read_us`, `unit_module_parse_us`, `unit_module_compile_us`, and `unit_module_classify_us` divide `unit_module_us` into artifact stat/cache lookup, wasm read, custom-section parse, deserialize-or-compile, and immutable import classification. The phase sum can be below the total because allocation and cache publication overhead remain in the aggregate.
`unit_module_cache_hit_count` and `unit_module_cache_miss_count` divide module loads by the worker's compiled-module cache. A miss is further identified by `unit_module_serialized_cache_hit_count` when Wasmtime deserializes the current `.cwasm`; `unit_module_compile_count` means it fell back to compiling the `.wasm`. `unit_module_lookup_us`, `unit_module_read_us`, `unit_module_read_bytes`, `unit_module_parse_us`, `unit_module_compile_us`, and `unit_module_classify_us` divide `unit_module_us` into artifact stat/cache lookup, wasm metadata/full-artifact read volume, custom-section parse, deserialize-or-compile, and immutable import classification. A current serialized-module hit scans only section headers and the `dylink.0`/`uce.abi` payloads; compilation fallback reads the complete wasm. The phase sum can be below the total because allocation and cache publication overhead remain in the aggregate.
`unit_symbol_resolve_count` and `unit_symbol_resolve_us` isolate function and data symbol lookup within `unit_import_us`. The remainder of import time covers import-vector construction, Wasmtime Globals, GOT function table placement, and related bindings.
site/tests/io.uce
+5 -1
View File
@@ -151,7 +151,11 @@ RENDER(Request& context)
check("to_u64() / to_s64() / to_f64() / to_bool()", to_u64(cfg["u64"], 7) == 42 && to_u64(cfg["bad"], 7) == 7 && to_s64(cfg["s64"], 5) == -12 && to_s64(cfg["bad"], 5) == 5 && to_f64(cfg["f64"], 1.0) > 3.49 && to_f64(cfg["bad"], 1.25) == 1.25 && to_bool(cfg["yes"], false) && !to_bool(cfg["no"], true) && !to_bool("unknown", false), var_dump(cfg));
DValue perf = request_perf();
bool unit_module_profile_valid = perf["unit_module_cache_hit_count"].to_u64() + perf["unit_module_cache_miss_count"].to_u64() == perf["unit_load_count"].to_u64() && perf["unit_module_serialized_cache_hit_count"].to_u64() + perf["unit_module_compile_count"].to_u64() == perf["unit_module_cache_miss_count"].to_u64() && perf["unit_module_lookup_us"].type != 'S' && perf["unit_module_read_us"].type != 'S' && perf["unit_module_parse_us"].type != 'S' && perf["unit_module_compile_us"].type != 'S' && perf["unit_module_classify_us"].type != 'S';
u64 module_misses = perf["unit_module_cache_miss_count"].to_u64();
u64 serialized_hits = perf["unit_module_serialized_cache_hit_count"].to_u64();
u64 module_read_bytes = perf["unit_module_read_bytes"].to_u64();
bool serialized_reads_bounded = module_misses == 0 || serialized_hits != module_misses || (module_read_bytes > 0 && module_read_bytes < module_misses * 1024 * 1024);
bool unit_module_profile_valid = serialized_reads_bounded && perf["unit_module_cache_hit_count"].to_u64() + module_misses == perf["unit_load_count"].to_u64() && serialized_hits + perf["unit_module_compile_count"].to_u64() == module_misses && perf["unit_module_lookup_us"].type != 'S' && perf["unit_module_read_us"].type != 'S' && perf["unit_module_read_bytes"].type != 'S' && perf["unit_module_parse_us"].type != 'S' && perf["unit_module_compile_us"].type != 'S' && perf["unit_module_classify_us"].type != 'S';
bool perf_stable = unit_module_profile_valid && perf["worker_pid"].to_u64() > 0 && perf["running_us"].to_f64() > 0 && perf["dispatch_us"].type != 'S' && perf["workspace_setup_us"].type != 'S' && perf["workspace_birth_us"].type != 'S' && perf["context_apply_us"].type != 'S' && perf["hostcall_count"].to_u64() > 0 && perf["hostcall_us"].type != 'S' && perf["guest_us"].to_f64() > 0 && perf["component_resolve_count"].type != 'S' && perf["component_path_us"].type != 'S' && perf["component_artifact_us"].type != 'S' && perf["component_load_us"].type != 'S' && perf["component_link_us"].type != 'S' && perf["unit_load_count"].to_u64() > 0 && perf["unit_module_us"].type != 'S' && perf["unit_allocate_us"].type != 'S' && perf["unit_import_us"].type != 'S' && perf["unit_symbol_resolve_count"].type != 'S' && perf["unit_symbol_resolve_us"].type != 'S' && perf["unit_instantiate_us"].type != 'S' && perf["unit_initialize_us"].type != 'S';
u64 profiled_hostcalls = perf["hostcall_count"].to_u64();
f64 profiled_hostcall_us = perf["hostcall_us"].to_f64();