webkernel/config

Benchmark & Performance Profiling

Sur cette page

Measure cold CLI access, a new key in a loaded section, a new section, PHP-FPM APCu access, full boot and repeated lookups separately. The target is 90 µs or less for the first keyed config() call; current cold CLI measurements are higher.

Measurement boundaries

Measurement	Included work
Missing or stale cache	Discovery, source execution, compilation, cache write and lookup
Cold CLI, first keyed access	Read and decode one section file, then resolve the requested key; freshness validation is deferred until shutdown
New key, loaded section	Traverse that section's in-memory tree once and memoize the key
PHP-FPM, warm section	Stat one section file and fetch its decoded tree from APCu, then resolve the requested key
PHP-FPM, cold section	Read and decode the section file, then place its decoded tree in APCu for later requests
Full `Config::boot()`	Read compiled PHP cache, compare inputs, check filesystem metadata and load all values
Subsequent access	In-memory lookup
`config_path()`	Configured path access and optional concatenation

Do not call Config::boot() or read configuration before timing a cold keyed access. Doing so excludes section loading. Measure a new key in an already loaded section separately from a key in a new section. In PHP-FPM, measure cold and APCu-warm section access in separate requests. Time Config::boot() and shutdown freshness work separately.

Example

Run this after the application's Composer autoloader has been required, before any configuration access:

hrtime(true);

$start = hrtime(true);
$value = config('paths.app');
$first_end = hrtime(true);

config('paths.app');
$second_end = hrtime(true);

$iterations = 1_000_000;
$loop_start = hrtime(true);
for ($index = 0; $index < $iterations; $index++) {
    config('paths.app');
}
$loop_end = hrtime(true);

printf("First access: %.3f µs\n", ($first_end - $start) / 1_000);
printf("Second access: %.3f µs\n", ($second_end - $first_end) / 1_000);
printf("Repeated mean: %.3f µs\n", ($loop_end - $loop_start) / $iterations / 1_000);

The repeated mean includes loop overhead. Print values after timing, and confirm the requested key exists. A missing key exercises the fallback path. A fresh CLI process cannot reuse its process memory; its disk-cache timing differs from later PHP-FPM requests that share APCu. Include process startup and shutdown separately when measuring command wall time.

OPcache and filesystem conditions

Record PHP version, SAPI, OPcache settings, whether bytecode is warm, source count, filesystem type, and whether the compiled cache was current. Performance measurements should enable OPcache. A fresh CLI process with opcache.enable_cli=1 still pays initial bytecode compilation; it does not represent a warmed PHP worker.

Use multiple runs and report their spread. An isolated first-call sample is sensitive to scheduling and filesystem state. A fast repeated mean does not establish the first-call budget.