Mullion 2.5.0 — Performance

Mullion 2.5.0

A Performance tab on every project: per-route timings with budgets and baselines, a PHP profiler with flame graphs (macOS), production-readiness checks, a built-in load test, production-like mode and worker health.

Upgrade: Settings → About & updates → Upgrade · brew upgrade mullion · scoop update mullion

Added

• PHP profiler: flame graphs for a site's requests (macOS). The
  project page's Performance → Profiler tab (and mullion profile <site> on|off|status|list|show|clear) records where PHP requests spend their
  time with SPX 0.4.22
  (GPLv3 — built on your Mac from source, never shipped by Mullion; about
  a minute the first time per PHP version, with the Xcode Command Line
  Tools). spx.so is loaded disabled, which costs nothing; turning a site
  on passes its SPX settings through Caddy for that site only (other
  sites on the same php-fpm are explicitly kept off). Profile only when
  asked (SPX's control panel for your browser session, or
  ?SPX_ENABLED=1&SPX_KEY=… on any URL), the next N requests, requests
  matching a path (/api, /orders/*), or every request. Each profile
  shows its request, wall time, memory and call count, a zoomable,
  searchable flame graph and the top functions by self and inclusive
  time, with a link to SPX's own viewer; the newest 100 profiles (at most
  256 MB) per site are kept. mullion php update rebuilds SPX for the new
  patch when the old one had it. Not available on Windows yet: SPX has no
  Windows build.

• Performance tab on the project page (and mullion perf <site> routes|check|load|prod|workers). Routes: per-route count,
  average, p50/p95/p99, queries, query time, peak memory, error rate and
  N+1 from the Watch data (Laravel's route names with Deep on), with
  per-site budgets, saved baselines and a regressions-first comparison.
  Readiness: a production-readiness audit — debug/env/log settings,
  config and route cache readiness, dev packages in require, Composer
  autoloader, strict models, OPcache, PHP end of life, database indexes
  and primary keys (read-only, through the .env connection), the N+1
  and slow queries Watch saw, and the frontend build — each finding with
  the why, the exact fix, Run fix/Simulate and Ignore. Load test: a
  built-in load tester against the site's own hosts with live RPS,
  percentiles, status codes and a latency histogram, results kept and
  compared with the previous run. Production-like mode: APP_ENV and
  APP_DEBUG passed by Caddy (the .env stays as it is), OPcache
  timestamps frozen for the site, optional artisan optimize, long
  cache headers for fingerprinted assets; frontend sites serve their
  build. Workers: memory/CPU of each worker sampled every 10 s, leak
  detection, restarts, and per-job timings and repeated failures from
  queue workers' logs.

• Watch: the request event now carries its query count and total query
  time (kept even past the 3,000-events-per-request cap).

Fixed

• Explain: SQL Server on Windows failed with "An unsupported attribute
  was designated on the PDO object" (pdo_sqlsrv rejects
  PDO::ATTR_TIMEOUT).