Sizing the headless Adobe Commerce Cloud for a large number of API requests
summary
Adobe Commerce Cloud handles a headless, non-cacheable personalized-offer flow for 7‑Eleven using Adobe Campaign and experiences PHP‑FPM contention when API spikes hit pm.max_children limits. Monitoring via New Relic and Fastly logs plus nginx timeouts surfaces queued requests (listen_q) and max_children_reached events. Load tests model 5,000 concurrent sessions (180s) and calculate a theoretical minimum of 514 PHP workers and a practical target of ~1,028 workers (×2 safety); use static PM, distributed load generators, and queue metrics to size vCPU instances.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:58API calls
- 3:55Agenda
- 5:16Maximum concurrency
- 9:18Queueing
- 15:15Status Page
- 16:59PHP Configuration
- 18:40Dynamic vs Static Configuration
- 20:10Qn Theory
- 21:32Load Test Scenario
- 25:04Load Generators
- 26:48Summary
related talks
50mBuilding and Operating Online Shops under Extreme Load
34mHow We Handle Over 1 Trillion HTTP Requests ...
58mMassively Scaled High Performance Web Services with PHP
35mMaximize Performance: Asynchronous PHP Explained!
45mSurviving a Prime Time TV Commercial
44mIntroduction to application scaling
27mScaling for the Expected and Unex
56mLarge scale websites performance optimisation tricks