Architecting High-Density Performance Testing with Container-Based Simulation
summary
Replaces 4,000–5,000 VM-style agents with thousands of lightweight containers orchestrated on a few powerful hosts using Podman on a few powerful hosts and Kubernetes, claiming spin-up times near 1s versus ~32s for VMs. Implements stateful container agents that run real work (Ansible playbooks) and emits timestamps to measure fan‑out intensity metric and completion distribution. Practical limits encountered: file descriptor, PID, sub UID/GID and inotify exhaustion; remedies include increase ulimit, pid and inotify limits, log rotation, central logging and aggressive container lifecycle cleanup.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:03API load testing basics
- 3:09Distributed execution testing
- 5:58Real system metrics
- 7:23Containers vs virtual users
- 8:50Testing architecture overview
- 11:15Fan out intensity measurement
- 15:11Containers vs VMs
- 16:36High density scaling challenges
- 19:25PID exhaustion fix
- 20:53Inotify exhaustion fix
- 21:38Disk exhaustion prevention
- 24:06Disk space recovery
- 25:08Key takeaways and conclusion
- 25:50Q&A
related talks
13mPerformance and Energy Efficiency in Docker
01h 08mPerformance testing in the cloud
16mLiving without pre-production environments
47mBuilding Highly Scalable Appium Infra with Terraform &Kubernetes
23mDistributed performance testing at the push of a button
46mDeploying and maintaining software for today's infrastructure
41mShift Left your Performance Testing
32mP to V to C: The Value of Bringing “Everything” to Containers