Testing Ceph: Pains & Pleasures
summary
Ceph proves resilient because a homegrown test harness runs combinatorial suites and thousands of machine‑hours daily using teuthology automated testing system to drive RADOS, OSDs and monitors over SSH, inject failures with Thrashers, archive logs and cores, and queue jobs via Beanstalk in a shared sepia lab. Results are public, unit and API fuzzing exist, but performance and deployer integration are weak; a practical fix is subset functionality to cover 124k combos incrementally.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:31CEPH COMPONENTS
- 1:18RADOS COMPONENTS
- 2:49Eight Years Ago
- 12:43Ceph Branch Testing
- 13:28Ceph Pull Request Testing
- 16:10Test Coverage Gaps
- 17:45Test Coverage Fixes
- 34:32Comprehensive Changes Coming
related talks
36mTesting Ceph: Status, Development, & Opportunities
34mTesting Ceph for the Cloud, in the Cloud
02h 59mCeph Storage: a Caffeinated Primer
05mCeph BenchOps: Following DevOps Practices for Benchmarking Ceph
27mFrom Chaos To Clarity: Systematic Ceph Cluster Analysis for Rapid Troublesho...
29mCeph Operations at Scale
34mModernizing Ceph Deployments at CERN: CephFS and Object Storage Across Data Centres
35mMaking Ceph Fast in the Face of Failure