Deep Dive Into Cilium Resilient Architecture
summary
Cilium documents 680 contributors, ~470k lines of Go and ~35k lines of eBPF and introduces an in‑memory StateDB that stores immutable revisioned snapshots with lockless readers and watch channels, a reusable reconciler that retries failing BPF map writes (ENOSPC), updates metrics and writes failing status back into StateDB, and a new agent Health export. Recovery demo shows service BPF maps filling, drops rising, deleting services clears maps and restores traffic; StateDB is not durable and is rebuilt from BPF maps and the API server.
AI-written summary · about
chapters (7)
- 0:00Cilium resilient architecture intro
- 4:23Historical context and complexity
- 7:13BPF maps reconciliation
- 9:04StateDB for state management
- 12:34Reconciler pattern and failures
- 14:45Resilience improvements and future
- 16:12Q&A on StateDB
related talks
36mCilium: Welcome, Vision and Updates
38mUnderstanding and Troubleshooting the eBPF Datapath in Cilium
28mA Decade of Cilium Around the World
22mOne Project, Different Angles: How to Secure and Observe with Cilium
20mTackling Container Networking at Scale
31m10 Years of Cilium: Connecting, Securing, and Simplifying the Cloud Nat...
04mClosing Remarks
24mHigh-Scale Networking for ML Workloads With Cilium