Reusing bazel’s analysis cache by cloning micro-VMs

on YouTube ↗28m

" summary

Build system warm state is preserved by serializing Firecracker VMs so subsequent CI runs reuse Bazel’s in-memory analysis cache, cutting median pipeline time ~8x (multi‑minute runs drop to seconds). Key tools are Firecracker snapshots, plus splitting huge snapshot files (20+ GB) into cacheable chunks and serving memory/disk on demand with userfaultfd and FUSE. Implementation adds copy‑on‑write to avoid clobbering, compresses chunks, and supports remote sharing for scale; remaining costs are snapshot size, initial snapshot creation, and engineering complexity to safely update master snapshots.

AI-written summary · about

" chapters (11)

1. 0:00MicroVMs and CI bottlenecks
2. 1:34Slow CI workflows with Bazel
3. 3:40Cloning CI runners for Bazel caches
4. 5:26Firecracker microVMs for CI cloning
5. 8:55Firecracker snapshot challenges
6. 11:22Snapshots for remote caching
7. 13:51Userfaultfd for efficient memory access
8. 15:38FUSE for disk access and caching
9. 17:45CI cloning live demo
10. 19:59Production status and integration
11. 21:08Q&A