Lightning Talk: How to Get Slim with Soci and Stars: Minifying Containers with Lazy Image Loading in ContainerD
summary
Explores using ContainerD’s lazy image loading via SOCI index and stargz per-file tars to derive exact files loaded at runtime and feed that telemetry into SlimToolkit to produce tiny, runnable images. Shows that snapshotters use HTTP range requests and FUSE callbacks, requires disabling background fetching and lowering default chunk sizes (SOCI defaults to 4MB), and needs an instrumented registry (one-line log) to map ranges back to file offsets and generate minified images.
AI-written summary · about
chapters (12)
- 0:00Minimal container images
- 1:11Lazy loading with snapshots
- 2:20Container image layers
- 3:23On-demand loading with Starz
- 4:50SOI index structure
- 6:20StarGZ index rebuilding
- 7:24Disabling background fetching
- 8:12Minified images with Darker Slim
- 8:55Live demo setup
- 10:14Range requests demo
- 12:50File resolution requirements
- 13:54Conclusion
related talks
05mLightning Talk: What Made Your Container Fat? Visualizing the Size of Container Layers
20m⚡️ FEDC 2018 - Eric Portis
30mFaster Container Image Distribution on a Variety of Tools with Lazy Pulling
06mLightning talk from Andrey Sitnik
45mLightning talks
41mHow we make Web Apps lightning fast at Microsoft!
26mFast and Beautiful Modern Image Delivery Techniques
09mLightning talk from Kamlesh Chandnani