Let’s Farm Out Our Image Builds!
summary
Podman farm uses the Podman unix socket over SSH to send a single build to remote nodes of different CPU architectures, run native builds in parallel on those nodes, and pushes images to the registry then create a manifest list locally. A recorded demo builds a llama cpp python model server image in about 3m08s vs 11m35s when using QEMU-user-static emulation locally, supports a podman farm build flag to clean up node images, and uses registry creds forwarded from the local machine.
AI-written summary · about
chapters (19)
- 0:00Multi-arch image builds intro
- 1:02What is multi-arch and why
- 3:51Manual builds and emulation limits
- 4:54Podman Farm for seamless builds
- 7:01Demo setup and Podman sockets
- 9:06First Podman Farm build
- 11:56Farm vs emulation comparison
- 13:21Podman Farm commands
- 15:10Local caching behavior
- 16:32Cleanup and resource management
- 17:36Farm simplifies multi-arch workflows
- 18:21Multi-arch in AI and edge computing
- 19:27Resources and community
- 19:47Q&A on socket auth
- 20:31Q&A on caching and optimization
- 21:55Q&A on cost optimization
- 23:01Q&A on config sharing
- 23:43Q&A on Kubernetes integration
- 24:26Q&A on arch-specific images
related talks
36m10 Cool features in Podman and Podman Desktop
15mBuilding multi-arch container images with buildah
31mPodman Health Check: A Data-Driven Look at Community, Performance, and Trends
15mFive podman features you didn’t know about or that you should consider start using
37mPodman 5 & Long-Term Software Maintenance
34mWhat's new in Image Builder?
34mPodman 101 and Beyond!
41mThe Android Device Farm That Fits in a (Cloudy) Pocket