Conversation
Adds a GPU variant of the demo-bundled image so users with NVIDIA GPUs can run `flyte start demo --image ghcr.io/flyteorg/flyte-demo:gpu-latest` and submit tasks with `Resources(gpu=1)`. - Dockerfile.gpu stages NVIDIA Container Toolkit v1.19.x binaries and their shared libs into the rancher/k3s final image. Libs are copied into /usr/lib/<triple>/ because the nvidia-ctk OCI hook runs without inheriting LD_LIBRARY_PATH. A statically-linked /sbin/ldconfig is also staged (rancher/k3s ships none) because the toolkit's update-ldcache hook bind-mounts it into workload pods. - containerd-config.toml.tmpl sets nvidia as the default containerd runtime. Pods requesting nvidia.com/gpu get GPUs without needing runtimeClassName in their spec; non-GPU pods are unaffected (nvidia-container-runtime is a passthrough when no GPU is requested). - nvidia-device-plugin.yaml installs a RuntimeClass and the NVIDIA k8s-device-plugin DaemonSet so nvidia.com/gpu is advertised on the node. Auto-applied by k3s at startup. - Makefile gains a build-gpu target producing flyte-demo:gpu-latest. - CI gains a build-and-push step publishing gpu-latest, gpu-nightly, and gpu-<sha> tags to both flyte-demo and flyte-sandbox-v2. The GPU plumbing was verified end-to-end with a layered test image on an A10G (torch 2.11.0+cu130 reported cuda_available=True). The full multi-stage Dockerfile.gpu has not been built locally; the CI run here is the first end-to-end test of the production Dockerfile and may need fixup iterations. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Removes the duplicated builder/bootstrap/pg-cache stages and final-stage setup by making Dockerfile.gpu a thin layer on top of flyte-demo:latest (parameterized via ARG BASE_IMAGE). CI now builds the CPU image first and passes its sha-tag in as BASE_IMAGE to the GPU build. - Dockerfile.gpu shrinks from ~165 to ~75 lines; inherits flyte-binary, embedded postgres, staging manifests, and k3d entrypoint from the base image unchanged. - Makefile build-gpu target now depends on build (not the full prereq chain) and passes BASE_IMAGE=flyte-demo:latest. - CI gates the GPU build on push/workflow_dispatch since PR builds don't push the CPU image to ghcr.io (nothing to pull for BASE_IMAGE). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Drops the `if:` gate and conditions `push:` on the same expression the CPU build uses, so both steps always build and only push on v2-branch pushes or workflow_dispatch. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
On pull_request events the CPU build step runs with push=false, so the GPU build's FROM ghcr.io/.../flyte-demo:sha-<sha> fails to resolve (image not found in the registry). Fix by producing an OCI archive of the CPU image locally and passing it to the GPU build as a named build context (build-contexts: base=oci-layout://...) with BASE_IMAGE=base. Registry push happens in a separate step that only runs on push / workflow_dispatch, so PR builds no longer need ghcr credentials for the GPU step.
Add GHA cache (type=gha) to the three docker/build-push-action steps in build-and-push-demo-bundled-image. CPU archive and CPU push share the demo-cpu scope so the push reuses layers from the archive build; GPU gets its own demo-gpu scope.
The oci-layout:// build-context source requires Dockerfile frontend 1.5+. CI was failing with 'unsupported context source oci-layout for base'. Signed-off-by: Kevin Su <pingsutw@apache.org>
Rename all references of flyte-demo to flyte-devbox and demo-bundled to devbox-bundled across Helm charts, Docker files, Makefiles, CI workflows, Go source, and Kubernetes manifests. Signed-off-by: Kevin Su <pingsutw@apache.org>
This was referenced Apr 21, 2026
Draft
Resolve merge conflicts from GPU build steps added in v2, applying the flyte-demo → flyte-devbox rename to the incoming changes. Signed-off-by: Kevin Su <pingsutw@apache.org>
Signed-off-by: Kevin Su <pingsutw@apache.org>
- Tag nightly on push to v2, latest only on manual workflow_dispatch - Fix local build-gpu by using --build-context to pass CPU image into the docker-container BuildKit driver Signed-off-by: Kevin Su <pingsutw@apache.org>
Signed-off-by: Kevin Su <pingsutw@apache.org>
cosmicBboy
approved these changes
Apr 21, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
TL;DR
Rename all references of
flyte-demotoflyte-devboxanddemo-bundledtodevbox-bundled.Type
Are all requirements met?
Complete description
Renames across the entire codebase:
flyte-demo→flyte-devbox(Helm chart name, Docker image tags, container/volume names, labels, service names, config values)FLYTE_DEMO→FLYTE_DEVBOX(env vars)flytedemo→flytedevbox(Helm template release name)demo-bundled→devbox-bundled(directory names, Go module path, CI workflow references)demo-build/run/stop→devbox-build/run/stopcharts/flyte-demo→charts/flyte-devbox,docker/demo-bundled→docker/devbox-bundledTracking Issue
NA
Follow-up issue
NA
main