fix(souslet): wire a real fetch image - every weight fetch was broken - #4
Merged
Conversation
cmd/souslet/main.go constructed fetch.Manager without ever setting its Image field, unlike cmd/sous-api/main.go's own -fetch-image flag (defaulted and wired). Every fetch job souslet ever started ran against Docker's zero-value image reference - the container was created with no real rootfs, and its "python3" entrypoint failed with "executable file not found in $PATH" (surfaced as ExitCode 127, docker ps showing "<no image>"), not a python-level error. Found live: the first-ever deploy of a model needing a fetch through the new souslet path (unsloth/gemma-4-12b-it-NVFP4 on aorus-ubuntu) failed with exactly this signature. This affected every souslet-managed node, not just this one - asus-gx10's souslet has simply never yet needed to fetch anything itself (its one deployed model's weights already exist on disk from the single-node Sous era). New -fetch-image flag, matching cmd/sous-api's own flag name/purpose, but defaulting to the MANIFEST-LIST digest (not a single platform's own digest, which is what sous-api's default is) - this fleet is multi-arch (asus-gx10 arm64, aorus-ubuntu amd64) and one souslet binary/flag value has to resolve correctly on either. Verified via `docker buildx imagetools inspect vllm/vllm-openai:v0.28.0` that the chosen digest is genuinely the list covering both platforms, not one arch's sub-manifest. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
`cmd/souslet/main.go` constructed `fetch.Manager` without ever setting its `Image` field. Every fetch job souslet started ran against Docker's zero-value image reference - created successfully, then failed at start with `exec: "python3": executable file not found` (ExitCode 127, `docker ps` showing ``).
Found live tonight: the first-ever deploy needing a fetch through souslet (gemma-4-12b on aorus-ubuntu) failed with exactly this signature. Affected every souslet-managed node.
New `-fetch-image` flag, defaulting to the multi-arch MANIFEST-LIST digest for vllm/vllm-openai:v0.28.0 (covers both linux/arm64 and linux/amd64) rather than a single platform's own digest, since this fleet is multi-arch and one flag value has to work on either node.
Reviewed (single fast pass): safe to merge, no Critical/Important findings. Reviewer independently reproduced the exact failure signature against a real Docker Engine API call with Image:"" and independently verified the digest is genuinely a manifest list covering both platforms (not a single-arch sub-digest). go build/vet/test clean, 539/28 passing.
🤖 Generated with Claude Code