Auto prod accept - #16
Merged
Merged
Conversation
Remove gpus: all and NVIDIA_VISIBLE_DEVICES from the portable base stack so ProSeqGo boots on CPU-only hosts (GitHub Actions, CPU cloud VMs). Add docker-compose.gpu.yml to opt into NVIDIA for embedding-worker, go-prediction-api, and trainer-worker. Add docker-compose.ci.yml to force CAFA_DEVICE=cpu, use CPU PyTorch wheels, and skip postgres-backup/ backup-offload during serving smoke via the ci-skip profile.
Introduce COMPOSE_DEV_FILES (base + gpu when nvidia-smi works) and COMPOSE_CI_FILES (base + ci overlay) so local dev and CI smoke share one command surface. Add ci-up (ci-env + gateway-auth + compose up) and ci-down (down -v). Route training/monitoring/all-up targets through the same dev file set. Rename up-all/down-all to all-up/all-down for consistency.
…es in README and CI.md. Document make ci-up / make smoke / make ci-down in smoke README. Move gateway vars to the top of .env.example and note CAFA_DEVICE behavior across base, gpu, and ci overlays.
Replace heredoc stdin parsing with python -c + PRED_RESP env var so large JSON responses are validated correctly. Default BASE_URL to http://localhost for consistency with gateway docs.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Introduces new Design for ProSeqGo Compose: a CPU-portable base stack with optional GPU and CI overlays. The serving stack can start on CPU-only hosts (e.g. GitHub Actions) while local dev still uses GPU automatically when NVIDIA is available.
What changed
Compose architecture
docker-compose.yml— Refactored to a portable base: removedgpus: allandNVIDIA_VISIBLE_DEVICESfromembedding-worker,go-prediction-api, andtrainer-worker.docker-compose.gpu.yml(new) — GPU overlay addinggpus: allfor inference/training workers on NVIDIA hosts.docker-compose.ci.yml(new) — CI overlay forcingCAFA_DEVICE=cpu, CPU PyTorch wheels (TORCH_INDEX_URL=.../cpu), and skipping backup sidecars via the ci-skip profile.Makefile
make up/make down/training/monitoringtargets.make ci-up(runsci-env+gateway-auth+ CPU compose stack) andmake ci-down(tear down with volumes).Renames
up-all/down-all→all-up/all-down.Documentation
README.md,.github/CI.md,tests/smoke/README.md, and.env.exampleto document the overlay pattern and CI smoke workflow.Smoke test fix
tests/smoke/smoke_embedding_api.sh— Fixespredict-go-from-fastaJSON validation (env-based parsing instead of heredoc stdin) and defaultsBASE_URLtohttp://localhost.Why
gpus: allin the base compose file blocked the stack from starting on CPU-only machines, which blocked Phase 3 CI smoke on GitHub-hosted runners. Application code already supported CPU viaCAFA_DEVICE=auto, but Compose enforced GPU at container startup.