chore: Migrate from Poetry to UV, added health checkpoint, and improved dockerimage #15
Conversation
spa5k
commented
Mar 4, 2025
•
spa5k
commented
- Add .dockerignore to optimize Docker build context
- Refactor docker-compose files for CPU and GPU modes to add health check.
- Update Dockerfile with multi-stage build and improved caching
- Add Makefile for simplified project management
- Implement health check endpoint in document converter route
- Optimize container runtime and model download process
- Improve documentation and setup instructions
- Add .dockerignore to optimize Docker build context - Refactor docker-compose files for CPU and GPU modes - Update Dockerfile with multi-stage build and improved caching - Add Makefile for simplified project management - Implement health check endpoint in document converter route - Optimize container runtime and model download process - Improve documentation and setup instructions
|
@drmingler do you mind taking a look? |
spa5k
changed the title
@spa5k Awesome work! Thank you. I will have a look and get back. Meanwhile can we have a unified lint rule so that we can avoid modifying the code formatting? |
|
Sure, I tried not to change much formatting, will reset wherever it happened. UV also provides a formatter, maybe it can be used |
|
What about this? Let's merge this, then I will create a PR to implement Ruff, as a GitHub action, and a pre commit in next PR, otherwise, there will be too many changes in this PR. |
Okay that makes sense. I will merge this over the weekend. Thank you |
…type safety and API documentation - Add comprehensive response examples and status codes - Introduce JobStatus and ImageType enums for better type safety - Enhance route error handling and status reporting - Add detailed field descriptions and validation - Improve health check endpoint with more granular service status reporting
|
Thanks, the PR is now in quite good stage, I tested the caching, and schema changes, including generating SDK using it for Go, and everything seems to be working well. |
| ```bash | ||
| curl -sSL https://install.python-poetry.org | python3 - | ||
| curl -LsSf https://astral.sh/uv/install.sh | sh |
There was a problem hiding this comment.
Can we also add instructions on how to install uv for those using windows or add a link to the uv installation doc? @spa5k
README.md
Outdated
| make docker-run-gpu | ||
|
|
||
| # Or build and run with multiple workers | ||
| make docker-run-gpu WORKER_COUNT=3 | ||
| ``` |
There was a problem hiding this comment.
Should be "docker-start-cpu" not "docker-run-gpu"
…build and runtime performance - Simplify Dockerfile with single-stage build - Improve PyTorch and EasyOCR model installation based on system architecture - Update docker-compose files to remove runtime target - Add auto-detection for CPU/GPU docker start in Makefile - Enhance .dockerignore with additional Python-related exclusions - Add entrypoint script for consistent container startup
- Introduce detect_gpu.sh script for dynamic GPU and Docker configuration detection - Update docker-compose.cpu.yml and docker-compose.gpu.yml to use uv for running commands - Enhance Dockerfile with multi-stage build, improved GPU/CPU detection, and model downloading - Add NVIDIA GPU capabilities and platform configuration in Docker Compose files - Improve runtime and build performance with better caching and environment setup
|
Can you check the GPU part now? If it is not working, please move the pytorch and model downloads part to the runtime stage |
| try: | ||
| result = document_converter_service.get_single_document_task_result(job_id) | ||
|
|
||
| # Return 202 Accepted if job is still in progress | ||
| if result.status in ["IN_PROGRESS"]: | ||
| return JSONResponse( | ||
| status_code=status.HTTP_202_ACCEPTED, | ||
| content=result.dict(exclude_none=True) | ||
| ) | ||
|
|
||
| # Return 422 for failed jobs | ||
| if result.status == "FAILURE": | ||
| return JSONResponse( | ||
| status_code=status.HTTP_422_UNPROCESSABLE_ENTITY, | ||
| content=result.dict(exclude_none=True) | ||
| ) | ||
|
|
||
| # Return 200 OK for successful jobs | ||
| return JSONResponse( | ||
| status_code=status.HTTP_200_OK, | ||
| content=result.dict(exclude_none=True) | ||
| ) | ||
| except KeyError: | ||
| raise HTTPException( | ||
| status_code=status.HTTP_404_NOT_FOUND, | ||
| detail=f"Job not found: {job_id}" | ||
| ) |
There was a problem hiding this comment.
Please take this off, during job processing we do not raise errors, we return a status of FAILURE if the job failed and the reason for the failure in the error field. It was intentionally done that way.
| result = document_converter_service.get_batch_conversion_task_result(job_id) | ||
|
|
||
| # Return 202 Accepted if the batch job or any sub-job is still in progress | ||
| if result.status in ["IN_PROGRESS"] or any( | ||
| job.status in ["IN_PROGRESS"] | ||
| for job in result.conversion_results | ||
| ): | ||
| return JSONResponse( | ||
| status_code=status.HTTP_202_ACCEPTED, | ||
| content=result.dict(exclude_none=True) | ||
| ) | ||
|
|
||
| # Return 422 for failed batch jobs | ||
| if result.status == "FAILURE" or any( | ||
| job.status == "FAILURE" | ||
| for job in result.conversion_results | ||
| ): | ||
| return JSONResponse( | ||
| status_code=status.HTTP_422_UNPROCESSABLE_ENTITY, | ||
| content=result.dict(exclude_none=True) | ||
| ) | ||
|
|
||
| # Return 200 OK for successful batch jobs (all success) | ||
| return JSONResponse( | ||
| status_code=status.HTTP_200_OK, | ||
| content=result.dict(exclude_none=True) | ||
| ) | ||
| except KeyError: | ||
| raise HTTPException( | ||
| status_code=status.HTTP_404_NOT_FOUND, | ||
| detail=f"Batch job not found: {job_id}" | ||
| ) |
I will test on gpu and get back to you. @spa5k |
