I am deploying an ML model using Docker in Beanstalk. Firstly, I uploaded my Docker image (which contains my ML model) to Docker Hub. Then, I deployed it into Beanstalk using docker-compose.yml. In Beanstalk, I am using Docker as a platform, and my model requires GPU support. For that, I have used the Deep Learning AMI GPU CUDA 11.5.2 (Amazon Linux 2) 20230104, which is built with NVIDIA CUDA, cuDNN, NCCL, GPU Driver, Docker, NVIDIA-Docker, and EFA support. However, when I am building the environment using this configuration, I encounter the following error:
**[ERROR]** An error occurred during execution of command [app-deploy]
- [Track pids in healthd]. Stop running the command. Error: update processes
[docker eb-docker-compose-events eb-docker-compose-log eb-docker-events cfn-hup healthd]
pid symlinks failed with error read pid source file /var/pids/docker.pid failed with
error:open /var/pids/docker.pid: no such file or directory.
Means my environment is building but it is giving the error message like:
Env build successfully but with some errors.
I found this error message in the eb.engine.log.
Additionally, I checked the EC2 instance through SSH, it showed that NVIDIA drivers, NVIDIA CUDA, and Docker are already installed (verified using these commands: nvidia-smi, docker -v). I have tried multiple different deep learning AMIs, but I encounter the same issue with all of them. Also, I noticed a strange thing while trying out different AMIs: if I build the environment with default settings like Docker platform with the default Docker AMI, it builds successfully without any error. However, when I pass a different AMI ID in the configuration, it is not able to build the environment properly.
How to solve this error?