This is an experimental work to check deploying the whole SQLFlow service mesh on Windows, Linux, or macOS desktop.

The general architecture of SQLFlow is as the following:

In this deployment, we have Jupyter Notebook server, SQLFlow server, and MySQL running in a container executing the sqlflow/sqlflow:latest image. Argo runs on a minikube cluster running on the VM. The deployment is shown in the folllowing figure:

I chose this deployment plan for reasons:

1. We don't have a well-written local workflow engine, and at the right moment, we need to focus on the Kubernetes-native engine. So, we use minikube and install Argo on minikube.

2. We can install minikube directly on users' desktop computers running Windows, Linux, macOS. However, writing a shell script to do that requires us to consider many edge cases. To have a clear deployment environment, I introduced VM.

3. To make the VM manageable in a programmatic way, I used Vagrant. Please be aware that Vagrant is the only software users need to install to use SQLFlow on their desktop computer. And Vagrant provides official support for Windows, Linux, and macOS.

4. We can run the SQLFlow server container (sqlflow/sqlflow:latest) on minikube as well. But that would add challenge to export ports. Running the container directly in the VM but out of minikube, we

   1. expoe the in-container port by adding EXPOSE statement in the Dockerfile, and

   2. expose the docker port for accessing from outside of the VM by adding the following code snippet to the Vagrantfile.

      config.vm.network "forwarded_port", guest: 3306, host: 3306
      config.vm.network "forwarded_port", guest: 50051, host: 50051
      config.vm.network "forwarded_port", guest: 8888, host: 8888