Change deployment method from Docker to Kubernetes

[ema-pe]

With the evolution of the prototype, the next step is to update the deployment method from a Docker-based solution to Kubernetes. This step is necessary not only because it is necessary to integrate DFaaS into the WatchEDGE platform (i.e., Kubernetes-based), but also to enable DFaaS to be deployed with Kubernetes and take advantage of its capabilities.

The migration involves some decisions to be made, in particular:

• faasd is the official OpenFaaS provider for single-host deployments using only Docker. For Kubernetes, there is the faas-netes provider.

• With the migration from faasd to faas-nets, the OpenFaaS architecture changes and there is the ability to run asynchronous functions using NATS:

  

• There should be no problems using OpenFaaS on Kubernetes, there are a lot of resources to deploy it on Kubernetes. The same goes for HAProxy since there is an ingress controller for it.

• The latter is not true for the DFaaS agent. This is perhaps the most important step in moving to Kubernetes, as it requires writing operators and custom resources.

• Since OpenFaaS CE (especially faasd, with the limit of 15 functions) has some limitations for the non-commercial version, we should evaluate some other alternatives that use Kubernetes as a platform, such as Apache OpenWisk. If we can extract some important metrics (like queue status), this can be a valid reason to change the FaaS platform.

• We should also evaluate an additional tool to help with migration and deployment, such as Helm or KubeVela.

This issue covers general information about migration; each topic should be covered in more detail in a separate issue.