Continuous Delivery/Orchestration Sytem

Once we have the image built and ready to be run, we need to find a way to get it to the server. There were three great options: Rancher, Nomad/Kubernetes and Ansible.

I will divide all 3 into their own segments and talk about the pro-cons of each option. Spoiler alert, I went with Ansible.

Rancher:

Rancher (v1.6) is a simple orchestration system. You can define services you want to run with a YAML file (similar to Docker-compose) and submit it to your rancher endpoint, rancher will then start said service in one of its registered hosts and automatically forward traffic, load balance, etc for you.

Pro: Automatic load balancing, easy to define, GUI based management

Cons: Hard to run on a 1 host system, We would need to define and setup an ingress, High overhead for a small payoff

 

Nomad/Kubernetes:

Before anyone gets offended, I know the two aren't the same. But they are both "multi-node" setup systems.

Nomad: Closer to Rancher than it is to kubernetes, You can define a single-host and mention what services you want it to run, the host then schedules itself and restarts connections if needed.

Kubernetes: Typically made for multi-node systems, super efficient and highly scalable system. Its totally the bees-knees as the young people say.

Pro: Very easily scalable, high availability is pretty much guaranteed, I'm used to using kubernetes so quick turn-around time

Cons: Suuuuper overkill for what we need, single node clusters tend to be "Dev-mode" and not production ready since the systems are built expecting components to fail

Ansible:

Ansible is an automation tool that can be scripted to manage anything. It can be hence setup to perform Continuous Delivery but cannot perform orchestration itself. Its lightweight. Its run once and forget. We let docker inherently handle orchestration, since all the orchestration we need is a restart when the service dies.

Ansible can be run via a Github Action using the shell-run mode.

Pro: Run and forget, performs CD without too much overhead

Cons: Since its run once and forget, it cannot retry when things don't go as expected. No orchestration possible with it, just delivery

Since its the best fit for what we want, I decided to go ahead with just using Ansible and figuring out the cons through other methods.