Richard Bucker

IT infrastructure best practices

Posted at — Oct 1, 2018

Yikes! Do the masses get lost and confused in the buzz word and buzz tech of the day. Let’s consider:configuration as codeidempotentmicro servicesCI/CDfull stack automated deployfail early fail fastagile - scrumsix sigmaAnd then there’s what Kubernetes got right but that we do not really understand and how it might be wrong after all.Taken separately each item in the list seems like a good idea and together they feel like powerful tools. In practice, however, there is no magic brush that allows you to paint the Mona Lisa in a single stroke. Even in the most ideal HA cluster scenario there are so many dependencies that have to be addressed and at some point you have to make decisions on the many single points of failure and risk.For example; building a docker swarm from scratch from a batch file is pretty simple. Just configure and license one or more VMware or bare metal docker servers and deploy to your heart’s content. Make sure that the containers you install are trusted then you are still OK. Somewhere you need a git server and some shared volumes and some sort of CI/CD to deploy your apps and services on the swarm/cluster. That’s gong to work for a while until you add a second customer or a second cluster or when a package needs to be upgraded or report deployed.Configuration in code and full stack deployment means that if one service changes then the entire stack needs to be relaunched. This way DEVOPS knows, verifies and trusts that they can recover from a failure. But when your services are a combination of OLTP and OLAP services then a full redeploy may have other types of unwanted side effect. Also depending on the size of the system it could take hours to redeploy a system.But if you have to deploy manually it’s also a challenge to keep the docs and scripts in order.Six sigma is a waste of a good tree because you can still be in compliance if you schedule the downtime. So what’s the point of that?The other problem with an all in one approach is that we NEVER seem to execute the promise of move everything at once. Management always seems to change their mind and move things one piece at a time which actually creates it’s own set of problems.