Red Hat
May 7, 2015
by Christina Lin
This post is part of short session I spoke in Japan, strictly my opinion.

Automation plays an essential role in DEVOPS, the entire cycle will cripple without automation, the whole idea of DEVOPS is to speed up the process inside enterprise, and relief developer and operation from day to day repetitively meaningless routines. Automation meaning that anything the machine can do, make the machine do it. In the old day, due to technology limitation, politics or if it's just simply not having time to do it, there are many jobs that needs to have human involve. If you seriously want to introduce DEVOPS, You will find that you spend majority of time dealing with automations.

There are few aspect in Automation in the DEVOPS lifecycle, let's take a look at my real life experience, it's one of the most anti-automation case I have ever seen in many years of my consulting career. This happens in one of the SI company, where they were responsible for maintaining a banking solution for a bank, due to many reason, their project has failed to deliver the full function on time, so their project manager decided to move the current version to production first as phase one, then deliver the rest of the services later when it's ready. It was a large monolith application, the developer in the project will push out a version of the application, then a quality assurance personnel will come in and test that version of the application, the scope of the application was huge, plus the new function, it take a day or 2 for the QA to test and return the feedback to the developers. By the time the developer receives the feedback, he/she has already pushed out 2 more version, now he has to go back to the problematic version, and fix it, for me as a developer, I hate being interrupt in the middle of something, and go back to the code 2 week before and try to remember, pick up and fix bug. It's not the problem of fixing the bug, if the feed back could of given to developer right after the code to is out, it'll save a lot of time from all these recaps. This process has taken 2 weeks in the QA cycle before it is pushed to production. The operation then take the application, create an pre-production environment, and then it is tested again, of course by human again. A normal function for this bank will normally take about a month.



This is an extreme case, and I see many improvements can be done, but most importantly, eliminating the human testing routines and should be the top priority. To help speed up the automatic testing, in Red Hat JBoss xPaaS, you can setup an CI server such as Jenkins, that detect from our version control repo, which is built-in in xPaaS using Git. So when everything there is a push to the git, the CI server can detect it, then automatically create an production like container to hold the application, then it can automatically deploy and test the current version. Of course, to make the automation of testing easier, one might also want to consider breaking down applications into smaller pieces, and have these smaller pieces deploy on separate container, this makes testing unit simple and quick. Because of the nature in xPaaS, it is very easy for xPaaS to quickly spin up a container, without heavy involvement of developer and operations. Allowing this will make the whole architecture to become more agile and more flexible.

Normally automating these processes are difficult, because you will need to take in account on different version, software, networks and other environment factors. That makes automation complex, Red Hat JBoss xPaaS has a range of middleware already built-in so there is no need to care about the installation script, that make the entire automation much quicker and easier to maintain.
That's all for automation, next time it's the last part of the series, Continuous Delivery.




Original Post