Graham King

Solvitas perambulum

A developer goes to a DevOps conference

As a professional software engineer and enthusiastic pro-am systems administrator, I have long been curious about DevOps.

I spent the last two days at PDX DevOpsDays. Across organised talks, Ignite’s, Open Spaces, and as many hallway conversations as I could fit in, here is what I learned about what “DevOps” is and does.

DevOps is Ops with new tools. I went to DevOpsDays thinking DevOps means developers and operations merged into one team. I was wrong.

I did not meet or hear from any developers. It was a gathering of system administrators who use version control, and write a lot of YAML. Programming languages were not a topic; I heard Python mentioned once in passing, that’s it.

DevOps means the veteran admins had to check in their personal scripts, and everyone is expected to automate more things. They relate to the software their business runs (“app”, “asset”) the way a merchant-navy captain relates to what’s in the containers on his ship.

AWS is king. Whenever a hosting provider was mentioned, it was AWS. It was usually mentioned as if it was a basic understanding that we’re both on AWS. Google Cloud was a sponsor but didn’t otherwise get mentioned, Digital Ocean was mentioned once, and one of my lunch tables had someone who uses Azure. Otherwise it was all AWS.

Slack is king. “Chatops” is big. Being a gathering of system administrators I’d expected more IRC, there was no mention of it.

Everyone hates YAML. Everyone writes a lot of YAML.

On-call / incident-response is a big part of the DevOps job, and several of the sessions focused solely on that. As a result they also talk about mental health more than programmers do.

Cloud monitoring is a saturated market. Maybe half the vendors/sponsors were cloud monitoring (i.e. Prometheus competitors), although most of them called it something else.

Kubernetes, Terraform, Ansible and Packer were the tools I heard mentioned the most, in that order. Prometheus was popular for monitoring, but many (most?) used a vendor, such as Datadog.

The most common job title seemed to be SRE (Site Reliability Engineer), although there was a long tail, and they don’t care much about job titles.

Many were curious about service-meshes (basically a smart network proxy?) such as Istio or Linkerd, but almost none were using one.

The consensus on how software releases happen is that your CI/CD pipeline builds and tests the “asset” resulting in a Docker container, which is they deployed in a blue-green fashion using either Kubernetes or Terraform. The existence of a CI/CD pipeline that produces a Docker container was assumed in most conversations I had. Everyone loves containers.

PDX DevOpsDays was one of the best run conferences I’ve been to (MC Alice Goldfuss was phenomenal), and probably the one I learnt the most at. A++++ would conference again.