Would ChatGPT be the future Kubernetes SRE?

I tested this AI to find out how much it knows about Kubernetes!

The first impression is that it is still not yet time for Skynet to catch us! Also, the fear of losing our jobs to an artificial intelligence is not yet a reality. On the other hand, I was enthusiastic about the potential that OpenAI has to revolutionize our daily lives.  The chat brings much more direct and simpler-to-understand results than a bunch of open tabs in Google.

Recently launched by OpenAI, ChatGPT is an artificial intelligence that can help obtain information about specific subjects, solve practical problems, or learn something new. In its own words, ChatGPT is:

”I am an artificial language model created by OpenAI. My function is to help people find answers and perform tasks using natural language. I do not have a personality or desires of my own, and my objective is to provide accurate and useful information to help people solve their problems or better understand a subject. I cannot see, hear, or feel like a person, and my answers are based only on the information provided to me.”

Testing ChatGPT 

As promised, let's head to our experiment to find out how much ChatGPT knows about Kubernetes cluster administration and if it can help us with daily tasks. 

To begin with, I tested a bit of its ability to “understand problems and guide in natural language”:

Cool, the tool gave some simple guidance that we would also find in a quick search of the documentation. As for the way it communicates, the robot really has a natural language, similar to human!

Now, let's see how it handles a real error:

Up to this point, ChatGPT gave good guidance, but nothing very practical, because I wanted something more than just a direction. With this log, I expect and need something more practical. Therefore, let's be more direct based on this guidance and the meaning of the error:

Now we're talking, a very direct tip on how to find the offending “pods"! But let's not stop at problem remediation. Let's go further: how can we prevent this from happening?

Wonderful!

This is a very important recommendation: define resource limits for your pods so they don't destroy your cluster and affect your applications. On this subject, I have some more tips in this KubiLab. 

To finish, I decided to ask the million-dollar question:

Well, that is a complex answer, but it was great. We have also already talked about these topics on our blog. If you want to know a bit more, access:

1 - Configuring limits and resources (KubiLab)
2 - Using Quotas in your Namespaces (Doc.Kubernetes)
3 - Kubicast on best practices (Listen here)
4 - Monitoring: observability series (Listen here)

Moving on, a great function I noticed is that OpenAI can help with the administration and creation of resources with examples and templates of application deployments. Let's look at some examples here:

OpenAI can also easily provide the steps for installing an application or deploying to your cluster, for example, the Prometheus and Grafana stack, using Helm:

Conclusion

Even though ChatGPT gave good answers to my questions, it is important to remember that every process starts and ends with people. Before marveling at pasteurized and rigid frameworks, we need to consider individuals, the different realities of companies, and their teams. Therefore, it is always essential to gather the team, understand the project requirements, and work together with the real situation.

We still have a way to go in terms of development, but being able to outsource the task of remembering things to a tool can make our lives much simpler and more convenient. Finally, the natural language used by OpenAI will certainly be a great help in this regard, and, as always, review, research, and do not copy and paste directly into your environment, after all, its information came directly from a cluster of other information that may or may not be correct.