After a *very suspicious* Zoom servers problem :fa...
# general
After a very suspicious Zoom servers problem 😶‍🌫️ last week, the rescheduled webinar “Destroy latency and kill pods: how not to manage your K8s resources” with @Natan Yellin is finally happening today. Natan is going to list some of the most common misconceptions about Kubernetes resource management and how not to get them wrong. He'll do his best to destroy some clusters and performance along the way. Be sure to tune in! November 8, 7 PM CET, 1 PM CDT
And we are live with Natan on YouTube with Natan and Mariya!

check it out here!
Question from Natan on YouTube: what is the best approach to run a database? As a pod? Or inside a machine? Or some paas service?
PaaS is easiest if it works, but sometimes that’s not viable, in which case use a stateful set or an appropriate Kubernetes Operator
There’s a misconception that Kubernetes isn’t for stateful sets, despite the fact that large companies do run stateful sets on kubernetes
If you’re running Prometheus on Kubernetes, for example, you’re already running a stateful set which is storing all your metrics!
To give a concrete example of the tradeoffs, we’re using a postgres as a service for the cloud, but for our on-prem customers we deploy a postgres instance inside their clusters. So there is no one solution! Depends on the scenario
Thanks @Natan Yellin, that was a great overview!
My pleasure, I hope it helped!
@Natan Yellin have you done testing around limits with different kernel versions? It looks like there have been several CFS kernel changes over the past few years, with some changes correlated with Kubernetes issues.
@Dave Thompson I haven’t tested enough, but afaik the fundamental issue remains. There was a rather nasty CFS bug once, but the issue with limits occurs regardless.