Реальный кейс настройки Pod Autoscaling в k8s с точки зрения разработчика

На носу 2026 год, а я хочу поделиться своим путешествием по переводу приложения на инфраструктуру Kubernetes. И самая сложная и интересная часть, как раз, настройка автоскейлинга. Не слишком ли заезженная тема? Думаю нет, потому что я буду рассказывать именно с позиции разработчика приложения, а не девопса. Мне повезло, я без понятия как это всё настраивается. Я буду рассказывать как это всё работает. Конфигов кубера будет минимум, рассуждений и погружений в метрики максимум. В конце оставил TL;DR. Поехали?

https://habr.com/ru/articles/973936/

#kubernetes #hpa #horizontal_pod_autoscaler #keda #ec2 #cadvisor #k8s

Google released #cAdvisor version 0.54.1. https://github.com/google/cadvisor

Google released #cAdvisor version 0.54.0. https://github.com/google/cadvisor

@techviator You're welcome. Don't hesitate to ping me back. I migrated to podman in the previous 3-4 months and I still need to understand a lot (how to do #healthcheck on quadlet, how can I replace #cadvisor to #monitor #quadlet)

Ich monitore Docker Umgebungen gerne via #cadvisor und #Prometheus in #Grafana.

Gerade aus Spaß mal #Telegraf statt #cadvisor zum sammeln von Container Metriken eingebaut und die gleichen Werte erwartet...

Aber die Werte sind bei weitem nicht gleich!

Telegraf zeigt mir wesentlich geringere Werte an als der cadvisor...

Hat da eins Erfahrungen mit?

Welchem Collector will ich eher trauen?

Finally completed my rebuild of my #Grafana #prometheus #vps stack.

The old one was hosed in a 3 month battle with a #cryptominer
It was #docker but they kept fucking the Prometheus container.

I rebuild everything from scratch. The panels are integrated into a single JSON file, rather than in libraries.

The stack is now #podman. Rootless execution.
But I couldn't get it to get #cadvisor to feed it.
So I got a dodgy scraper script.
But even with nice, it loads the low tier VPS to 14%

#selfhosting

Google released #cAdvisor version 0.53.0. https://github.com/google/cadvisor

After upgrading my VictoriaMetrics cluster, I've found that I no longer get container resource usage for worker nodes (but still get it from control plane nodes).

From what I can see vmagent is still scraping the worker nodes and isn't returning errors, but the scraped metrics just aren't going anywhere?

I've completed removed all drop configs so it doesn't appear to be related to that.

Just really confused. Tempted to file a bug report but not skilled enough with the product to be sure its not a skill issue.

#VictoriaMetrics #cadvisor #Prometheus #Observability #Kubernetes

I like deep dives and I cannot lie #kubecon #prometheus #cadvisor

Google released #cAdvisor version 0.52.1. https://github.com/google/cadvisor

Google released #cAdvisor version 0.52.0. https://github.com/google/cadvisor

Мониторинг контейнеров с помощью cAdvisor

cAdvisor — это инструмент, предназначенный для сбора информации о состоянии контейнеров. В этой статье мы посмотрим, как с помощью cAdvisor можно отслеживать использование ресурсов контейнеров и их производительность.

https://habr.com/ru/companies/otus/articles/887396/

#cadvisor #docker #мониторинг #it_инфраструктура

Google released #cAdvisor version 0.49.2. https://github.com/google/cadvisor

#telegraf or #cadvisor
To monitor #docker stacks?

#portainer is nice for standalone, but if you want to plug into #grafana?

I finally finished writing up a post on #Kubernetes Memory metrics.

Technically, it should be called #cAdvisor memory, but then the only people who would know that it's about kubernetes probably don't need the contents of the post.

https://hegdenu.net/posts/kubernetes-memory/

#k8s

Hmm #cadvisor doesn't seem to pickup #podman containers running as a regular user.

Since I'm having trouble figuring it out on my own, and the docs are useless, and the comments in the code are tautological:

Does anyone on here know how frequently kubelet scrapes the CPU usage metrics from the filesystem?

#kubernetes #k8s #cadvisor #kubelet

@akashrajpurohit Yes, I have a question about #cAdvisor.

I configured it for each of my multiple Docker with #Prometheus and #Grafana.

In Grafana, I would like to get the hostname instead of the IP:port. Can we use tags in the Prometheus configuration?

Do you have a Grafana dashboard you prefer, or do you build your own?

Thanks

@akashrajpurohit I didn’t know this one. Until now I was using telegraf, until now. But the data was limited. I will test #cadvisor. Thanks for the advice!

If you #selfhost services via #docker and use #grafana and #prometheus then one amazing exporter that you can try is #cadvisor.

It helps you collect stats from your docker containers and I have been using it for quite some time.

Recently, I realised that the cAdvisor container ends up taking a lot of CPU resources which did not feel right, if you are facing the same then don't worry, I have a solution for you.

https://akashrajpurohit.com/blog/optimizing-cadvisor-for-lower-cpu-usage/

This helped me drop the CPU usage to more than 50%