jkremser

cluster setup

Spawn the cluster or just use existing kubeconfig

git clone https://github.com/jkremser/kubecon-2025-eu && cd kubecon-2025-eu/infra/gcp
source .env .secret
./setup-gcp.sh bootstrap
k kc use demo
cd so && setup-llama.sh

jkremser

# helm upgrade -i kserve community-charts/kserve --version 1.0.1 --set kserve.controller.deploymentMode=RawDeployment

# kubectl apply -f https://github.com/cert-manager/cert-manager/releases/download/v1.17.0/cert-manager.yaml
# kubectl rollout status -ncert-manager deploy/cert-manager-webhook
# helm upgrade -i kserve-crd oci://ghcr.io/kserve/charts/kserve-crd --version v0.15.0-rc1
# helm upgrade -i kserve oci://ghcr.io/kserve/charts/kserve --version v0.15.0-rc1 --set kserve.controller.deploymentMode=RawDeployment
# kubectl rollout status deploy/kserve-controller-manager
apiVersion: v1
kind: Secret
metadata:

jkremser

// polygon
linear_extrude(20, twist=140,slices=500,scale=0)
translate([-1,0,0])
polygon([[-5,-5],[10,0],[6,4],[5,10],[3,6],[-5,2]]);




// svg
linear_extrude(800, scale=.4, twist=120, slices=11)

jkremser

// ==UserScript==
// @name         to ical
// @namespace    http://tampermonkey.net/
// @version      2025-01-20
// @description  Make ical
// @author       jkremser
// @match        https://objednavky.obedyvbrne.cz/faces/secured/objednavky.jsp
// @icon         https://www.google.com/s2/favicons?sz=64&domain=obedyvbrne.cz
// @require      https://raw.githubusercontent.com/matthiasanderer/icsFormatter/refs/heads/master/icsFormatter.js
// @grant        none

jkremser

pre-fetch images on nodes:

• https://github.com/che-incubator/kubernetes-image-puller
• https://github.com/senthilrch/kube-fledged
• https://github.com/stackrox/image-prefetcher
• https://github.com/dcherman/image-cache-daemon

kubelet param: If you would like to enable parallel image pulls, you can set the field serializeImagePulls to false in the kubelet configuration. With serializeImagePulls set to false, image pull requests will be sent to the image service immediately, and multiple images will be pulled at the same time.

jkremser

# two clusters
k config get-contexts -oname

# eu@aws
k --context cluster-eu@aws get no
k --context cluster-eu@aws describe node | grep -A5 "System Info"
k --context cluster-eu@aws get svc -n ingress-nginx
dig +short a1f9cb3961e1a4d11af80365d0c5fa28-31f9021cb00929cd.elb.eu-west-2.amazonaws.com
curl -s http://ip-api.com/json/18.132.100.90  | jq

jkremser

1*
curl -s https://raw.githubusercontent.com/zroubalik/test-data/main/resources/create_resources.sh | \
OTHER_DEPLOYMENTS=9 \
OTHER_STATEFUL_SETS=1 \
bash -s -- 6 1 0 t



zaple vsechno to neleakovalo, dokad sem neudelal SOs (1*) - monitoring (1,1,1)

jkremser

apiVersion: v1
kind: Service
metadata:
  annotations:
    cloud.google.com/neg: '{"ingress":true}'
  labels:
    k8s-app: nvidia-gpu-device-plugin
  name: gpu-metrics
 namespace: kube-system

jkremser

Kedify AI use-cases

general areas

1. GPU scheduling (DRA or some better support for nvidia plugin)
2. AI related scalers (metrics)
3. Glue - for operating the auto-scaleable models
4. Adding and releasing GPU enabled nodes dynamically to k8s cluster
5. Multi-cluster (overflow to a cluster that have GPUs 'attached' say spotinstances that are cheaper)
6. Maybe introduce a new CRD on the KEDA level that will support heterogeneous settings across replicas (currently all replicas of a single deployment or statefulset are identical)

jkremser

k exec -ti deployments/psql -n kedify -c psql -- pg_dump --user [email protected] --host 127.0.0.1 -d kedify --inserts > db.dump
psql --user postgres --host 127.0.0.1 -f db.dump


#docker cp db.dump 02a749fdba3a:.

#docker exec -ti 02a749fdba3a  bash
#psql --user postgres --host 127.0.0.1 -f db.dump