#7190 [V2] Upgrade Kubernetes version to 1.34 in demo and dev environments Pull request opened by pingsutw Why are the changes needed? The demo/sandbox environment was using Kubernetes 1.29 (k3s

v1.29.0-k3s1

), which is outdated. Upgrading to 1.34 brings security patches, performance improvements, and access to newer Kubernetes features. What changes were proposed in this pull request? • Bump k3s base image from

v1.29.0-k3s1

to

v1.34.6-k3s1

in

docker/demo-bundled/Dockerfile

. • Bump kind node image from

v1.26.0

to

v1.34.0

in

executor/DEVELOPMENT.md

and

dataproxy/DEVELOPMENT.md

. • Fix

docker/demo-bundled/bin/k3d-entrypoint-cgroupv2.sh

by replacing

busybox xargs

with the standalone

xargs

. Why the cgroupv2 script change is needed After bumping to

rancher/k3s:v1.34.6-k3s1

,

make sandbox-run

hung at

Waiting for kubeconfig...

because the container exited with code 1 before k3s ever started. Root cause: 1. The k3d entrypoint runs

/bin/k3d-entrypoint-cgroupv2.sh

before launching k3s. That script (copy-pasted from moby/moby's dind helper) calls

busybox xargs -rn1 < /sys/fs/cgroup/cgroup.procs > /sys/fs/cgroup/init/cgroup.procs || :

to evacuate processes from the root cgroup into

/init

so that writing to

cgroup.subtree_control

does not fail with

EBUSY

. 2. The busybox shipped inside

rancher/k3s:v1.34.6-k3s1

was built without the

xargs

applet, so

busybox xargs

fails with

xargs: applet not found

. The

|| :

swallows the error, so the script appears to continue. 3. But because the processes were never actually evacuated, the next line —

sed ... > /sys/fs/cgroup/cgroup.subtree_control

— fails with

sed: write error

(EBUSY).

set -o errexit

is in effect, so the entrypoint exits 1. k3s never starts, the kubeconfig is never written, and

make sandbox-run

waits forever. 4. The standalone

/bin/xargs

binary does exist in the k3s v1.34 image, so simply dropping the

busybox

prefix is enough — the cgroup evacuation then works and k3s boots normally. Entrypoint log captured from the failing container before the fix:

[2026-04-10T06:55:06+00:00] Running k3d entrypoints...
[2026-04-10T06:55:06+00:00] Running /bin/k3d-entrypoint-cgroupv2.sh
[2026-04-10T06:55:06+00:00] [CgroupV2 Fix] Evacuating Root Cgroup ...
xargs: applet not found
sed: write error

How was this patch tested? Rebuilt the demo image locally and ran `make sandbox-run`:

$ kubectl --context flyte-demo get nodes
NAME           STATUS   ROLES           AGE   VERSION
3494856c3ac2   Ready    control-plane   23s   v1.34.6+k3s1

$ kubectl --context flyte-demo get pods -n flyte
NAME                               READY   STATUS              RESTARTS   AGE
docker-registry-5b9f54d5d5-5wwpz   0/1     ContainerCreating   0          16s
flyte-binary-7b57dd88c6-dxcqg      0/1     Init:0/1            0          16s
flyte-console-69b86957f6-dpcd6     0/1     Running             0          16s
minio-6cb99cfc49-rbv5p             0/1     Running             0          16s

Cluster comes up on

v1.34.6+k3s1

, kubeconfig is written and merged into

~/.kube/config

, and all demo workloads schedule normally. Labels • changed: Kubernetes version upgrade in demo and dev environments. Check all the applicable boxes • I updated the documentation accordingly. • All new and existing tests passed. • All commits are signed-off. •

main

• #6583 • [V2] Upgrade Kubernetes version to 1.34 in demo and dev environments 👈 flyteorg/flyte