kcore now has initial container operations. This post documents what you can run today, with conservative expectations.
This is early support. Treat it as an operator workflow for labs and controlled environments, not as a full container orchestration platform.
What this currently covers
- Node-scoped container lifecycle via
kctl --node .... - Create, list, inspect, start, stop, and delete container workloads.
- Container runtime CLI auto-detection with
nerdctlpreferred (ordockerfallback). - Optional network attachment by name when the network already exists and is bridge-backed.
What this does not claim yet
- No claim of Kubernetes-style scheduling or orchestration semantics.
- No claim of production-grade HA behavior for this container path yet.
- No claim of full parity with the VM workflow in all dimensions.
Prerequisites
- An installed kcore node reachable at
<NODE_IP>:9091. kctlconfigured on your operator machine.nerdctl(preferred) ordockeravailable on the node runtime.- If you pass
--network, ensure that network already exists.
1) Create a container
kctl --node <NODE_IP>:9091 create container web-01 \
--image nginx:alpine \
--port 8080:80 \
--env NGINX_ENTRYPOINT_QUIET_LOGS=1
You can also pass command arguments with repeated --cmd flags.
Keep initial tests simple.
2) List containers
kctl --node <NODE_IP>:9091 get containers
3) Inspect one container
kctl --node <NODE_IP>:9091 get containers web-01
4) Stop and start
kctl --node <NODE_IP>:9091 stop container web-01
kctl --node <NODE_IP>:9091 start container web-01
5) Delete
kctl --node <NODE_IP>:9091 delete container web-01
# force delete if needed:
kctl --node <NODE_IP>:9091 delete container web-01 --force
Troubleshooting notes
- If runtime detection fails, install
nerdctlon the node first. - If network attach fails, verify the target network name and bridge-backed setup.
- Prefer small, stateless test workloads while this support matures.
For technical planning details and phased rollout notes, see the design doc in the main repository: support-for-containerd.md.