Lammps on Google Cloud

This basic tutorial will walk through creating a MiniCluster to run LAMMPS! You should have already setup your workspace

Create Cluster

Let’s use gcloud to create a cluster, and we are purposefully going to choose a very small node type to test. Note that I choose us-central1-a because it tends to be cheaper (and closer to me). First, here is our project for easy access:


Then create your cluster as follows:

$ gcloud container clusters create flux-cluster --project $GOOGLE_PROJECT \
    --zone us-central1-a --machine-type n1-standard-1 \
    --num-nodes=4 --enable-network-policy --tags=flux-cluster --enable-intra-node-visibility

If you need a particular Kubernetes version:

$ gcloud container clusters create flux-cluster --project $GOOGLE_PROJECT \
    --zone us-central1-a --cluster-version 1.23 --machine-type n1-standard-1 \
    --num-nodes=4 --enable-network-policy --tags=flux-cluster --enable-intra-node-visibility

Note that not all of the flags above might be necessary - I did a lot of testing to get this working and didn’t go back and try removing things after the fact! If you want to use cloud dns instead (after enabling it)

$ gcloud beta container clusters create flux-cluster --project $GOOGLE_PROJECT \
    --zone us-central1-a --cluster-version 1.23 --machine-type n1-standard-1 \
    --num-nodes=4 --enable-network-policy --tags=flux-cluster --enable-intra-node-visibility \
    --cluster-dns=clouddns \

In your Google cloud interface, you should be able to see the cluster! Note this might take a few minutes.


I also chose a tiny size (nodes and instances) anticipating having it up longer to figure things out.

Get Credentials

Next we need to ensure that we can issue commands to our cluster with kubectl. To get credentials, in the view shown above, select the cluster and click “connect.” Doing so will show you the correct statement to run to configure command-line access, which probably looks something like this:

$ gcloud container clusters get-credentials flux-cluster --zone us-central1-a --project $GOOGLE_PROJECT
Fetching cluster endpoint and auth data.
kubeconfig entry generated for flux-cluster.

Finally, use cloud IAM to ensure you can create roles, etc.

$ kubectl create clusterrolebinding cluster-admin-binding --clusterrole cluster-admin --user $(gcloud config get-value core/account) created

At this point you should be able to get your nodes:

$ kubectl get nodes
NAME                                            STATUS   ROLES    AGE     VERSION
gke-flux-cluster-default-pool-f103d9d8-379m   Ready    <none>   3m41s   v1.23.14-gke.1800
gke-flux-cluster-default-pool-f103d9d8-3wf9   Ready    <none>   3m42s   v1.23.14-gke.1800
gke-flux-cluster-default-pool-f103d9d8-c174   Ready    <none>   3m42s   v1.23.14-gke.1800
gke-flux-cluster-default-pool-f103d9d8-zz1q   Ready    <none>   3m42s   v1.23.14-gke.1800

Deploy Operator

To deploy the Flux Operator, choose one of the options here to deploy the operator. Whether you apply a yaml file, use flux-cloud or clone the repository and make deploy you will see the operator install to the operator-system namespace.

For a quick “production deploy” from development, the Makefile has a directive that will build and push a test tag (you’ll need to edit DEVIMG to be one you can push to) and then generate a yaml file targeting that image, e.g.,

$ make test-deploy
$ kubectl apply -f examples/dist/flux-operator-dev.yaml

or the production version:

$ kubectl apply -f examples/dist/flux-operator.yaml
... created created created created created created
configmap/operator-manager-config created
service/operator-controller-manager-metrics-service created
deployment.apps/operator-controller-manager created

Ensure the operator-system namespace was created:

$ kubectl get namespace
NAME              STATUS   AGE
default           Active   6m39s
kube-node-lease   Active   6m42s
kube-public       Active   6m42s
kube-system       Active   6m42s
operator-system   Active   39s
$ kubectl describe namespace operator-system
Name:         operator-system
Labels:       control-plane=controller-manager
Annotations:  <none>
Status:       Active

Resource Quotas
  Name:                              gke-resource-quotas
  Resource                           Used  Hard
  --------                           ---   ---
  count/ingresses.extensions         0     100
  count/  0     100
  count/jobs.batch                   0     5k
  pods                               1     1500
  services                           1     500

No LimitRange resource.

And you can find the name of the operator pod as follows:

$ kubectl get pod --all-namespaces
operator-system   operator-controller-manager-56b5bcf9fd-m8wg4               2/2     Running   0          73s

Create Flux Operator namespace

Make your namespace for the flux-operator custom resource definition (CRD), which is the yaml file above that generates the MiniCluster:

$ kubectl create namespace flux-operator

Now let’s run a short experiment with LAMMPS!

Custom Resource Definition

The Custom Resource Definition (CRD) defines our Mini Cluster, and is what we hand to the flux operator to create it. Here is the CRD for a small lammps run.

kind: MiniCluster
  name: flux-sample
  namespace: flux-operator
  # Number of pods to create for MiniCluster
  size: 4

  # Disable verbose output
    quiet: true

  # This is a list because a pod can support multiple containers
    # The container URI to pull (currently needs to be public)
    - image:

      # You can set the working directory if your container WORKDIR is not correct.
      workingDir: /home/flux/examples/reaxff/HNS
      command: lmp -v x 2 -v y 2 -v z 2 -in in.reaxc.hns -nocite

You can save the above file as minicluster-lammps.yaml to get started.

Create the Lammps Job

Now let’s apply the custom resource definition to create the lammps mini cluster! The file we generated above should be in your present working directory. Importantly, we have set localDeploy to false because we need to create volume claims and not local host mounts for shared resources.

$ kubectl apply -f minicluster-lammps.yaml

There are different ways to see logs for pods. First, see pods running and state. You probably want to wait until the state changes from ContainersCreating to Running because this is where we are pulling the chonker containers.

$ kubectl get -n flux-operator pods

If you need to debug (or see general output for a pod about creation) you can do:

$ kubectl -n flux-operator describe pods flux-sample-0-742bm

And finally, the most meaty piece of metadata is the log for the pod, where the Flux Operator will be setting things up and starting flux.

# Add the -f to keep it hanging
$ kubectl -n flux-operator logs flux-sample-0-742bm -f

To shell into a pod to look around (noting where important flux stuff is)

$ kubectl exec --stdin --tty -n flux-operator flux-sample-0-742bm -- /bin/bash
ls /mnt/curve
ls /etc/flux
ls /etc/flux/config

To get logs for the operator itself:

$ kubectl logs -n operator-system operator-controller-manager-56b5bcf9fd-j2g75

If you need to run in verbose (non-test) mode, set test to false in the minicluster-lammps.yaml. And make sure to clean up first:

$ kubectl delete -f minicluster-lammps.yaml

and wait until the pods are gone:

$ kubectl get -n flux-operator pods
No resources found in flux-operator namespace.

Observations about comparing this to MiniKube (local):

  • The containers that are large actually pull!

  • The startup times of the different pods vary quite a bit.

If you want to run the same workflow again, use kubectl delete -f with the file and apply it again. I wound up running with test set to true, and then saving the logs:

$ kubectl -n flux-operator logs flux-sample-0-qc5z2 > lammps.out

For fun, here is the first successful run of Lammps using the Flux Operator on GCP ever!


Then to delete your lammps MiniCluster:

$ kubectl delete -f minicluster-lammps.yaml

Clean up

Whatever tutorial you choose, don’t forget to clean up at the end! You can optionally undeploy the operator (this is again at the root of the operator repository clone)

$ make undeploy

Or the file you used to deploy it:

$ kubectl delete -f examples/dist/flux-operator.yaml
$ kubectl delete -f examples/dist/flux-operator-dev.yaml

And then to delete the cluster with gcloud:

$ gcloud container clusters delete --zone us-central1-a flux-cluster

I like to check in the cloud console to ensure that it was actually deleted.

Customization and Debugging


When I first created my cluster, the nodes could not see one another. I added a few flags for networking, and looked at firewalls as follows:

$ gcloud container clusters describe flux-cluster --zone us-central1-a | grep clusterIpv4Cidr

I didn’t ultimately change anything, but I found this useful.

Last update: Mar 30, 2023