Skip to content

Google Cloud Platform

Google Cloud Platform provides seamless support for Kubernetes. Therefore, Pachyderm is fully supported on Google Kubernetes Engine (GKE). The following section walks you through deploying a Pachyderm cluster on GKE.


If this is the first time you use the SDK, follow the Google SDK QuickStart Guide.


When you follow the QuickStart Guide, you might update your ~/.bash_profile and point your $PATH at the location where you extracted google-cloud-sdk. However, Pachyderm recommends that you extract the SDK to ~/bin.


You can install kubectl by using the Google Cloud SDK and running the following command:

gcloud components install kubectl

Deploy Kubernetes

To create a new Kubernetes cluster by using GKE, run:

CLUSTER_NAME=<any unique name, e.g. "pach-cluster">

GCP_ZONE=<a GCP availability zone. e.g. "us-west1-a">

gcloud config set compute/zone ${GCP_ZONE}

gcloud config set container/cluster ${CLUSTER_NAME}

MACHINE_TYPE=<machine type for the k8s nodes, we recommend "n1-standard-4" or larger>

# By default the following command spins up a 3-node cluster. You can change the default with `--num-nodes VAL`.
gcloud container clusters create ${CLUSTER_NAME} --scopes storage-rw --machine-type ${MACHINE_TYPE}

# By default, GKE clusters have RBAC enabled. To allow 'pachctl deploy' to give the 'pachyderm' service account
# the requisite privileges via clusterrolebindings, you will need to grant *your user account* the privileges
# needed to create those clusterrolebindings.
# Note that this command is simple and concise, but gives your user account more privileges than necessary. See
# for the complete list of privileges that the
# pachyderm serviceaccount needs.
kubectl create clusterrolebinding cluster-admin-binding --clusterrole=cluster-admin --user=$(gcloud config get-value account)


You must create the Kubernetes cluster by using the gcloud command-line tool rather than the Google Cloud Console, as you can grant the storage-rw scope through the command-line tool only.

This migth take a few minutes to start up. You can check the status on the GCP Console. A kubeconfig entry is automatically generated and set as the current context. As a sanity check, make sure your cluster is up and running by running the following kubectl command:

# List all pods in the kube-system namespace.
kubectl get pods -n kube-system

System Response:

NAME                                                     READY     STATUS    RESTARTS   AGE
event-exporter-v0.1.7-5c4d9556cf-fd9j2                   2/2       Running   0          1m
fluentd-gcp-v2.0.9-68vhs                                 2/2       Running   0          1m
fluentd-gcp-v2.0.9-fzfpw                                 2/2       Running   0          1m
fluentd-gcp-v2.0.9-qvk8f                                 2/2       Running   0          1m
heapster-v1.4.3-5fbfb6bf55-xgdwx                         3/3       Running   0          55s
kube-dns-778977457c-7hbrv                                3/3       Running   0          1m
kube-dns-778977457c-dpff4                                3/3       Running   0          1m
kube-dns-autoscaler-7db47cb9b7-gp5ns                     1/1       Running   0          1m
kube-proxy-gke-pach-cluster-default-pool-9762dc84-bzcz   1/1       Running   0          1m
kube-proxy-gke-pach-cluster-default-pool-9762dc84-hqkr   1/1       Running   0          1m
kube-proxy-gke-pach-cluster-default-pool-9762dc84-jcbg   1/1       Running   0          1m
kubernetes-dashboard-768854d6dc-t75rp                    1/1       Running   0          1m
l7-default-backend-6497bcdb4d-w72k5                      1/1       Running   0          1m

If you don't see something similar to the above output, you can point kubectl to the new cluster manually by running the following command:

# Update your kubeconfig to point at your newly created cluster.
gcloud container clusters get-credentials ${CLUSTER_NAME}

Deploy Pachyderm

To deploy Pachyderm we will need to:

  1. Create storage resources,
  2. Install the Pachyderm CLI tool, pachctl, and
  3. Deploy Pachyderm on the Kubernetes cluster

Set up the Storage Resources

Pachyderm needs a GCS bucket and a persistent disk to function correctly. You can specify the size of the persistent disk, the bucket name, and create the bucket by running the following commands:

# For the persistent disk, 10GB is a good size to start with.
# This stores PFS metadata. For reference, 1GB
# should work fine for 1000 commits on 1000 files.
STORAGE_SIZE=<the size of the volume that you are going to create, in GBs. e.g. "10">

# The Pachyderm bucket name needs to be globally unique across the entire GCP region.
BUCKET_NAME=<The name of the GCS bucket where your data will be stored>

# Create the bucket.
gsutil mb gs://${BUCKET_NAME}

To check that everything has been set up correctly, run:

gsutil ls
# You should see the bucket you created.

Install pachctl

pachctl is a command-line utility for interacting with a Pachyderm cluster. You can install it locally as follows:

# For macOS:
brew tap pachyderm/tap && brew install pachyderm/tap/pachctl@1.11

# For Linux (64 bit) or Window 10+ on WSL:

$ curl -o /tmp/pachctl.deb -L && sudo dpkg -i /tmp/pachctl.deb

You can then run pachctl version --client-only to check that the installation was successful.

pachctl version --client-only

Deploy Pachyderm on the Kubernetes cluster

Now you can deploy a Pachyderm cluster by running this command:

pachctl deploy google ${BUCKET_NAME} ${STORAGE_SIZE} --dynamic-etcd-nodes=1

System Response:

serviceaccount "pachyderm" created
storageclass "etcd-storage-class" created
service "etcd-headless" created
statefulset "etcd" created
service "etcd" created
service "pachd" created
deployment "pachd" created
service "dash" created
deployment "dash" created
secret "pachyderm-storage-secret" created

Pachyderm is launching. Check its status with "kubectl get all"
Once launched, access the dashboard by running "pachctl port-forward"


Pachyderm uses one etcd node to manage Pachyderm metadata.


If RBAC authorization is a requirement or you run into any RBAC errors see Configure RBAC.

It may take a few minutes for the pachd nodes to be running because Pachyderm pulls containers from DockerHub. You can see the cluster status with kubectl, which should output the following when Pachyderm is up and running:

kubectl get pods

System Response:

NAME                     READY     STATUS    RESTARTS   AGE
dash-482120938-np8cc     2/2       Running   0          4m
etcd-0                   1/1       Running   0          4m
pachd-3677268306-9sqm0   1/1       Running   0          4m

If you see a few restarts on the pachd pod, you can safely ignore them. That simply means that Kubernetes tried to bring up those containers before other components were ready, so it restarted them.

Finally, assuming your pachd is running as shown above, set up forward a port so that pachctl can talk to the cluster.

# Forward the ports. We background this process because it blocks.
pachctl port-forward &

And you're done! You can test to make sure the cluster is working by running pachctl version or even creating a new repo.

pachctl version

System Response:

pachctl             1.9.7
pachd               1.9.7

Increasing Ingress Throughput

One way to improve Ingress performance is to restrict Pachd to a specific, more powerful node in the cluster. This is accomplished by the use of node-taints in GKE. By creating a node-taint for pachd, you configure the Kubernetes scheduler to run only the pachd pod on that node. After that’s completed, you can deploy Pachyderm with the --pachd-cpu-request and --pachd-memory-request set to match the resources limits of the machine type. And finally, you need to modify the pachd deployment so that it has an appropriate toleration:

- key: "dedicated"
  operator: "Equal"
  value: "pachd"
  effect: "NoSchedule"

Increasing upload performance

The most straightfoward approach to increasing upload performance is to leverage SSD’s as the boot disk in your cluster because SSDs provide higher throughput and lower latency than HDD disks. Additionally, you can increase the size of the SSD for further performance gains because the number of IOPS increases with disk size.

Increasing merge performance

Performance tweaks when it comes to merges can be done directly in the Pachyderm pipeline spec. More specifically, you can increase the number of hashtrees (hashtree spec) in the pipeline spec. This number determines the number of shards for the filesystem metadata. In general this number should be lower than the number of workers (parallelism spec) and should not be increased unless merge time (the time before the job is done and after the number of processed datums + skipped datums is equal to the total datums) is too slow.

Last update: July 16, 2020