Version v0.5 of the documentation is no longer actively maintained. The site that you are currently viewing is an archived snapshot. For up-to-date documentation, see the latest version.


Finding and fixing problems in your Kubeflow deployment

TensorFlow and AVX

There are some instances where you may encounter a TensorFlow-related Python installation or a pod launch issue that results in a SIGILL (illegal instruction core dump). Kubeflow uses the pre-built binaries from the TensorFlow project which, beginning with version 1.6, are compiled to make use of the AVX CPU instruction. This is a recent feature and your CPU might not support it. Check the host environment for your node to determine whether it has this support.


grep -ci avx /proc/cpuinfo


Some components requirement AVX2 for better performance, e.g. TF Serving. To ensure the nodes support AVX2, we added minCpuPlatform arg in our deployment config.

On GCP this will fail in regions (e.g. us-central1-a) that do not explicitly have Intel Haswell (even when there are other newer platforms in the region). In that case, please choose another region, or change the config to other platform newer than Haswell.


On Minikube the Virtualbox/VMware drivers for Minikube are recommended as there is a known issue between the KVM/KVM2 driver and TensorFlow Serving. The issue is tracked in kubernetes/minikube#2377.

We recommend increasing the amount of resources Minikube allocates

minikube start --cpus 4 --memory 8096 --disk-size=40g
  • Minikube by default allocates 2048Mb of RAM for its VM which is not enough for JupyterHub.
  • The larger disk is needed to accommodate Kubeflow’s Jupyter images which are 10s of GBs due to all the extra Python libraries we include.

If you just installed Minikube following instructions from the quick start guide, you most likely created a VM with the default resources. You would want to recreate your Minikube with the appropriate resource settings.

minikube stop
minikube delete
minikube start --cpus 4 --memory 8096 --disk-size=40g

If you encounter a jupyter-xxxx pod in Pending status, described with:

Warning  FailedScheduling  8s (x22 over 5m)  default-scheduler  0/1 nodes are available: 1 Insufficient memory.
  • Then try recreating your Minikube cluster (and re-apply Kubeflow using ksonnet) with more resources (as your environment allows):

RBAC clusters

If you are running on a K8s cluster with RBAC enabled, you may get an error like the following when deploying Kubeflow:

ERROR Error updating roles kubeflow-test-infra.jupyter-role: "jupyter-role" is forbidden: attempt to grant extra privileges: [PolicyRule{Resources:["*"], APIGroups:["*"], Verbs:["*"]}] user=&{  [system:authenticated] map[]} ownerrules=[PolicyRule{Resources:["selfsubjectaccessreviews"], APIGroups:[""], Verbs:["create"]} PolicyRule{NonResourceURLs:["/api" "/api/*" "/apis" "/apis/*" "/healthz" "/swagger-2.0.0.pb-v1" "/swagger.json" "/swaggerapi" "/swaggerapi/*" "/version"], Verbs:["get"]}] ruleResolutionErrors=[]

This error indicates you do not have sufficient permissions. In many cases you can resolve this just by creating an appropriate clusterrole binding like so and then redeploying kubeflow

kubectl create clusterrolebinding default-admin --clusterrole=cluster-admin
  • Replace with the user listed in the error message.

If you’re using GKE, you may want to refer to GKE’s RBAC docs to understand how RBAC interacts with IAM on GCP.

Problems spawning Jupyter pods

If you’re having trouble spawning Jupyter notebooks, check that the pod is getting scheduled

kubectl -n ${NAMESPACE} get pods
  • Look for pods whose name starts with juypter
  • If you are using username/password auth with Jupyter the pod will be named
  • If you are using IAP on GKE the pod will be named

    • Where USER@DOMAIN.EXT is the Google account used with IAP

Once you know the name of the pod do

kubectl -n ${NAMESPACE} describe pods ${PODNAME}
  • Look at the events to see if there are any errors trying to schedule the pod
  • One common error is not being able to schedule the pod because there aren’t enough resources in the cluster.

Pods stuck in Pending state

There are three pods that have 10Gi Persistent Volume Claims (PVCs) that will get stuck in pending state if they are unable to bind their PVC. The three pods are minio, mysql, and vizier-db. Check the status of the PVC requests

kubectl -n ${NAMESPACE} get pvc
  • Look for the status of “Bound”
  • PVC requests in “Pending” state indicate that the scheduler was unable to bind the required PVC.

If you have not configured dynamic provisioning for your cluster, including a default storage class, then you must create a persistent volume for each of the PVCs.

You can use the example below to create a local 10Gi persistent volume.

sudo mkdir /mnt/pv1

kubectl create -f - <<EOF
kind: PersistentVolume
apiVersion: v1
  name: pv-volume1
    storage: 10Gi
    - ReadWriteOnce
    path: "/mnt/pv1"

Repeat two more times creating a new directory and changing the name and path fields to satisfy all three PVCs. Once created the scheduler will successfully start the remaining three pods. The PVs may also be created prior to running any of the commands.


If you are deploying Kubeflow in an OpenShift environment which encapsulates Kubernetes, you will need to adjust the security contexts for the ambassador and Jupyter-hub deployments in order to get the pods to run.

oc adm policy add-scc-to-user anyuid -z ambassador
oc adm policy add-scc-to-user anyuid -z jupyter-hub

Once the anyuid policy has been set, you must delete the failing pods and allow them to be recreated in the project deployment.

You will also need to adjust the privileges of the tf-job-operator service account for TFJobs to run. Do this in the project where you are running TFJobs:

oc adm policy add-role-to-user cluster-admin -z tf-job-operator

Docker for Mac

The Docker for Mac Community Edition now ships with Kubernetes support (1.9.2) which can be enabled from their edge channel. If you decide to use this as your Kubernetes environment on Mac, you may encounter the following error when deploying Kubeflow:

ks apply default
ERROR Attempting to deploy to environment 'default' at '', but cannot locate a server at that address

This error is due to the fact that the default cluster installed by Docker for Mac is actually set to https://localhost:6443. One option is to directly edit the generated environments/default/spec.json file to set the “server” variable to the correct location, then retry the deployment. However, it is preferable to initialize your ksonnet app using the desired kube config:

kubectl config use-context docker-for-desktop
ks init my-kubeflow

403 API rate limit exceeded error

Because ksonnet uses GitHub to pull kubeflow, unless user specifies GitHub API token, it will quickly consume maximum API call quota for anonymous. To fix this issue first create GitHub API token using this guide, and assign this token to GITHUB_TOKEN environment variable.

export GITHUB_TOKEN=<< token >>

ks apply produces error “Unknown variable: env”

Kubeflow requires a specific version of ksonnet. If you run ks apply with an older version of ksonnet you will likely get the error Unknown variable: env as illustrated below:

export KF_ENV=default
ks apply ${KF_ENV}
ERROR Error reading /Users/xxx/projects/devel/go/src/ /Users/xxx/projects/devel/go/src/ Unknown variable: env

  namespace: if params.namespace == "null" then env.namespace else params.namespace

You can check the ksonnet version as follows:

ks version

If your ksonnet version is lower than what is specified in the requirements, please upgrade it and follow the guide to recreate the app.

ksonnet on Windows

There are some known issues with ksonnet and Windows. You might consider alternative solutions.

  • construct base object: Failed to filter components (kubeflow #481)
  • “ks apply” fails to correctly process paths in Windows shell (ksonnet #382)
Last modified 27.04.2019: Capitalized brand names (#666) (3409d72)