GKE on Google Cloud Platform Deployment

OpenMetadata supports the Installation and Running of Application on Google Kubernetes Engine through Helm Charts. However, there are some additional configurations which needs to be done as prerequisites for the same.

Google Kubernetes Engine (GKE) Auto Pilot Mode is not compatible with one of OpenMetadata Dependencies - ElasticSearch. The reason being that ElasticSearch Pods require Elevated permissions to run initContainers for changing configurations which is not allowed by GKE AutoPilot PodSecurityPolicy.

All the code snippets in this section assume the default namespace for kubernetes.

Prerequisites

Cloud Database with CloudSQL and ElasticCloud for GCP as Search Engine

It is recommended to use GCP Cloud SQL services for Database and Elastic Cloud GCP for Search Engine for Production.

We support -

Cloud SQL (MySQL) engine version 8 or higher
Cloud SQL (postgreSQL) engine version 12 or higher
ElasticSearch Engine version 8.X (upto 8.10.X)

We recommend -

CloudSQL to be Multi Zone Available
Elastic Cloud Environment with multiple zones and minimum 2 nodes

Once you have the Database and Search Engine configured and available, update the helm values below for OpenMetadata kubernetes deployments to connect with Database and ElasticSearch.

For Database as PostgreSQL, the use the below config for database values -

Make sure to create CloudSQL and ElasticSearch credentials as Kubernetes Secrets mentioned here.

Also, disable MySQL and ElasticSearch from OpenMetadata Dependencies Helm Charts as mentioned in the FAQs here.

Persistent Volumes with ReadWriteMany Access Modes

OpenMetadata helm chart depends on Airflow and Airflow expects a persistent disk that support ReadWriteMany (the volume can be mounted as read-write by many nodes).

The workaround is to create nfs-server disk on Google Kubernetes Engine and use that as the persistent claim and deploy OpenMetadata by implementing the following steps in order.

Create NFS Share

Provision GCP Persistent Disk for Google Kubernetes Engine

Run the below command to create a gcloud compute zonal disk. For more information on Google Cloud Disk Options, please visit here.

Deploy NFS Server in GKE

Code Samples

Run the commands below and ensure the pods are running.

We create a ClusterIP Service for pods to access NFS within the cluster at a fixed IP/DNS.

Provision NFS backed PV and PVC for Airflow DAGs and Airflow Logs

Update <NFS_SERVER_CLUSTER_IP> with the NFS Service Cluster IP Address for below code snippets. You can get the clusterIP using the following command

Code Samples for PV and PVC for Airflow DAGs

Create Persistent Volumes and Persistent Volume claims with the below command.

Code Samples for PV and PVC for Airflow Logs

Create Persistent Volumes and Persistent Volume claims with the below command.

Change owner and permission manually on disks

Since airflow pods run as non root users, they would not have write access on the nfs server volumes. In order to fix the permission here, spin up a pod with persistent volumes attached and run it once.

Airflow runs the pods with linux user name as airflow and linux user id as 50000.

Run the below command to create the pod and fix the permissions

Once the permissions pod is up and running, execute the below commands within the container.

Create OpenMetadata dependencies Values

Override openmetadata dependencies airflow helm values to bind the nfs persistent volumes for DAGs and logs.

For more information on airflow helm chart values, please refer to airflow-helm.

When deploying openmeteadata dependencies helm chart, use the below command -

The above command uses configurations defined here. You can modify any configuration and deploy by passing your own values.yaml

Once the openmetadata dependencies helm chart deployed, you can then run the below command to install the openmetadata helm chart -

Troubleshooting

Pods are stuck in Pending State due to Persistent Volume Creation Failure

If you came across invalid access type while creating the pvc, and the permission pod is stuck in "pending" state.

The above error might have occurred due to the pvc volumes not setup or pvc volumes are not mounted properly.

Permission pod events

Please validate:

all the prerequisites mentioned in this section
the configuration of dags_pv_pvc.yml file
storageClassName field in YAML file