The following provides a means to run ClusterODM on your master node and a series of NodeODM instances on worker nodes in a kubernetes cluster. The configuration uses GPU's if they are setup on your kubernetes cluster.
Before deploying it is important to make the following changes In the file k8clusternodeodm.yml
change EXTERNALIPADDRESS to the external ip adddress of your k8 cluster change the number of replicas (REPNUM) for nodeodm to the number of nodes in your cluster (I suggest 1 nodeodm per cluster node)
On the master/head node run the command:
kubectl apply -f k8clusternodeodm.yml
This will then start up 2 deployments
- Deployment 1 with ClusterODM
- Deployment 2 with N replicas of NodeODM.
Once the pods are running you should be able to connect to ClusterODM on http://EXTERNALIPADDRESS:10000 and then add the NodeODM nodes.
On the master/head node issue the command:
kubectl get pods -o wide
This will supply you with the IP Address details that you need to enter into the ClsuterODM configuration page or add via telnet session.
The following shows the type of output you should get and the IP address details NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES nodeodm-deployment-6d454b88d5-2z5rh 1/1 Running 0 157m 172.29.8.86 gigganode02 nodeodm-deployment-6d454b88d5-5cksx 1/1 Running 0 148m 172.29.26.145 gigganode03 nodeodm-deployment-6d454b88d5-78wd2 1/1 Running 0 157m 172.29.6.183 gigganode01 nodeodm-deployment-6d454b88d5-7m8pm 1/1 Running 0 148m 172.29.26.130 gigganode03 nodeodm-deployment-6d454b88d5-bfgps 1/1 Running 0 148m 172.29.26.153 gigganode03
Add the ClusterODM as a Processing Node using the details EXTERNALIPADDRESS:3000