Using Job Composer App with kubernetes instead of Slurm

Get Help
1

Hello there,

is it possible to use the Job Composer App to submit job using Kubernetes adapter. Looking at the below error I received after submitting the job, it seems that the back-end is still using Slurm which is not configured.

Is there anything I need to configure for using the “Job Composer App” to submit in kubernetes. oidc and the hook for kubernetes is working.

Here is my Resource Manager Config for the kubernetes adapter.

--- 

v2: 

metadata: 

title: "cluster" 

# you may not want a login section. There may not be a login node 

# for your kuberenetes cluster 

login: 

host: "clustercom.com" 

job: 

adapter: "kubernetes" 

config_file: "/home/$USER/.kube/config" 

cluster: "kubernetes" 

context: "kubernetes-admin@kubernetes" 

bin: "/usr/local/bin/kubectl" 

username_prefix: "" 

namespace_prefix: "" 

all_namespaces: false 

auto_supplemental_groups: false 

server: 

endpoint: "https://api.clustercom.com" 

# cert_authority_file: "/etc/pki/tls/certs/kubernetes-ca.crt" 

auth: 

type: "oidc" 

mounts: [] 

batch_connect: 

ssh_allow: true 

basic: 

header: "#!/bin/bash"

error after submitting a job from the Job Composer


 App 971 output: [2023-03-23 20:42:27 +0000 ]  INFO "method=PUT path=/pun/sys/myjobs/workflows/1/submit format=html controller=WorkflowsController action=submit status=500 error='NoMethodError: undefined method `[]' for nil:NilClass' duration=26.79 view=0.00 db=7.16" 

App 971 output: [2023-03-23 20:42:27 +0000 ] FATAL "" 

App 971 output: [2023-03-23 20:42:27 +0000 ] FATAL "NoMethodError (undefined method `[]' for nil:NilClass):" 

App 971 output: [2023-03-23 20:42:27 +0000 ] FATAL "" 

App 971 output: [2023-03-23 20:42:27 +0000 ] FATAL "app/models/resource_mgr_adapter.rb:36:in `qsub'\napp/models/workflow.rb:291:in `each'\napp/models/workflow.rb:291:in `submit_jobs'\napp/models/workflow.rb:259:in `submit'\napp/controllers/workflows_controller.rb:181:in `block in submit'\napp/controllers/workflows_controller.rb:176:in `submit'"

I only have the default files in the compose APP (myjobs). Do I need a submit.yml.erb ? for it.

[root@ondemand apps]# ls /etc/ood/config/apps/myjobs/templates/python/
hello.py      manifest.yml  script.sh

Please bear with me :slight_smile: , my first week with Ondemand, still learning…

thanks,
mohbarry

2

Hi and welcome!

While I’m not quite sure what that error means, I would say no that the kubernetes adapter is built mostly for interactive work. That is, it doesn’t really have support for batch jobs.

3

thanks Jeff for the clarification !

4

Jeff, Yinping Ma from Peking University is here in the booth asking about possible support for the Job Constructor on Kubernetes resources. Is this something that could be done?

5

Maybe at some point. The OOD job composer is being rewritten at the moment without kubernetes support. Once the UI is updated, it’s possible we could start to support k8s batch processing.

6

This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.