How to connect two Kubernetes Jobs without using a Service?

Question

I have a deployment that consists on one Job that contains two containers. Due to distribution and economical reasons, I want these two containers to be deployed independently. However, they are still interconnected. This means there should be network communication between both of them.

As far as I know, you can not create a Service over a Job. You need it to be at least a simple Pod. However, this adds another element and modifies the equation: I want to split the Job into two Jobs, not into one Job, one Pod and one Service.

So, with this as context, is there a way to connect to Jobs directly without creating a service? They both are going to be on the same cluster, using the same internal DNS, in different machines.

Basically, something like, from Job1, connect to job1.job1namespace:8080.

Answer 1

You might be able (assuming you're running at least Kubernetes 1.20) to leverage a beta feature for the Pod spec that allows a Pod FQDN to be set in DNS.

I am uncertain if this extends to Job resources in the minimum version (v1.20), but because it is part of the Pod specification, I would assume so between Jobs. In this example, assuming setHostnameAsFQDN: true is set for that definition, a FQDN like busybox-1.default-subdomain.my-namespace.svc.cluster-domain.example (for a name metadata field called busybox for that spec) would be valid, and should be accessible from a different Job.

I was able to test that Job resources would accept setHostnameAsFQDN in the definition, and it seemed to without issue:

apiVersion: batch/v1
kind: Job
metadata:
  name: pi
spec:
  template:
    spec:
      containers:
      - name: pi
        image: perl
        command: ["perl",  "-Mbignum=bpi", "-wle", "print bpi(2000)"]
      restartPolicy: Never
      setHostnameAsFQDN: true
  backoffLimit: 4

on Kubernetes v1.21.1.