Add support for 1:1 job per topology assignment

[danielvegamyhre]

Fixes #27

Adds Exclusive struct to the API which has a TopologyKey field. If specified, the JobSet controller will append pod affinity and anti-affinity terms to each job pod template spec, which ensures 1:1 job per topology assignment, and ensures all pods for each job land on the same topology.

Note: I also had to update the make install command in the Makefile to use kubectl create instead of kubectl apply as a workaround for this issue.

Did a manual test on a cluster with 3 node pools, using the topology key cloud.google.com/gke-nodepool to place 1 job per node pool, and it worked.

Test JobSet with 3 replicated jobs, 1 replica each:

apiVersion: batch.x-k8s.io/v1alpha1
kind: JobSet
metadata:
  name: jobpertopology
  labels:
    app.kubernetes.io/name: jobset
    app.kubernetes.io/instance: jobset-example
    app.kubernetes.io/part-of: project
    app.kubernetes.io/managed-by: kustomize
    app.kubernetes.io/created-by: project
spec:
  jobs:
  - name: worker0
    network: 
      enableDNSHostnames: true
    exclusive:
      topologyKey: cloud.google.com/gke-nodepool
    template:
      metadata:
        name: "worker1"
      spec:
        completionMode: Indexed
        parallelism: 2
        completions: 2
        template:
          spec:
            restartPolicy: Never
            containers:
            - name: busybox-short-sleep
              image: busybox:latest
              command:
              - sleep
              - "1000" 
  - name: worker1
    network: 
      enableDNSHostnames: true
    exclusive:
      topologyKey: cloud.google.com/gke-nodepool
    template:
      metadata:
        name: "worker1"
      spec:
        completionMode: Indexed
        parallelism: 2
        completions: 2
        template:
          spec:
            restartPolicy: Never
            containers:
            - name: busybox-short-sleep
              image: busybox:latest
              command:
              - sleep
              - "1000" 
  - name: worker2
    network: 
      enableDNSHostnames: true
    exclusive:
      topologyKey: cloud.google.com/gke-nodepool
    template:
      metadata:
        name: "worker2"
      spec:
        completionMode: Indexed
        parallelism: 2
        completions: 2
        template:
          spec:
            restartPolicy: Never
            containers:
            - name: busybox-short-sleep
              image: busybox:latest
              command:
              - sleep
              - "1000" 

Pod placement:

$ k get pods -o wide
NAME                               READY   STATUS    RESTARTS   AGE   IP         NODE                                     NOMINATED NODE   READINESS GATES
jobpertopology-worker0-0-0-6tg7d   1/1     Running   0          7s    10.8.7.3   gke-jobset4-np2-782cc200-vxq8            <none>           <none>
jobpertopology-worker0-0-1-262vf   1/1     Running   0          7s    10.8.8.3   gke-jobset4-np2-782cc200-946b            <none>           <none>
jobpertopology-worker1-0-0-lc2q6   1/1     Running   0          7s    10.8.5.4   gke-jobset4-np1-004ddd96-58v7            <none>           <none>
jobpertopology-worker1-0-1-4wgcz   1/1     Running   0          7s    10.8.3.4   gke-jobset4-np1-004ddd96-cvzx            <none>           <none>
jobpertopology-worker2-0-0-849c6   1/1     Running   0          7s    10.8.1.9   gke-jobset4-default-pool-238f9ca7-8s4k   <none>           <none>
jobpertopology-worker2-0-1-mdlqv   1/1     Running   0          7s    10.8.2.6   gke-jobset4-default-pool-238f9ca7-hc67   <none>           <none>

Test JobSet with 1 replicated job with 3 replicas:

apiVersion: batch.x-k8s.io/v1alpha1
kind: JobSet
metadata:
  name: jobpertopology
  labels:
    app.kubernetes.io/name: jobset
    app.kubernetes.io/instance: jobset-example
    app.kubernetes.io/part-of: project
    app.kubernetes.io/managed-by: kustomize
    app.kubernetes.io/created-by: project
spec:
  jobs:
  - name: worker
    network: 
      enableDNSHostnames: true
    exclusive:
      topologyKey: cloud.google.com/gke-nodepool
    replicas: 3
    template:
      metadata:
        name: "worker"
      spec:
        completionMode: Indexed
        parallelism: 2
        completions: 2
        template:
          spec:
            restartPolicy: Never
            containers:
            - name: busybox-short-sleep
              image: busybox:latest
              command:
              - sleep
              - "1000"

Pod placement:

$ k get pods -o wide
NAME                              READY   STATUS    RESTARTS   AGE   IP          NODE                                     NOMINATED NODE   READINESS GATES
jobpertopology-worker-0-0-tdbfp   1/1     Running   0          11s   10.8.8.4    gke-jobset4-np2-782cc200-946b            <none>           <none>
jobpertopology-worker-0-1-rzrnb   1/1     Running   0          11s   10.8.7.4    gke-jobset4-np2-782cc200-vxq8            <none>           <none>
jobpertopology-worker-1-0-pkqgt   1/1     Running   0          11s   10.8.4.4    gke-jobset4-np1-004ddd96-c0b6            <none>           <none>
jobpertopology-worker-1-1-fdgtn   1/1     Running   0          11s   10.8.3.5    gke-jobset4-np1-004ddd96-cvzx            <none>           <none>
jobpertopology-worker-2-0-q7wbr   1/1     Running   0          11s   10.8.1.10   gke-jobset4-default-pool-238f9ca7-8s4k   <none>           <none>
jobpertopology-worker-2-1-kp4n7   1/1     Running   0          11s   10.8.2.7    gke-jobset4-default-pool-238f9ca7-hc67   <none>           <none>

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: danielvegamyhre

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [danielvegamyhre]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[ahg-g]

few nits

[ahg-g]

/label tide/merge-method-squash

[ahg-g]

/lgtm