Multiple Zone Deployment

Version: 17.07

Supported Since: 17.01

Using the concept of zones, IPS provides you the ability to give priority to a certain set of nodes (collectively referred to as Primary Data Center or PDC zone) for your cluster deployments, rescheduling them on other nodes (collectively referred to as Secondary Data Center or SDC zone) if the aforementioned nodes are either unavailable or have run out of resources.

Unlike node-group based deployments, this does not constrain the deployment to a specific set of nodes; rather, it affects the instance scheduling process such that the PDC has priority over SDC. As an added advantage, this gives rise to a fail-over behaviour where instances will get scheduled on the PDC as long as its nodes are available, but get rescheduled on to SDC if the PDC nodes become unavailable (e.g. due to network issues or node failures).

Althought the other deployment types can be applied on a per-cluster basis, the multiple-zone deployment configurations are applied globally across all clusters in IPS, as it ultimately boils down to a platform-level configuration.
As the demo IPS installer is inherently a single-node set-up, it does not support multiple-zone deployments.

In order to create a multiple-zone deployment,

  1. Ensure that the orchestration platform underlying your IPS installation is configured for zone-based scheduling. For example, under Kubernetes, this can be achieved by providing a custom policy configuration file under the --policy-config-file parameter of the kube-scheduler:

    {
      "predicates": [
        {
           "name": "HostName"
        },
        {
           "name": "MatchNodeSelector"
        },
        {
           "name": "PodFitsResources"
        }
      ],
      "priorities": [
        {
           "name": "NewNodeLabelPriority",
           "weight": 1,
           "argument": {
              "labelPreference": {
                  "label": "pdc",
                  "presence": true
              }
           }
        }
      ]
    }
  2. Configure the PDC and SDC zones to utilize appropriate sets of nodes.

  3. Follow the uniform distribution deployment guide to obtain an active deployment that can scale across all worker nodes of your IPS installation.

  4. In order to see the fail-over mechanism in action, take the PDC nodes offline (one after the other, if you have multiple PDC nodes) while tracking the nodes where instances are getting deployed under the instances view.

    To compensate for the instances in the lost nodes, new instances will be spawned in the remaining PDC nodes, as long as they collectively have sufficient resources to host the new instances. Once the PDC runs out of resources (due to a sufficiently large number of nodes having been shut down), instances will start getting scheduled on the SDC, which effectively means a fail-over from PDC to SDC.

In this topic
In this topic
Contact Us