Deploying CVP OVA on ESX

Deploying the CVP OVA file should be the first step in any setup. After the CVP OVA file is deployed, you can chose between the two configuration methods for CloudVision Portal (CVP).

Note: Arista does not support VMware Snapshots on CloudVision virtual machines. For more information, refer to VMware vMotion and Snapshot Support.

Pre-requisites:

Use of the Deploy OVF Template requires the VMware Client Integration plugin, which is not supported by the Chrome browser after versions 42.

  1. The OVA file can be deployed as a VM in a VMware environment by using the drop menu under the Actions heading and selecting Deploy the OVA template.
    Note: For multi-node setups, the following steps must be completed once for each VM, three times to launch three VMs.
    Figure 1. Deploy the OVA template
  2. Having selected the Deploy OVF Template option, VCenter will prompt for the location of the OVA file; this can be either on a local hard disk, network share, or Internet URL. The location of the OVA file should be entered or selected.
    Figure 2. Location of the OVA file
  3. Click Next to go to the next task.
  4. Review the OVA template details.
    Figure 3. Review OVA template details
  5. Click Next to go to the next task.
  6. Type the name for the OVA file in the Name field and select the folder for the OVA file.
    Figure 4. Select name and folder location for OVA file
  7. Click Next to go to the next task.
  8. Select the resource where you want the deployed template (OVA file) to be run.
    Figure 5. Select the resource
  9. Click Next to go to the next task.
  10. Select the location where you want the files for the deployed template to be stored.
    Figure 6. Select the destination storage
    Note:

    It is recommended to select Thick provision lazy zeroed under the Select virtual disk format dropdown menu.

  11. Click Next to go to the next task.
  12. Setup the networks that the deployed template should use.
    Figure 7. Setup the networks
  13. Click Next.

    VCenter loads the OVA and displays the configuration settings.

    Figure 8. Select the Finish button to accept these settings
  14. Review the configuration settings, and click Finish to accept and save the configuration.

    VCenter begins to deploy the virtual appliance. Once the appliance is deployed, you can configure the CVP application using either Shell-based Configuration or ISO-based Configuration.

VMware vMotion and Snapshot Support

CloudVision includes the following infrastructure components that are used as the basis for the application services and database. This is not an exhaustive list, but the key components as it relates to this topic.
  • Hadoop - open source framework from Apache that is used to store and process large datasets distributed across a cluster of servers
  • Hbase - open source database from Apache that runs on Hadoop cluster
  • Zookeeper - centralized service for maintaining configuration information, naming, providing distributed synchronization, and providing group services

The CloudVision database (hbase/hadoop) is deployed across the three nodes within the CloudVision cluster. The integrity of this database is critical to the correct functioning of CloudVision, and thus, there are specific requirements on the hypervisor and storage for these virtual machines forming these nodes.

VMware Snapshots

Within the CloudVision infrastructure, data is constantly being written to Apache hadoop by all nodes. Disk snapshots used by VMware have no hooks into the hbase quiesce states, meaning a snapshot of a disk state would almost always be inconsistent and lead to database corruption during a restore process. This results in a snapshot having no meaningful use as a restore point due to the nature of the database, which is typical for database application performance using VMware Snapshots (VMware reference).

VMware Snapshots are very I/O intensive, leaving almost no I/O for the virtual machines during the snapshot process. Impact on resources, such as disk, can lead to hbase and zookeeper failures. These symptoms are evident in multiple cases where the support team has identified snapshots that were in progress before failures.

VMware does not recommend using VM Snapshots as backups (https://kb.vmware.com/s/article/1025279), therefore other backup mechanisms are recommended by Arista as outlined below.

Note: For these reasons, Arista does not support VMware Snapshots on CloudVision virtual machines.

VMware vMotion

CloudVision supports VMware vMotion under specific configuration and operational criteria as follows:
  • The virtual machine disks are shared between the source and target ESXi host
  • Latency between ESXi hosts is less than 5ms
  • Only one CloudVision node may be vMotioned at a time
Note: CVP 2020.1.0 and future releases support host-to-host vmotion where the storage is shared between ESXI hosts. Only one host can be in vMotion at a given time.

Backup Solutions for CloudVision

Daily backups of the CloudVision provisioning data are automatically scheduled to be taken at 2AM UTC. This backup file is stored locally on the CloudVision cluster. Common practice by customers is to schedule a copy of this backup file from the CloudVision cluster to some external data store.

There is an example script to help automate the copying of the backup file available on the Arista Github site (link).

CloudVision telemetry data received from switches is replicated between the CloudVision clusters. In the event a single node becomes unavailable and a new node is added to the cluster, this telemetry data is replicated to the new node.

Arista EOS with the Streaming Telemetry agent (TerminAttr v1.7.1 and later) supports establishing connections to multiple CloudVision clusters. This enables the user to send the telemetry data to a backup CloudVision instance, to maintain an up-to-date redundant store.

There is a detailed explanation of this deployment model available on the Arista EOS Central site (link), which would assist with the design and deployment of this HA solution.