1.2 Virtual Appliances (VA) Deployment & Clustering

Key Takeaways

  • For standard ISC source connectivity, deploy at least two VAs per cluster and maintain separate production and sandbox clusters.

  • Place a cluster close to its target systems. Use the source network requirements and the virtualization platform’s sizing guidance when selecting deployment locations.

  • The standard SailPoint VA image runs Flatcar. Pairing a VA with a tenant and cluster gives it an operational role beyond simply starting the virtual machine.

  • Separate hosts and sufficient remaining capacity reduce disruption when a VA becomes unavailable.

Last updated: November 2026

Choose the deployment before registering appliances

A virtual appliance cluster groups the VAs responsible for work on its assigned sources. Each source that uses a VA-based connector points to a specific cluster. A healthy cluster can have more than one running VA polling its queue. This provides continuity when one member is unavailable, provided other members remain healthy, have adequate capacity, and can reach the affected systems. A separate customer load balancer is not required for this design.

For standard ISC connectivity, SailPoint calls for at least two VAs in each cluster and at least one cluster for production and one for sandbox. Sandbox gives the team a place to check connectivity and configuration changes separately from production. Keep the sandbox design similar enough to exercise the same dependencies that matter in production, such as egress rules and the connector path. A sandbox with permissive internet access gives weak evidence about a production design that uses an HTTP proxy.

Placement questionDecision criterion
Which host platform can run the image?SailPoint documents local deployments on vSphere and Hyper-V, and cloud deployments on AWS, Azure, and Google Cloud Platform. Follow requirements for the chosen platform.
Where should the cluster sit?Prefer a location close to the sources with approved network routes. SailPoint advises against a demilitarized zone (DMZ) for a standard deployment.
How many machines are needed?Meet the per-cluster minimum and size for workload after a member is unavailable.
Can a single host take down several VAs?Spread VAs across separate hosts wherever you control placement, following SailPoint’s host distribution requirement.
Does the selected cluster run extra components?Check their resource additions against the platform-specific sizing requirements.

The standard image uses Flatcar. The customer provides the host infrastructure, network, and health monitoring. SailPoint ordinarily manages appliance software and updates. SailPoint documents manual updates and customer-provided operating systems as special options with separate requirements. Deployment planning for a standard image should use the normal appliance model and only assume an exception when the scenario specifies it.

Worked placement and failure example

A private payroll database and the directory it uses are on the same internal network. The engineering team proposes two VAs as separate guests on one vSphere host. That layout gives two guest instances but one underlying host failure would remove both. Move the guests onto separate hosts near those systems, verify that each can reach the required cloud endpoints and target ports, and assign the database source to the cluster intended for it. The engineering test should show that the surviving appliance can still communicate with the target when its peer is down.

Now consider fictional peak demand of 24 units of work and two VA hosts each measured at 15 units under this workload. Together they offer 30 units under an assumption of additive capacity in this example. Losing one leaves 15, so peak demand exceeds surviving capacity by 9 units. The two-appliance minimum protects against loss of the only member but does not prove that the cluster can meet the workload target after a failure. Reassess capacity through load tests or add capacity with independently reachable members. These figures are planning inputs for the example and do not represent product throughput limits.

Account for updates and interrupted work

In the standard automatic-update model, SailPoint performs rolling appliance updates. Members reboot one at a time. Continuity at cluster level does not guarantee that every operation already in progress completes. A long aggregation may be interrupted, so review its outcome and confirm whether it needs to be run again. A disconnected member can also prevent the cluster’s automatic maintenance from proceeding. Treat a disconnected standby as an actionable issue before counting it in a resilience plan.

A scenario may ask whether moving a functioning but remote cluster closer to its source is the first fix for a single permission error. Changing location does not change the permissions held by a target service account. Identify the evidence and the failure domain before redesigning infrastructure.

See SailPoint’s deployment and sizing requirements, supported deployment paths, and update behavior.

Test Your Knowledge

A team runs two VAs in one cluster, both on the same hypervisor host. Which design change best addresses loss of that host?

A

Place a customer HTTP load balancer in front of the two current guests

B

Assign the two guests different hostname aliases on the same host

C

Create a second cluster by dividing the guests on the same host

D

Place the two VAs on separate hosts and verify both host paths to the source

Test Your Knowledge

In a normally updating two-VA cluster, the first member starts rebooting during a long-running aggregation. What should the engineer check?

A

Only the source credentials, because a VA reboot cannot affect an in-progress task

B

Check the aggregation result separately from cluster availability, since the ongoing aggregation can be interrupted

C

Check whether both VAs were forced to reboot simultaneously as required by rolling updates

D

Check whether the tenant reassigned the source to a new cluster automatically

Sections you finish are checked off in the contents.