1.3 High Availability & Connectivity Models
Key Takeaways
Appliance high availability addresses member loss and available processing capacity. Disaster recovery also covers failure of a deployment location.
A single cluster can span sites so running members in either site can take work. Check target reachability and normal-operation latency from the recovery site.
A separate running recovery cluster requires source reassignment and reentry of source credentials during a switch.
Standby appliances deployed into the primary cluster during recovery trade less normal-operation exposure to remote latency for a slower start.
Design for the failure the business cares about
High availability (HA) covers the loss of an appliance or temporary interruption such as maintenance. Disaster recovery (DR) covers an event that takes a deployment location out of service. Two healthy VAs in one room can provide member redundancy. A loss of that whole room leaves the cluster with no available workers unless another location is part of the recovery design. For any pattern, verify that surviving VAs can reach both SailPoint and the relevant targets after the failure.
A recovery time objective is how quickly the business expects service to resume after an outage. Compare that target time with the measured duration of the required recovery steps, which may include reestablishing routes, retrieving source credentials, or pairing replacements. Estimate and measure the time for the representative set of sources. An appliance cluster marked healthy in the recovery site is evidence about the cloud-facing leg. A source test from that site provides separate evidence about the target-facing leg.
| SailPoint deployment pattern | Normal operation | After primary site loss | Main tradeoff |
|---|---|---|---|
| All appliances running in one cluster | Appliance members in either site can take requests for sources assigned to the cluster | Already running members in the recovery site can keep polling the queue | Slow paths from the recovery site can affect requests even before a disaster |
| Switch clusters | Source points to the primary cluster. Recovery cluster is running and receives updates | An operator reassigns affected sources to the recovery cluster and reenters their credentials | Avoids routine execution over the remote path but requires reconfiguration during an outage |
| Standby reactive deployment | Primary cluster runs. Tested reserve VAs at the recovery site have yet to join it | Deploy or register the reserves into the primary cluster so they can take its work | Turnaround depends on standby readiness and time to bring VAs online |
A separate recovery cluster has its own queue. Powering it on does not cause it to consume work from sources assigned to the primary cluster. For an all-running design, verify latency and access from each member site before placing them in the same cluster. For a switch-clusters design, arrange authorized access to the target credentials as part of the exercise. Access testing after reassignment should include a representative read and, where write capability is expected, a controlled write to a test account.
Work through a recovery decision
An enterprise measures 10 milliseconds between its primary VAs and a directory, and 85 milliseconds between its recovery VAs and that directory. The team measures unacceptable performance on the more distant path. Its highest-priority directory source can tolerate a measured manual switch, and its operations staff have access to source credentials during an outage.
The switch-clusters pattern fits those premises. With all members running in one cluster, any running member could pick up the work, so the slow route might be used even on a normal business day. A separate cluster lets the primary members handle normal work. The recovery plan must account for entering credentials and checking the moved connection. It also has to consider what happens if the directory itself is in the failed site. A reachable DR appliance cannot restore a directory service that has no reachable recovery endpoint.
Change the premise to a directory reachable quickly from both sites and a requirement to resume processing without an operator changing source settings. An all-running cluster across the sites better fits that case. The members at the alternate site are already working from the assigned queue. Confirm that they also have adequate capacity and that their route to the directory survives the same event that disabled the primary site.
Keep egress and redundancy decisions distinct
A standard VA can connect directly to the required endpoints through the firewall. SailPoint also documents an HTTP proxy configuration and a network tunnel option for some restricted networks. An egress option solves a network policy problem. The number of independent sites, the assigned cluster, and continued access to target systems determine the failover behavior. If all VAs rely on one proxy service in the primary site, include loss of that dependency in the recovery exercise even when appliance compute spans two sites.
See SailPoint’s HA and DR patterns and VA configuration options.
An alternate VA cluster is active at the recovery site. A source is still configured to use its failed primary cluster. What makes the source use the alternate cluster?
Reassign the source to the alternate cluster and provide the source credentials again
Wait for the source to discover a higher-priority cluster automatically
Restart the browser so the recovery tenant sees both clusters
Run two connection tests, which prompts the alternate cluster to claim the primary queue
A source remains reachable with acceptable latency from both sites. The business requires appliances at the recovery site to continue polling without source reconfiguration. Which pattern meets the requirement?
Separate clusters with a manual source switch
Undeployed standby appliances at the recovery site
Running VAs across both sites in the source’s currently assigned cluster
Two VAs at the primary site using different HTTP proxies
Sections you finish are checked off in the contents.