Multiple approaches are possible to ensure the appropriate level of availability for a customer’s use case.
HA within a deployment
All AWS Outposts Rack deployments offer local HA through the use of redundant networking and power equipment. This redundant Outposts equipment can be paired with dual customer-provided power sources and redundant network connectivity for even greater resilience. AWS recommends combining the built-in network redundancy with customers’ network redundancy to further ensure constant connectivity for the Outpost.
Additional built-in and always active capacity can be provisioned within a deployment, offering further improvements to availability. AWS offers production-ready deployment configurations that support an N+1 capacity level for EC2 instances.
AWS Outposts Rack deployments provide two instance placement options for minimizing the impact of hardware failures:
Dedicated hosts – This offers customers the ability to use auto-placement configurations to dictate that instances be deployed onto a specific host or any available host matching certain requirements. This is done using host affinity.
Placement Groups: These work similarly to the same feature available within a region, with some modifications.
Cluster placement groups ensure instances wind up on the same physical server. This is normally done to ensure the lowest possible latency between two instances.
Customers with more than one rack in their deployment can use a spread placement group at the rack level to ensure that the instances in a group are spread across multiple racks.
The spread placement strategy of partition is used to distribute multiple instances across multiple racks within partitions. Single partitions can house multiple instances and customers can target specific partitions for instance placement or they can use automatic distribution to spread the instances across the partitions. The following figure shows an example of this approach:
Figure 5.18 – A spread placement group using targeted partitions
Additionally, an AWS Outposts Rack deployment offers a unique spread placement strategy at the host level. This is meant for single-rack deployments so that customers can safely spread deployments of instances across physical servers within a single rack.
Customers should utilize Amazon CloudWatch to monitor capacity availability metrics and set alarms for application health, create CloudWatch actions to perform automated recovery, and monitor the utilization of your Outpost deployment. When instances are marked as unhealthy, they need to be migrated to a healthy host. These CloudWatch capabilities allow customers to enable automated migrations of unhealthy hosts based on instance status checks. Auto-scaling groups can also be enabled to achieve a similar effect.