Prerequisites
Scope
Applies to every DynamoAI product. Cloud-agnostic; see Cloud Reference for the commands that verify each item on your platform.
Prerequisites
Each of the following is required. Missing any one reduces the deployment to node-level tolerance at best.
| Requirement | Why |
|---|---|
| Node pools spanning three availability zones | The availability zone is the failure domain the deployment is measured against. Replicas that all sit in one zone tolerate a node failure and nothing larger, however many of them there are. A third zone keeps a datastore quorum alive after one zone is lost |
| Spare capacity for one zone's workload | Losing a zone reschedules its pods onto the surviving zones, which then carry half as much load again. Size the pools for that state, since capacity that exists only in the failed zone cannot be scheduled onto |
| Customer-managed datastores with HA enabled | The bundled PostgreSQL, MongoDB and Redis run as single in-cluster instances with no standby and no failover, and are intended for evaluation. Platform availability cannot exceed the availability of the stores it is pointed at |
| A storage class appropriate to each workload | Locally redundant block storage exists in one zone and pins any pod that mounts it to that zone, so losing the zone takes that workload down whatever its replica count. These volumes are also single-attach, so moving one means a detach and reattach cycle during which the workload serves nothing. Zone-redundant classes remove both constraints |