Failing over – almost as good
Architecture is highly available if it survives a component failure automatically. Either another running component takes over the workload from the faulty part, or a new member is brought to the active state to handle the load. The former is called the active-active scheme, while the latter is named active-standby.
When one of the instances inside a cluster experiences downtime, the LB will stop sending new requests to it. In an active-active mode, the LB will select the next working instance. However, for the active-standby mode, a defect instance triggers a failover process that moves the standby instance to the active state while it tries to recover the in-fault instance.
The failover operation is an intense computing process because it needs to spin off a new instance with the same resources and datasets. Since it takes some time to complete, users will experience an impact during this transition period. For that reason, it’s preferable...
