Normal Operation of Asynchronous Commit

The database functions entirely normally within the active data center(s). As transactions make changes, these changes are replicated to observers in the passive data center(s). When transactions commit, these commits are replicated to observers.

Replication messages flow over the network from engines in the active data center to observers, spending time in flight that is determined by long-distance network latency. Transactions do not wait for observers to respond, nor for them to make these changes durable on disk.

Asynchronous Storage Managers are passive and do not participate in NuoDB’s performance-sensitive network protocols, so high network latency between remote data centers does not slow down the workload. Performance-sensitive protocols execute entirely inside the active data center, where network latency is low.

As long as there is sufficient network bandwidth, the presence of observers does not slow down or interfere with execution of the database workload. If the network does not have sufficient bandwidth to keep up, active TEs will buffer replication messages, forcing them to throttle their performance or run out of memory. Before that happens, an asynchronous SM at the other end of a low-bandwidth network link might be evicted by ping timeout. In that case the active data center continues running but the passive data center stops providing for recovery from disasters. Administrators should monitor their databases to remain confident that everything is working as intended.