- If the metadata that needs to be stored is small and can reside on a single machine, then it’s replicated on each cluster server. Subsequently, the request can be served from any random server. In this approach, a load balancer can also be introduced between the front-end servers and metadata services.
Does the above para mean that the metadatastore can be replicated? It sounds like the data store is a separate component from metadata service. When the datastore is replicated to multiple nodes, service can reach out to any of these nodes. Here the load is between metadata service and the datastore servers. But what is the point of a load balancer between front end and metadata service? They make sense only when you have multiple instances of metadata services running. But we are talking about the multiple instances of data stores here