+1 to Jan's "clustered" vs "non clustered". If we clean up terminology, I suggest we also clarify the meaning and use of Slice vs Shard vs Leader vs Replica vs Core. Here's my understanding:
I consider Slice == Shard (and would happily drop Slice): a logical concept of a specific subset of a collection. A Shard then has one or multiple copies of the data called Replicas (if a shard has no copy of the data there's an issue). The Leader is one such Replica. A shard with a replication factor of 1 has a single Replica that happens to be the Leader. "Replica" does therefore not imply "replication". A Core is an in memory instantiation of a disk index representing a Replica. I believe that often the on disk index is referred to as "Core" as well (I'm not bothered by this, there's no associated confusion IMO). Overseer is a central place where a fair bit of the cluster management logic is implemented today (Collection API, Autoscaling, Cluster state change). It is therefore a cluster manager. Note that a different implementation of "Clustered Solr" (a.k.a. SolrCloud) can most likely be done without the need of a central process in addition to the already centralized storage backend (currently ZooKeeper). In other words, Overseer is not IMO the defining characteristic of SolrCloud, it is one implementation choice, and there are others. To keep in mind for clarity and to guide renaming. On Fri, Jun 19, 2020 at 3:23 PM j.s. <jsay...@brandeis.edu> wrote: > hi > > solr is very helpful. > > On 6/18/20 9:50 PM, Rahul Goswami wrote: > > So +1 on "slave" being the problematic term IMO, not "master". > > but you cannot have a master without a slave, n'est-ce pas? > > i think it is better to use the metaphor of copying rather than one of > hierarchy. language has so many (unintended) consequences ... > > good luck! >
