Page History

...

A proposed architecture is as follows. We have one ZooKeeper ensemble called main, which is dedicated to keeping metadata of other ensembles and of containers. All other servers map to one or more ensembles. The ensembles are the ones keeping containers and executing operations on them. Main has three branches: a.

1. "/ensembles": Contains a set of children mapping ZooKeeper servers to ensembles;

...

1. "/containers": Contains a set of children that maps containers to ensembles;

...

1. "/servers": A map from server to set of ensembles.

The clients only need to know the servers of the main ensemble to connect to everyone else. For example, "/subtree/apps/yaview" has the name of the ensemble that stores "/apps/yaview" and "/ensemble/A" has the list of servers that make up ensemble A. Assuming data stored in the main ensemble is relatively static and small, such an ensemble does not have to process a pretty heavy workload.

...