A topology master server receives any configuration change from other servers in the topology, verifies the change, then makes the change available to all connected servers when they poll the master. The master always sends a digest of its subtree contents on each update. If the node has a different digest than the master, it knows it's not synchronized. The servers will pull the entire subtree from the master if they detect that they are not synchronized. A server may detect it is not synchronized with the master under the following conditions:
- At the end of its periodic polling interval, if a server's subtree digest differs from that of its master, then it knows it's not synchronized.
- If one or more servers have been added to or removed from the topology, the servers will not be synchronized.
The master of the topology is selected by prioritizing servers by minimum supported product version, most available, newest server version, earliest start time, and startup UUID (a smaller UUID is preferred).
After determining a master, the topology data is reviewed from all available servers (every five seconds by default) to determine if any new information makes a server better suited to being the master. If a new server can be the master, it will communicate that to the other servers, if no other server has advertised that it should be the master. This ensures that all servers accept the same master at approximately the same time (within a few milliseconds of each other). If there is no better master, the initial master maintains the role.
After the best master has been selected for the given interval, the following conditions are confirmed:
- A majority of servers is reachable from that master. (The master server itself is considered while determining this majority.)
- There is only a single master in the entire topology.
If either of these conditions is not met, the topology is without a master and the peer
polling frequency is reduced to 100 milliseconds to find a new master as quickly as
possible. If there is no master in the topology for more than one minute, a
mirrored-subtree-manager-no-master-found alarm is raised. If one of
the servers in the topology is forced as master with the
force-as-master-for-mirrored-data option in the Global
Configuration configuration object, a
mirrored-subtree-manager-forced-as-master-warning warning alarm is
raised. If multiple servers have been forced as masters, then a
mirrored-subtree-manager-forced-as-master-error critical alarm will