A brief introduction to the following redis-sentinel:
Redis-sentinel is the management tool for Redis clusters, which is the monitoring management, notification and instance failover service for Redis instances. In the general distributed central Node database, the function of Redis-sentinel is the work of the central node, monitoring the work of each other node and failing recovery to improve the high availability of the cluster. (Introduction from Link: http://www.searchdatabase.com.cn/showcontent_71572.htm)
Here is my test summary for the Redis 2.9.11-non-stable version, which may not have the following issue.
1. Using Redis-sentinel,redis instances must be run in non-clustered mode
2. If the default master-slave relationship is hung, then start failover, in the failover process, there are n slave is still alive, then the new master has n slave, the old master restart can also join a cluster, However, other slave reboots that have been suspended before or during failover are not allowed to join the cluster unless their configuration files are modified to point master to the new master.
3. If the default master-slave relationship changes, one of the slave hangs a reboot, its master still points to the configuration file specified, but it is also the slave of the new master, because the old master has become the slave of the new master. So the old master's slave is also the new master's slave, the new master's slave number is unchanged, this can be seen from the command: Sentinel Slaves MyMaster.
PS: As can be seen from here, slave reboot does not redirect master, the master-slave relationship is the Sentinel Send info request obtained
4. If the default master-slave relationship is hung and the failover ends normally, the old master reboot can join the cluster, but the old master will not be able to join the cluster for the second reboot.
5. Whenever a slave of master command is invoked, that is, the master-slave relationship is established, it triggers the primary and the persisted data from the Save method, whether or not you prohibit save.
6. In the cluster, if the master of the default master-slave relationship hangs and restarts immediately, if the master does not persist, the data is completely lost and the data is emptied.
7. In the cluster, if the master of the non-default principal-slave relationship hangs and restarts immediately, the master-slave relationship becomes chaotic and the cluster is unavailable.
So much for the first.
