CAP: HA and Consistency? #7336
|
In HA mode, Redis can lose writes when the primary failovers. It can lose writes on restarts even with persistence enabled. How about DragonflyDB? Under what circumstances can it drop acknowledged writes? Is it a consistent datastore? |
Replies: 1 comment
|
Dragonfly is an in-memory data store, like Redis, and as such it is not consistent. There is a very short period of time between when a master acknowledges a command, and until its replica(s) are updated with it. As such, if the master's machine / process terminate unexpectedly, it is possible that some mutations would not be seen on the replica, despite them having been acknowledged by the master. |
Dragonfly is an in-memory data store, like Redis, and as such it is not consistent. There is a very short period of time between when a master acknowledges a command, and until its replica(s) are updated with it. As such, if the master's machine / process terminate unexpectedly, it is possible that some mutations would not be seen on the replica, despite them having been acknowledged by the master.
Having said that, if the master remains alive, and the failover is done by promoting the replica, no commands should be lost.