Conjur Enterprise - auto-failover election/promotion

Hey Team,

Having a dumb question about the Auto-Failover process. As the failover process relies on the raft consensus algorithm in electing a new leader out of the available standbys, how do we ensure the data is consistent ? As there is only one synchronous standby, how do we make sure that the synchronous standby is elected as the next leader during the election process or vice versa.

Thanks,
Senthil Kothan

Hi @senko,

Thanks for the question!

The auto-failover process does use the consensus protocol to determine which nodes are candidates for election and promotion to Leader. However, we have a second step where the nodes all share their current data replication state, and the node with the furthest replication will always be selected as the new Leader (if multiple nodes are tied for furthest replication, then those nodes race to hold a Leader “lock” and the first to obtain it is promoted).

The synchronous replication Standby will always be in the “furthest replication” cohort, and will be a promotion candidate. However, in a cluster with good network performance and infrequent updates, it’s often the case that asynchronous Standbys will also be fully replicated and will be candidates for promotion as well. So it isn’t guaranteed that the synchronous standby will be promoted, but the promoted Leader will have the most recent available data.

Thanks,
Micah

1 Like

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.