These are chat archives for atomix/atomix

Jul 2017
Jordan Halterman
Jul 08 2017 00:49
Well, the problem with allowing state machines to stop the application of operations is the completion of command requests to the leader depends on operations being able to be applied to the state machine. If an operation can’t be successfully applied and needs to be later replied, that blocks all future operations from being completed until that operation succeeds. Retrying failed operations should be handled at a higher level that ensures we don’t block operations from being applied to the state machine
@kammd 2/3 nodes
Essentially, a node’s bootstrap or join call is completed once it finds a leader. If only 1/3 nodes is bootstrapped then no leader will ever be elected. If 2/3 are bootstrapped then they’ll elect a leader and both bootstrap calls will complete