Module openraft::docs::obsolete::fast_commit

Fast Commit(obsolete)

This algorithm is no longer used, because it would be more flexible treating local-IO the same as replication to a remote node. As a result, applying a log does not need to wait for it to be flushed on local disk.

Fast Commit Algorithm

The algorithm assumes that an entry will be committed as soon as it is appended if the cluster has only one voter.

However, if there is a membership log in the middle of the input entries, the condition to commit will change, and entries before and after the membership entry must be dealt with differently.

When a membership entry is seen, update progress for all former entries. Then upgrade the quorum set for the Progress.

E.g., if the input entries are 2..6, entry 4 changes membership from a to abc. Then it will output a LeaderCommit command to commit entries 2,3.

1 2 3 4 5 6
a x x a y y
      b
      c

If the input entries are 2..6, entry 4 changes membership from abc to a. Then it will output a LeaderCommit command to commit entries 2,3,4,5,6.

1 2 3 4 5 6
a x x a y y
b
c