Feature: Raft::trigger()::allow_next_revert() allow to reset replication for next detected follower log revert
This method requests the RaftCore to allow to reset replication for a
specific node when log revert is detected.
allow=true: This method instructs the RaftCore to allow the target
node's log to revert to a previous state for one time.
allow=false: This method instructs the RaftCore to panic if the
target node's log revert
This method returns Fatal error if failed to send the request to
RaftCore, e.g. when RaftCore is shut down.
Otherwise, it returns a Ok(Result<_,_>), the inner result is:
Ok(()) if the request is successfully processed,
or Err(AllowNextRevertError) explaining why the request is rejected.
Behavior
If this node is the Leader, it will attempt to replicate logs to the
target node from the beginning.
If this node is not the Leader, the request is ignored.
If the target node is not found, the request is ignored.
Automatic Replication Reset
When the loosen-follower-log-revert feature flag is enabled, the
Leader automatically reset replication if it detects that the target
node's log has reverted. This feature is primarily useful in testing
environments.
Production Considerations
In production environments, state reversion is a critical issue that
should not be automatically handled. However, there may be scenarios
where a Follower's data is intentionally removed and needs to rejoin the
cluster(without membership changes). In such cases, the Leader should
reinitialize replication for that node with the following steps:
Changelog
Feature:
Raft::trigger()::allow_next_revert()allow to reset replication for next detected follower log revertThis method requests the RaftCore to allow to reset replication for a specific node when log revert is detected.
allow=true: This method instructs the RaftCore to allow the target node's log to revert to a previous state for one time.allow=false: This method instructs the RaftCore to panic if the target node's log revertThis method returns
Fatalerror if failed to send the request to RaftCore, e.g. when RaftCore is shut down. Otherwise, it returns aOk(Result<_,_>), the inner result is:Ok(())if the request is successfully processed,Err(AllowNextRevertError)explaining why the request is rejected.Behavior
Automatic Replication Reset
When the
loosen-follower-log-revertfeature flag is enabled, the Leader automatically reset replication if it detects that the target node's log has reverted. This feature is primarily useful in testing environments.Production Considerations
In production environments, state reversion is a critical issue that should not be automatically handled. However, there may be scenarios where a Follower's data is intentionally removed and needs to rejoin the cluster(without membership changes). In such cases, the Leader should reinitialize replication for that node with the following steps:
Shut down the target node.
call [
Self::allow_next_revert] on the Leader.Clear the target node's data directory.
Restart the target node.
Fix: #1251
This change is