[KUDU-2246] Make "kudu tablet change_config move_replica" more reliable

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: consensus, ops-tooling
• Assignee: Unassigned
• Reporter: Mike Percy
• Created: 20/Dec/17 00:59
• Updated: 20/Dec/17 01:35

Description

As part of KUDU-1097 we changed the replica move tool to use the new 3-4-3 re-replication scheme when it's enabled. In certain edge-case scenarios, such as when there is a replica failure in the middle of a move operation, the reliability of the tool may be reduced, and the node slated for removal may not be removed. This is especially true if the user hits ctrl+c and kills the tool before the movement process is complete in the context of a simultaneous failure.

We can improve this by implementing leadership transfer (KUDU-2245) and having a leader gracefully step down when there is a suitable replacement and the leader is marked with the "replace=true" attribute in the Raft config.

Issue Links

• is blocked by: KUDU-2245 Leadership transfer (graceful step down)
• relates to: KUDU-2247 Update doc on the 'kudu tablet change_config move_replica' after 3-4-3 enabled by default