The MVP for QTL dynamic configuration #1576 does not have a automagic way to handle weird state. Like if a node is down while a DELETE request is being propagated, then that node will never get the delete request.
This ask is that there is some better handling put in place for strange states.
Potential solutions include:
- Dictator mode - The coordinator implementation FORCES its view of what lookups should be on a node by deleting lookups on nodes that are not expected to be there
- Changelog - The configuration propagation is modified to use a changelog with history instead of only maintaining a changelog of 1 state.
The MVP for QTL dynamic configuration #1576 does not have a automagic way to handle weird state. Like if a node is down while a DELETE request is being propagated, then that node will never get the delete request.
This ask is that there is some better handling put in place for strange states.
Potential solutions include: