Removing a Rollback Record

After all grid processes have been restarted and verified to be operational, it is recommended but not required to delete the rollback record from the grid configuration to prevent complications during future maintenance operations.

Leaving the rollback record in the grid indicates that a node completes recovery to that checkpoint upon starting. This happens once. However if a secondary node with a blank data directory is started without access to any checkpoints, an error such as the following may be observed:

[timestamp] seve node: Error during start up: File I/O error
[timestamp] seve node: Exception details:
TIBCO Exception:
Error Code = File I/O error
Description = rollback checkpoint ID <checkpoint_ID> not found on this node
Thread Name = tibdgnode
Stack Trace:
  	_performNodeRecoveryCheckpointRestore, 1834
	_performNodeRecovery, 1901
	_tibdgNode_Open, 3798
	main, 694
Before you begin Ensure that the restore process is complete. That is, all grid keepers, nodes, and proxies are restarted following the restore operation.
    Procedure
  1. Run the 'tibdg rollback delete' command. 
    tibdg -r <realm service URL> rollback delete

  2. Ensure that future invocations of state keepers do not supply the -R / --recovery-file command line argument. Any scripts that are used to automatically start keepers must be updated to remove the -R / --recovery-file argument if present.

ResultThe rollback delete command must indicate that the rollback record is deleted.
What to do nextWhenever you start the state keeper again, ensure that you start it without the -R argument.