Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Abandoned
-
None
-
None
-
None
Description
Tablet servers that lose their locks don't always kill themselves. In one case, the lock is lost, detected, but the class that is supposed to execute fails because it cannot be loaded from the file system. Other threads continue to run, unaware that they are no longer part of a valid server.
Consider firewalling, somehow, this rogue server from the rest of the cluster. Perhaps it can be an external utility that will power down the node, or actually set up firewall rules.
Attachments
Issue Links
- relates to
-
ACCUMULO-1165 Use a unique SystemToken (credentials) for each server instance (TServer, etc.)
- Open