Details
-
Bug
-
Resolution: Fixed
-
Major
-
None
Description
When a cluster is under heavy load, with concurrent changes happening on all nodes, it is possible that OrientDB will "fix up" some distributed transactions as part of the last phase of the commit. This work is usually done on a separate DB connection, but with the same thread that initiated the local commit.
To correctly track these fix updates we need to detect when changes happen on the same thread during the commit phase, even when a different DB connection is used, and fold them into any existing record changes so they can be dispatched via the post-commit hook.
Attachments
Issue Links
- relates
-
NEXUS-18185 Database out of sync in HA cluster - mitigation and recovery
-
- Done
-