Orient has a background task that detects minor cluster inconsistencies and attempts to bring them back in sync. Unfortunately it cannot automatically resolve all sync issues, sometimes manual intervention is required such as triggering a full sync across the cluster. These situations can lead to requests failing because the underlying transaction has failed due to excessive retries.
This story will look at adding a health check to flag when transactions fail due to excessive retries, so the admin can decide how to address it. They can then run the new "check database" endpoint to see if the indexes have missing or duplicate entries, or if there's been any corruption. They can also decide to re-install the database from other nodes in the cluster.