About the Retry and Back-Off Mechanism

NuoDB automatically and continually tries to restart a database process that has failed. For each database, there is a retry and back-off policy that specifies

These database options specify the database's retry and back-off policy:

See Setting Database Options Related to Retry and Back Off.

NuoDB tries to restart a database process that terminated and backs off trying to restart a database process according to the database's template and also its observation of required uptime for that process. You may update the template by supplying values for any of these database options. Uptime starts when a process starts and ends when a process terminates. NuoDB compares process uptime to the required minimum uptime for the database. If the required uptime is not met, then enforcement is delayed and that delay grows with each failed retry (delay time increases two times after each failed retry).

The uptime requirement is violated by a NodeFailed event or a NodeLeft event. A NodeFailed event occurs when a database process fails to start, for example because of an invalid database option. A NodeLeft event occurs if a process that was running terminates or if the broker on the process's host fails.

You can request to stop enforcement when the maximum retry count has been reached. A maximum retry of 1 means that as soon as any process leaves the domain with insufficient uptime, stop enforcement. Stopping enforcemet does not stop any process. If enforcement is stopped, a call to start the database does not restart it.

The statistics the broker tracks for retry and back-off are durable. If a broker is restarted during the back-off retry period, the broker tries to restart a process according to the durable retry history. NuoDB maintains a separate retry/fail count for each process that terminated and that it is trying to restart.