There is a known problem in versions earlier than 25.4.10 where device server restarts get triggered and cause jobs to pile up and not process.
This will be resolved with the 25.4.10 release of NCM