fortuna Can you please set the system property com.cloudbees.hudson.plugins.folder.computed.BACKUP_LOG_COUNT to something high enough to catch the failure, e.g.
24h / 15min = 96
Note, if you do not want to restart your master you can set this value by the Groocy console with the script:
(to clear it again just remove the system property or set the value to null via the groovy console)
Then wait for a failure to occur, and find the indexing/indexing.log.N which has the failure (will likely be the one with Will remove in the log text, e.g.
grep "Will remove" indexing.log*
but will have a Finished: SUCCESS. OTOH if you have an orphaned item strategy that is retaining them for a period of time it may be harder to find the one with the suppressed (but hopefully logged) stack trace.