WiredTiger engine is not using prepared log files for log file allocation when hot backup is in progress. The relevant logic is implemented in function __log_newfile.
Due to this logic prep_missed counter is increased which causes log file server thread (__log_file_server) to bump counter log_prealloc. log_prealloc controls number of pre-created log files (WiredTigerPreplog.*).
So with every new hot backup log_prealloc counter grows more and more thus creating more and more WiredTigerPreplog.* files in journal directory. For example, after a month of daily hot backups we have about 370 prep log files (each of 100MB => 37GB of junk) which I even cannot delete premanently as the log file thread re-creates them to the count of log_prealloc property.
My proposed solution is not to increment prep_missed counter when hot backup is active since we deliberately not using those prep files.
- related to
-
WT-3756 Adjust pre-allocated file amount downward if we're not using them quickly enough
- Closed