Say you have configured DailyRollingFileAppender with daily rotation (It can be configured for rotation every hour, minute etc). Say, it is 31-Dec-2014 today and log file name is sample.log. Log rotation will happen in the following way:
- First log message that is received after midnight (say at 1am on 1-Jan-2015) will trigger log file rotation.
- Log file rotation will first delete any existing file with previous day suffix. (i.e. It will delete any file with name sample-2014-12-31.log. Ideally no such file should exist.).
- It will then rename current file with suffix of previous day. i.e. it renames sample.log to sample-2014-12-31.log.
- It will create new log file without suffix. i.e. new sample.log
- It will start writing into the new file sample.log.
If two instances of Log Manager points to same log file then each instance will independently repeats above steps on the same file. This can happen in any of the following scenarios:
- If two or more WAR file deployed in same container points to same log file.
- If two or more processes points to same log file.
- etc
Such scenario leads to the issue mentioned in the question.
- On a windows machine, once first process has rotated the log file and acquired handle on the new file, second log appender will fail to write logs.
- On a linux machine, Second process will delete the archive file created by first process and rename the new file (currently being used by first process) to previous day file. Thus first process will start writing logs in previous day file, second process will write logs in new file. After midnight, log file used by first process will get deleted. So, logs from first process will keep getting lost.