I met the similar problems and finally I found that it is caused by 2 different process logging at the same file. 

I think it should be a bug of logback, which only use the active file size to check if need to rollover or not under the SizeBasedTriggeringPolicy. Once 2 different file descriptor point to the same file (inode), some interesting thing will happen when reach the max size: one fd point to the new file(inode) and the other continue write to the old file (inode). The one point to the old inode cannot be rolloed until the active file size is reached.