Comment 4 for bug 766607

The problem is that when a new table is added after taking a full backup *and* log files have been flushed before an incremental one, the new table data goes to a .delta file. When trying to apply that .delta to a full backup later, xtrabackup_apply_delta() fails because the corresponding data file cannot be found.

The difference with the situation when log files have not been flushed before taking an incremental backup is that in this case a file creation event is in the log file, so .ibd is then created when doing crash recovery and replaying the log. The corresponding .delta file in this case will be empty.

I think the fix is to explicitly create a tablespace before applying the .delta file when the corresponding .ibd is missing.