Sun 26 Dec 2004 01:50:14 PM UTC, original submission:
I'm currently using rdiff-backup 0.13.4 (on OS X). As long as I know of rdiff-backup, I was unable to do backups to nfs volumes as the target. Currently processing power on my "file server" is the bottleneck so running rdiff-backup remotely is no solution for me.
rdiff-backup crashes trying to delete a directory in the target directory if the directory being deleted contains at least one file.
This seems to be due to the fact that at this time some file handles are still open. (This is a problem of rdiff-backup, not of nfs or some other network file system)
While on a local filesystem it is possible to delete a directory with a locked file inside, nfs renames locked but deleted files so something like .nfs?????. Thus it is impossible to delete the directory.
As far as I was able to debug, before deleting a file, it is opened while creating an rdiff signature (somewhere in librsync.py, in SigFile, as far as I remember). This file handle seems to be never released causing the error described above.
My complete knowledge of python comes from reading rdiff-backup source code, so I am not completely sure when it comes to the patch I supply here.
All I can say is, that for me backing up over nfs now works fine.
Feel free to contact me.
Regards,
Maximilian Mehnert
maximilian.mehnert at charite.de
|