[PVFS-developers] Request for testing of fsck for pvfs

[Craig Tierney]

I have put together some code that resembles an filesystem check
(fsck) for PVFS.  I have used it to clean up my pvfs filesystem.  
One other person has tried it and not completely destroyed their
filesystem.

DISCLAIMER:  You are at your own risk using this code.  You should
always backup your files.

I would like to get others to try it out and see what it does
on their system.  It is currently configured to find problems,
and not correct them.  If people could try it out on their
system and send the results back to me I would appreciate it.

What it finds:

- Files with correct metadata, but incorrect data files on the
iods whether they are missing or not the right size.

- Finds data on iods with no corresponding metadata entry

- finds saveme files on the iods.

Currently the code is designed to only clean up problems, not
fix them.  Fixing is a bit harder, and I couldn't justify it
to myself.

- If metadata is missing, the file can be reconstructed if you
  can guess what the base iod was, but you don't have filename,
  directory, or ownership information.

- If datafiles are incorrect or missing, all that can be done is
  pad the files with zeros.  Might work for small files or text
  files to recover something, but you are getting pretty desparate   
  if you actually need this.

- Saveme files could be recreated if you assume that if the file
  that overwrote it is still there to assume the filename,directory,
  and ownership.

You can download the source code from aftp:divisions/fd/pvfs.  It
requires mpi for communication.

The current version is pvfs.fsck-0.60.tar.gz.

If you want to really get adventurous you can recompile the code
to fix problems, not just find them.  See the makefile.  It should
be obvious.

Let me know how it works.

Thanks,
Craig

-- 
Craig Tierney <ctierney@hpti.com>