rsync - Mar 2002 - Wierd timestamp problem

Greetings,

         We found out today that rsync has been missing some files.  Here 
is the senerio:

         We have two nfs servers in different datacenters, one being the 
backup to the other.  We want to keep them up to date so we use rsync 
inside crontabs.  We have Solaris, AIX, True64 and VMS systems reading and 
writing to the nfs drives.  I am running rsync on Solaris 8 servers.  What 
seems to happen is when a file gets updated in a certain way on a VMS 
system, it does not change the modification time of the file on the unix 
servers.  This seems to be a bug in multinet and vms file 
versioning.  Anyway, we need a work around quickly and we don't think
we'll
see one out of Compaq or Multinet any time soon.

         I think I can get away with specifing the "--checksum" flag,
but
I'm just not sure.

         Also, right now we are doing this copy a very bad way and plan on 
changing it to a daemon solution shortly.  We mount both data centers nfs 
on one server and rsync the dirs.  Here is the code segement that we are using:
>$WORKPATH/bin/rsync.7   --recursive             \
>                         --verbose               \
>                         --stats                 \
>                         --delete                \
>                         --one-file-system       \
>                         --times                 \
>                         --group                 \
>                         --owner                 \
>                         --perms                 \
>                         --archive               \
>                         --checksum              \
>                         --whole-file            \
>                         --bwlimit=10240         \
>                         --timeout=600           \
>                         --exclude-from=$EXCLUDE \
>                         $FROMDIR $TODIR >> $OUTLOG 2>&1
         Any thoughts and/or suggestions would be most apreciated.

         thanks,

         Paul

On 26 Mar 2002, Paul LaMadeleine <plamadeleine@lightbridge.com> wrote:
>         I think I can get away with specifing the "--checksum"
flag, but
> I'm just not sure.
Yes, I think that is the best solution.  It will of course be pretty
slow since rsync will have to read every file on the filesystem every
time, but without accurate mtimes I don't see at the moment what else
it could do.

If the problem only occurs in particular directories accessed by the
VMS machines then perhaps you can copy them using --checksum and leave
it off for the rest of the filesystem.
       
Perhaps you can hack something up involving your NFS server writing a
log of modified files, and then use that as a cue to rsync.

I think you will find using an rsyncd rather than copying between two
NFS filesystems should be much faster.
>$WORKPATH/bin/rsync.7   --recursive             \
>                        --verbose               \
>                        --stats                 \
>                        --delete                \
>                        --one-file-system       \
>                        --times                 \
>                        --group                 \
>                        --owner                 \
>                        --perms                 \
>                        --archive               \
>                        --checksum              \
>                        --whole-file            \
>                        --bwlimit=10240         \
>                        --timeout=600           \
>                        --exclude-from=$EXCLUDE \
>                        $FROMDIR $TODIR >> $OUTLOG 2>&1
You can leave off --recursive --group --owner --perms because they're
implied by --archive, and --whole-file because it's implied by both
directories being local (including NFS).

-- 
Martin