Hi Jakob,

Jakob Praher wrote:
> hi all,
> 
> first of all: I am a very happy user of xfs for a long time now.
> I use xfs in tandem with lvm2 and I have to say I am very pleased.
> 
> Now a question:
> 
> There are two modes of using xfs(dump,restore): non-cumulative and 
> cumulative. Is this cumulative mode also good to work with full dumps 
> (level0). E.g. the data I have to mirror is not large, and I don't want 
> to rely on inventory data. So I want for instance if between two full 
> dumps a file F was deleted, it should get deleted when I do xfsrestore 
> for the second time.
> 
> I don't keep the xfsdump output files but i just apply them via ssh. So 
> something like that would be great!
> 
> Maybe someone can shed some light on that.
> 

The cumulative mode really only relates to restore and not xfsdump.
xfsrestore in cumulative mode keeps its version of the directory tree
around (restoredir/xfsrestorehousekeepingdir/tree) so that it can use
it to compare with and decide if directory entries have been renamed,
deleted etc.. The code has been written to do this given an initial
level 0 dump and then delta dumps (incremental or resumed).
If you try to use a 2nd level 0 dump on restore then it will fail to
work - it won't restore anything.
It fails in xfsdump/restore/content.c/dumpcompat() when it
compares the restores uuid from persp->a.lastdumpid with the
incremental's uuid stored in scrhdrp->cih_last_id (and for a full
dump the cih_last_id won't be set).
Unfortunatly, a flag is set in restore so that it stops it from reporting
this error (not sure why).
I tried circumventing this uuid check (see the patch) and things
seemed to work ok (it did the renames and deletes).
However, I'm uncertain of any other repercussions - it's been a while
since I looked at cumulative restores. Maybe others might know.

I wonder if other tools might be useful to you... xfscopy,...?

--Tim