On Fri, 17 Apr 2009, Mark J Musante wrote:
The dependency is based on the names.
I should clarify what I mean by that. There are actually two dependencies
here: one is based on dataset names, and one is based on snapshots and
clones.
If there are two datasets, pool/foo and pool/foo/bar, then you cannot
destroy pool/foo without also destroying pool/foo/bar.
But, if pool/foo/bar is ALSO an ancestor (e.g. clone-of-a-snapshot) of
pool/foo, then you cannot destroy pool/foo/bar without destroying
pool/foo.
That's the situation that happens in the following sequence:
zfs create testpool/testfs2
zfs snapshot testpool/test...@snap1
zfs clone testpool/test...@snap1 testpool/testfs2/clone1
zfs promote testpool/testfs2/clone1
The 'clone' command creates a double-relationship: it created child
dataset of testpool/testfs2 at the same time it created a child dataset of
testpool/test...@snap1. And the snapshot is also a child dataset of
testpool/testfs2.
Thus far, everything is OK. You can still destroy
testpool/testfs2/clone1, because both the relationships to
testpool/testfs2 is just in one direction: the latter is the parent of the
former.
But, when the 'promote' command is used, it inverts *one* of those
relationships without inverting the other.
In other words, testpool/testfs2 was now a child dataset of the snapshot
testpool/test...@snap1 (and therefore a descendant of
testpool/testfs2/clone1), *but* it was also the parent of
testpool/testfs2/clone1 by virtue of its name.
So in the "namespace", testpool/testfs2 is still the parent of
testpool/testfs2/clone1. But in the "clonespace", testpool/testfs2 has
now become the CHILD of testpool/testfs2/clone1 through the
testpool/test...@snap1 snapshot.
Then when you try to destroy them both with -R, you end up with the
"recursive dependency" error message, otherwise known as CR 6622809.
See http://bugs.opensolaris.org/view_bug.do?bug_id=6622809
Regards,
markm
_______________________________________________
zfs-discuss mailing list
zfs-discuss@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
