zfs destroy dry-run lying about reclaimable space?

[Marcus Müller]

On 17.03.2015, at 11:18, Marcus Müller <znek at mulle-kybernetik.com> wrote:

> I was a bit stumped by the fact that the sum of the reclaimable space reported via dry-run of zfs-destroy-snapshots.sh wasn't anywhere near the supposed "usedbysnapshots" value of the filesystem, so I just deleted all snapshots for real to see what would happen. Unfortunately I cannot zfs diff any of them now to get any clues why the dry-run would be so wrong, but maybe anyone else knows? Is this a known bug already?

A bit more context:

root at fbsd:(/)# uname -a
FreeBSD fbsd.z.net 10.1-PRERELEASE FreeBSD 10.1-PRERELEASE #0 8e3e4f9(stable/10)-dirty: Wed Nov 19 16:06:59 CET 2014     root at fbsd:/usr/obj/usr/src/sys/tank  amd64
root at fbsd:(/)# zfs get version tank/usr/local
NAME             PROPERTY  VALUE    SOURCE
tank/usr/local   version   5        -
root at fbsd:(/)# zpool get all tank
NAME        PROPERTY                       VALUE                          SOURCE
[...]
tank        feature at async_destroy          enabled                        local
tank        feature at empty_bpobj            active                         local
tank        feature at lz4_compress           active                         local
tank        feature at multi_vdev_crash_dump  enabled                        local
tank        feature at spacemap_histogram     active                         local
tank        feature at enabled_txg            active                         local
tank        feature at hole_birth             active                         local
tank        feature at extensible_dataset     enabled                        local
tank        feature at embedded_data          active                         local
tank        feature at bookmarks              enabled                        local
tank        feature at filesystem_limits      enabled                        local

The pool was only upgraded recently, but it already had feature flag support before (don't recall which feature flags were new). Also, I don't recall when the filesystem version was upgraded from 4 to 5, but I presume that's been before the snapshots had been created (I did search in the pool history, but didn't find an upgrade statement. This pool had been created last year and populated with a recursive dump of another pool which has been destroyed in the meantime, thus no history logs for this.)


Cheers,

  Marcus

--
Marcus Müller  .  .  .  http://www.mulle-kybernetik.com/znek/



-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 2326 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-fs/attachments/20150317/ab2bc1af/attachment.bin>