Re: ZFS zombie pool haunting me, please send help (ZFS bugs?)

[mlelstv%serpens.de@localhost (Michael van Elst)]

el%horse64.org@localhost (Ellie) writes:

>2. When a zpool is created using a /dev/dkX numbered wedge, then zpool 
>export'ed, then seen on a new system with a different /dev/dkX wedge 
>(but same partition, can happen e.g. if it's on USB and devices end up 
>enumerated differently), then the new system sees nothing with zpool 
>import -a either. This also seems like a pretty hard to work around bug.


ZFS caches pool configuration in /etc/zfs/zpool.cache. If there is
such a cached configuration, it relies on it and doesn't scan devices.

Renaming devices causes confusion as it is not reflected in the cache.

Removing the cache file will scan disks (from /dev) again and store the
new information again in the cachefile. Setting the 'cachefile' attribute
of a pool to 'none' will prevent the configuration from being stored
into the cachefile, but then it also won't be imported at boot (and
importing seems to clear the attribute again).

N.B. I'm using persistent names also for disks (maintained by devpubd):

% ls -l /dev/disk/
total 0
lrwxr-xr-x  1 root  wheel  8 Nov  1 07:25 wwn-5000039AE8CBF2E1 -> /dev/wd0
lrwxr-xr-x  1 root  wheel  8 Nov  1 07:25 wwn-5000039AE8CBF33D -> /dev/wd1

# zpool list -v
NAME                            SIZE  ALLOC   FREE  EXPANDSZ   FRAG    CAP  DEDUP  HEALTH  ALTROOT
tank                             29T  4.49T  24.5T         -     4%    15%  1.01x  ONLINE  -
  raidz1                         29T  4.49T  24.5T         -     4%    15%
    disk/wwn-5000039AE8CBF2E1      -      -      -         -      -      -
    disk/wwn-5000039AE8CBF33D      -      -      -         -      -      -