ZFS: Can't find pool by guid

[Eugene Grosbein]

24.10.2018 13:35, Eugene M. Zheganin wrote:

> On 28.04.2018 17:46, Willem Jan Withagen wrote:
>> Hi,
>>
>> I upgraded a server from 10.4 to 11.1 and now al of a sudden the server complains about:
>>     ZFS: Can't find pool by guid
>> And I end up in the boot prompt:
>>
>> lsdev gives disk0 withe on p1 the partion that the zroot is/was.
>>
>> This is an active server, so redoing install and stuf is nog going to be real workable....
>>
>> So how do I get this to boot?
> The basic scenario for this is when you have a "shadow" pool on the bootable disks with actual root pool - for example once you had a zfs pool on some disks that were in dedicated mode, then you extracted these disks without clearing the zpool labels (and 'zpool destroy' never clears the zpool labels) and installed the system onto them. This way 'zpool import' will show the old pool which has no live replicas and no live vdevs. The system on it may be bootable (and will probably be) until the data gets redistributed in some way, after that gptzfsboot will start to see the old pool remains, will try to detect if this pool has bootfs on it - but in this case there's no valid pool - so it will fall into error and stop working. Actually, the newer 11.2 gptzfsboot loader has more support of this - it clearly states the pool found and mentions the error - thanks to all the guys that did a great work on this, seriously.
> 
> The way to resolve this is to detach disks sequentially from root pool (or offline them in case of raidz), making 'zpool labelclear' on them (please keep in mind that 'labelclear' is evil and ignorant, and breaks things including GPT table) and attaching them back, resilvering, and repeating this until 'zpool import' will show no old disassembled pools. Determining which disks have the old labels can be done with 'zdb -l /dev/<disk> | grep name:'.
> 
> I understand that your situation was resolved long ago, I'm writing this merely to establish a knowledge point if someone will step on this too, like I did yesterday.

There is quicker and easier way: just rename real pool to another name and use vfs.root.mountfrom=zfs:newpoolname/fs