How to make zpool refer to a disk using a readable string?

Question

There are loads of ways a system might uniquely identify a disk or partition, GUID/UUID, how it's connected 'usb-...', and the traditional directory structure '/dev/sda'. zpool seems to choose randomly between them. How can I get a zpool status to list the array using the directory structure as it is the only thing other tools know about?

Further Information:

The history reveals how the pool was created:

zpool history XX
History for 'XX':
YYYY-MM-DD.HH:MM:SS zpool create -f XX -m /XX raidz sda sdb sdc sdd sde

However status now reads:

zpool status XX
pool: XX
...
                                                    STATE     READ WRITE CKSUM
      XX                                                         0     0     0
        raidz1-0                                                 0     0     0
          ata-WDC_WD10EFRX-68PJCN0_WD-XXXXXXXXXXXX  ONLINE       0     0     0
...

The names used on build are not the same as those now listed. The array has been moved around a lot once it was created however.

Update and conclusion:

It looks like most utilities can use the long name ZFS uses in place of the short, via

/dev/disk/by-id/*

say

smartctl --all /dev/disk/by-id/ata-WDC_...

While more cumbersome, I agree it is more precise.

score 1 · Accepted Answer · answered Dec 07 '17 at 08:32

1

zpool uses the device names you have given at pool creation time and when modifying devices (for example attaching disks or adding vdevs to the pool). Therefore, you can either destroy/recreate the pool with your chosen names, or detach/attach all devices one after each other (this is only possible with pool layouts that have enough redundancy, of course).

This is how it works on Solaris, there might exist specific caveats on other systems like Linux or BSD.

answered Dec 07 '17 at 08:32

user121391

1,272
7
19

This isn't the case in my experience, I've basically universally created pools using the `/dev/sdx` notation, only to after a while get back the long unintelligible strings. Alternately, is there a way in zpool to 'identify' a drive a command to light up one drive light say..? – J Collins Dec 07 '17 at 09:41
@JCollins On which operating system did this happen? IIRC BSD has had problems in the past when using the shorthand notifier, because the system reordered those names on reboot (which does not happen on Solaris). – user121391 Dec 07 '17 at 09:51
It is CentOS 7, kernel 3.10.0. Yes the drives come in with different names on every reboot. – J Collins Dec 07 '17 at 10:01
@JCollins See related: https://serverfault.com/questions/854979/how-to-change-the-drive-reference-in-a-zfs-pool-from-dev-sdx-to-dev-disk-by-id / https://superuser.com/questions/1019528/zfs-change-device-path / https://unix.stackexchange.com/questions/288599/forcing-zpool-to-use-dev-disk-by-id-in-ubuntu-xenial – user121391 Dec 07 '17 at 13:22
@user212391 A great source of information. Curiously people seem to be asking for the opposite of what I am, wanting the long names. Is there some reason why this would be? What other disk tools use these long names? In nearly every case I've seen so far, the /dev/sdx naming style is what a tool has asked for. – J Collins Dec 07 '17 at 13:41
1

@JCollins Most likely reason is to make replacement of disks easier, because you can easily spot your new (or old, but reinserted) disk by the name of its id/serial number. Especially when using large arrays (24, 48 or more disks) and frequent replacements, or with the help of dependent scripts, it would be easier. Also, the reshuffling would drive someone with OCD certainly mad ;-) – user121391 Dec 07 '17 at 14:15

How to make zpool refer to a disk using a readable string?

1 Answers1