PR 11813

[bghira]

#11813

I don't think that naming a pool after a disk is "reasonable"..

There's a link in /dev/poolname/ to /dev/zdXYZ for zvol device paths. if you name a pool after a disk, it could appear as /dev/sda1/volume1.

[mskarbek]

But you can name pool after disk - both sda and sda1 work "perfectly well". It's only the Solaris-based disk naming convention that is currently filtered. Linked PR is not about naming pools after disk but allowing to name pool in a way that avoids putting in scripts workarounds such as mine:

function gen_pool_uuid {
    while true; do
        POOL_UUID=$(cat /proc/sys/kernel/random/uuid)
        if [[ $POOL_UUID =~ ^([a-b]|[d-z]).* ]]; then
            echo $POOL_UUID
            break
        fi
    done
}

Yes, I name my pools using UUID. 🤷‍♂️

[bghira]

yes, good point. it should be filtered for all platforms.

[mskarbek]

Why? How? Just mention in documentation that naming pools after disks is dumb and can hurt you. That's it. If I want to name my pool sda-something then why not? Simple filtering based on sd.* or nvme.* rules will behave similarly to the Solairs-based one preventing me from using such a name.

[bghira]

no one reads the documentation.

[bghira]

we're presently running restrictions on the bootfs property for zpools, across different platforms; making sure your bootloader doesn't get updated is something that can be simply mentioned in the docs, because it's "dumb and can hurt you". i see no reason why we shouldn't filter problematic names across platforms, too. of course, if we shouldn't do one then we shouldn't do the other.