Follow

by default uses device numbers for .

They're apparently not pinned to the controller and port.

I removed two devices on one controller, and every other device number shifted up by 2 causing the boot to fail 😒

Trying to convert a FreeBSD to use disk ids rather than numbered devices.

Somehow I've stumbled back into a hell I haven't visited in a long while

Show thread

@rtyler I was under the impression that #FreeBSD #ZFS would find the right disks even if disks are detected in different order than when a pool was created🤔

It has been while since I used raw device names ever since I got in habit of labeling the partitions; I create a large, single partition even when using the whole disk in a pool.

By chance are/were you using "/etc/fstab" to mount file systems?

@parvXm For better or worse, I was running off a ZFS pool created by the guided installer.

Trying to get the devices relabeled in the pool has proven to be quite a pain in the ass

@rtyler Yeah, #FreeBSD #installer not creating a label & using that label to create a #ZFS pool, even if it is "root" pool, is certainly a thorn in the back side😒
#bsdinstall

@rtyler Had the exact same issue on #NetBSD. The zpool is based on numbered dk devices. If they get renumbered for whatever reason, welcome to manual recovery

@bentsukun @rtyler Tried a few ways of setting up devices for #zfs on #netbsd - raw devices, numbered dk wedges, and named wedges with /etc/rc.d/devpubd (by default on netbsd-9 devpubd runs _after_ zfs, which means an additional reboot to an autoconfigured system), and named wedges have ended up working well, even in the case of renumbering devices

Oh, and concatenating mixed size disks into something that _looks lke_ a viable zfs setup, but thats another story :) netbsd0.blogspot.com/2022/05/t

@rtyler that shouldn't happen. Zfs has metadata and the device name doesn't matter per say. Devices can move around and it should be fine. Now I have had issues with zfs where if I replaced a drive, but left the old drive in the machine, the kernel would find the old drive and some of the new drives, but not all of them and not boot. The solution was to remove.thd old drives when booting so zfs wouldn't be confused.

The short is that zfs just uses the first drives it finds, even if there is an actual configuration that would work.

Sign in to participate in the conversation
Hacky Town!

Welcome to hacky.town, a small Mastodon server for free and open source data hackers