[Discuss] Debian 11 -> 12
Kent Borg
kentborg at borg.org
Thu May 30 16:26:32 EDT 2024
On 5/30/24 11:25, markw at mohawksoft.com wrote:
> I don't understand why you think ZFS has any more base complexity than
> something like LVM.
I admit it is a matter of taste that I find zfs ornery. It is trivial,
but I find it annoying that I can't use mount to mount a zfs volume.
I've got to use zfs for everything, and tell *it* to mount or umount.
Except for different things where I have to use zpool to do stuff.
I wanted to use zfs for encrypted external backups (no redundant disk,
just a single disk, but I wanted to have the checksummed features of zfs
to assure me my bits weren't rotting), but where my system already knows
how to deal with other kinds of external backups, I need to manage zfs
manually, and if I forgot the "export" the pool before unplugging I seem
to remember it left my computer in an unhappy state. (I thought
unplugging a disk at the wrong time risked damaging the disk, not
damaging the running computer.)
So I wrote a pair of scripts for mounting and unmounting, and I find
that annoying. Again trivial, but still annoying.
> Dude, I had ZFS running on a RPI4 and just upgraded it to an RPI5. Zero
> issues.
Checking back in my notes I was wrong, it wasn't a crash, I was getting
IO errors, from both devices, when doing a torture test of copying a
bunch of files. This was using a powered hub plugged into a Pi 4 fast
USB port. When the hub was plugged into a slow port it worked, with no
errors.
Using the same drives plugged into the some hub plugged into a 64-bit
Intel notebook, zfs worked, with no IO errors.
Maybe the Pi has broken fast USB ports, except the same hub and the same
drives plugged into the same fast port on the same Pi 4, it works for SW
raid 1, with no errors.
Maybe the Pi has a subtle fast USB problem that only happens with that
hub and zfs. Maybe I am not smart enough to use zfs. Clearly zfs works
in some circumstances, but eventually I ran out of patience and went
with something that did work for me.
Why do I compile my own Raspberry Pi 4 kernel? Because I don't trust SD
cards, I frequently manage this machine remotely and I want reboots to
work, so I want to boot from more robust devices. In my current
configuration if one disk is missing or completely dead the Pi will
still try the other, boot, and run. (Yes, if the first disk it tries
only *sorta* works it can certainly still fail to boot.)
In the normal case, when things are working, and the computer is up and
booted, the vfat boot volume and / are both sitting on SW raid 1
devices. Not only am I booting from a device that is more reliable an SD
card, I am booting from a redundant pair of them. The stock kernel
didn't seem capable of this, I think I needed stuff linked that they
only built as modules, I'd have to check my notes to be sure.
New Pi 4 kernel sources, at tag stable_20240529, happened to come out
yesterday, so earlier today I rebuilt the kernel with them and installed
it. I installed it just once, but on top of raid 1, so I now have
redundant copies. I like that.
-kb
More information about the Discuss
mailing list