(SOLVED) ZFS support in DietPi

General Discussion
21

There does not seem to be a way to format a drive in ZFS using dietpi-drive_manager. Or am I doing something wrong?

22

ZFS is not open source, so it is not part of the Linux kernel.

But you could try OpenZFS

23

Yup, I compiled the dkms module and installed the ancillary packages already. Are the options in the drive manager hardcoded or picked up from the available modules?

24

Nope they are not dynamic.

Anyway you can format the drive without drive manager same way as one any other Linux system. There is no need to use drive manager.

25

I’m just a bit worried that it would overwrite /etc/fstab with ZFS set up whenever I mount anything else. Unless ZFS works differently and doesn’t use fstab (I never used ZFS before).

26

ZFS is not mounted via /etc/fstab, regardless of DietPi.
I’m also curious what device you are using, ZFS needs some RAM and CPU time. 4 GB RAM is recommended for the use of ZFS.

27

Rock 5b with 13 GB of RAM (limited artificially because some chips seem to be damaged). I just wanted to try it with eMMC only to see what it’s about.

28

For testing it might be fine, but I wouldn’t recommend it on a productive system on eMMC. ZFS writes a lot of metadata and eMMC is pretty slow and wears out relatively quickly

29

Does ZFS do any writes to disk when the user is not doing any writes? From what I gathered, even with one drive it’s still basically one of the more reliable filesystems in general. And that any RAM is used mostly for caching, not that it is absolutely required for anything.

30

This is out of my knowledge so I asked chatGPT :sweat_smile:

1. Background writes (even when idle)

  • ZFS metadata updates
    • Even if no user data is written, ZFS may update internal metadata occasionally, like transaction groups.
    • Every txg (transaction group) is flushed to disk periodically (default: ~5 seconds on modern ZFS).
    • On a single-drive pool, this still happens, but the impact is minimal.
  • Pool state
    • ZFS periodically updates uberblocks and pool state information to ensure consistency.
  • Scrub/Resilver
    • Only runs if explicitly triggered.
  • No random I/O – if the system is idle, writes are generally just the transaction group commits.
31

There’s an issue with the zfs module when upgrading the kernel See the pastebin:

Compilation works only after running sudo apt -f install.

This seems to be an underlying problem in how the update process works. Any idea if it can be fixed?