Per title. From quick googling, I understand that zfs is a filesystem that provides some kind of software RAID and resiliency (eg CRC checks at the block level). Is my understanding right that it just good for detecting storage corruption, but not really self healing?
My home server just for a mini storage/ media server and I am broke to afford RAID. So at the moment, it’s just 2 consumer grade drives .
So pros with zfs I can think of, I get slightly better data resiliency.
But after reading the docs, the cons for me are that a) it’s not so straightforward/ somewhat complex to tune and use, b) it’s not default in mainline linux, and c) slightly lower performance?
What do you think and is it worth learning to use it vs just slapping a plain xfs/ ext4 ?
Thanks to btrfs I learned I had bad RAM that was corrupting files. Nothing was crashing, so I would never have noticed otherwise. 100% would recommend using a file system with checksumming, be it ZFS or btrfs.
it’s not default in mainline linux
This may or may not be a huge con depending on what distro you’re using. I guess a lot of folks like using something like TrueNAS, which I assume takes a lot of the pain of maintaining the ZFS install…
On the other hand, if you want freedom of distro choice… this might be more important to consider… I’m currently blocked on updating my Fedora IoT servers because the ZFS installation (that I had to manually do) some how broke… 😭
btrfs actually being part of Linux seems really nice. I just started with ZFS, so that’s what I know. I actually did look into switching to btrfs last year… but now I don’t remember why I stuck with ZFS… maybe I should check again.
I think BTRFS is easy enough (included in the kernel, pretty stable) and has most of the practical benefits of ZFS with little overhead. I like the snapshots and FS compression
Short answer: not for this machine. Your reading is right: on a single drive ZFS only detects corruption. It repairs a file only when a second copy exists, and with two drives that means a mirror. So the real question is what the second drive should do. I would give it a backup job rather than a mirror. A mirror survives one drive dying, but it also copies every deletion and every mistake to both drives at the same moment. A second drive holding a nightly copy (rsync, or restic if you want older versions) covers both cases, and then plain ext4 or XFS on the first drive is plenty. If you can, keep the backup drive in another room. ZFS or btrfs earns its place later, when you have more drives and want checksums and snapshots, and nothing you set up now stops you moving to it. Once the nightly copy runs, test it by restoring one file from the backup drive onto the first.
Snapshotting, which ZFS has, prevents any data loss from accidental deletion or mistakes.
Snapshots also occupy no space until you delete files
You could just use btrfs which as way less overhead
Alright gonna need some kind of citation for “way less overhead”
I love ZFS, and you don’t have to get deep in the technical stuff to enjoy the basic benefits. I’ve used it via TrueNAS, and it’s super easy to get set up.
Worth it 100%. Even for single-drive systems like laptops. It allows for a trivial full system backups while the system is running.
Zfs slaps and it’s so simple it doesn’t really need complex documentation, just get started and don’t fiddle with advanced options until you understand how they work
Both ZFS and BTRFS require having at least 2 copies for healing, however, unlike a standard RAID1 that is oblivious to which of the 2 copies is healthy, the 2 filesystems use CRC checksums to figure out which is corrupt and recover the good one. As long as you have at least 2 copies, they will check the checksum on every read and heal it if corruption is detected. Besides that, you’d schedule a periodic scrub to also check the data that hasn’t been read in a while.
In ZFS you can tell it to keep 2 copies of everything even if you only have 1 drive, which would give you the healing effect, but this obviously halves the drive capacity, and does not save you if that 1 drive dies, hence spreading it on multiple drives is better.
Personally, I find ZFS easier to use than BTRFS, it’s not that complicated once you get the hang of it, plus LLMs can help anyway. It does have many dials that can be tuned, but unless you are trying to optimize it for a specific usecase, the generally recommend setting are fine for most use cases.
It’s not built into the kernel, but server distros (e.g. Ubuntu) build it as a loadable module, which is kept in step with the kernel, so generally no issues there. On distros like Arch it can be really annoying though.
“Slightly lower” is a bit of an understatement, both for ZFS and BTRFS. Those are CoW (Copy on Write) filesystems, so they are much slower to write, which is a traid off for the benefits of CoW. On the other hand, reads are only slightly slower (since CRC is always checked), and thanks to transparent compression, they can be even faster to read.
One downside of ZFS over BTRFS is fragmentation, all CoW filesystems suffer from it, but BTRFS allows you to defragment in place (takes a while though) unlike ZFS.
If you can’t afford storage, the data you want to store is generally compressible, and speed is not critical/priority, it might be woth using ZFS/BTRFS just for the transparent compression to save on space and get more out of the drives you have. I’m using BTRFS on a Fedora system for this reason alone, helps me avoid buying a new SSD 😬
I’d go with zfs. You’ll get copy-on-write snapshots that take a split second to create and do not add any read or write overhead, online filesystem checking, detection of corrupt files down to the filename in the filesystem and all snapshots, lz4 file compression, and more. I even use zfs on single disk usb drives just because the features it offers are so good.
ZFS shines with large datastores and multiple drives. It has its uses, for sure.
As others have said, it puts all the layers: RAID blocks, file system, volume management, etc, all under a single umbrella. Because of this it can do some things that are more challenging with other approaches.
For your two drives, I wouldn’t bother with anything other than what they are, and make sure you have proper backups.
Drives die all the time. Just today one of my RAID boxes complained about a drive, in the process of replacing the drive I discovered 2 other bad drives that I had sitting around as spares.
Last week an external drive decided to die. Fortunately I was using it as one of three replicants of my primary system, so no big deal.
That is a good point. A backup is probably higher priority over choosing a filesystem assuming you already have something functional and no backup.
RAID is not a backup, what RAID gives is availability, what you should prioritize depends on the usecase, but generally proper backups are usually the priority
It’s my personal opinion but: no
Btrfs is integrated in the kernel already and pretty solid. I have been running it for the last few years. I use Linux mdarray to create a raid5 and on top of it, format btrfs.
Never had issues, and yes it did catch data corruption (and fixed it) so I am pretty happy .
Zfs on the other hand, needs specific kernel patches and I don’t like that too much.
ZFS has been running with upstream kernels for years. Ubuntu ships with precompiled module and Debian has a DKMS package which compiles ZFS against the stock kernel.
fully agree.
i used zfs on arch many yeara ago and my updates would fall months behind.
then i switched to freebsd for better zfs support. but lost support of some linux apps
later i started using truenas and it works great.
on a related note, i switched to cachyos on my desktop and laptop and used btrfs for the first time and i absolutely love it.
so if youre building a nas from scratch, id say use btrfs, but if youre using an os that manages that stuff for you, zfs is a great choice.
If you only require mirrors (RAID 1) then BTRFS works fine, but if you need something more and the system has good support for it (e.g. Ubuntu) then ZFS is better IMO. Yes you can use mdraid with BTRFS, but I prefer 1 ZFS over 10 different moving parts.
I use both, choosing which one depending on the conditions.
I use and highly recommend truenas. It uses zfs by default and has a nice web ui, so you don’t have to worry about setting up zfs. It has docker built in and can also do virtual machines and containers.
I’ve used it for about 4 years now. It is a good platform with a few limitations, but largely worth it. Do not use anything but TrueNAS SCALE or you’ll be disappointed by the availability of packages for it.
Iirc scale is now the mainline, even.
2 legends about ZFS seem never to die: that it is difficult in normal linux, and that it needs much extra RAM.
Truth is, it is availible in the normal kernel now and several distros offer it from start. There are tools for maintaining it, and after a while, you will find that it needs less routine maintenance that most traditional file systems.
Truth is also that it can absolutely work with less RAM. That thing is only important for servers with a high load. There it is recommended strongly to have lots of RAM, but even there it is not for the sake of ZFS itself, but it is for the performance of your server. It makes good use of all that RAM in ways that other file systems cannot.
Is my understanding right that it just good for detecting storage corruption, but not really self healing?
If you use it on just 1 drive, and the default setting, then this is true. It still detects bit rot, but it cannot repair it automatically. But it has also an option for the self healing on only 1 drive (for the price of drive space then).
If you use 2 mirrored drives, or more as a raidz, the self healing is always there.
Truth is, it is availible in the normal kernel now and several distros offer it from start.
You may want to rephrase this. ZFS license (CDDL) is still incompatible with the Linux kernel (GPL) so it’s not part of the kernel sources and needs to be compiled as an out-of-tree module against the kernel you’re running.
Some distros make such modules available by default with some of their official kernels, some don’t. You can get it installed on most distros but the manner of doing so and the amount of support you can expect from the distro varies wildly. If install/update is a manual operation you run the risk of the module falling out of sync with the running kernel.
That beeing said. For most popular ‘server’ distros its available in the packet sources and the hussle to install it is just a single command to the packet manager.
It you run custom kernels, or do anything that is not officially supported then obviously you do not have the support and have to risk having it falling onto your feet.
Mostly scarecity.
I chose btrfs over zfs, because you can add and remove disks from a volume (I’m not sure if zfs has this feature now) I use this to move the volume from old to new disks.
Yeah btrfs user here as well. But I never tried zfs, so cannot compare.
Self healing for bit-rot in files. So if your data is critical it is worth it.
it does report bitrot, but for self healing it needs duplication set up right?
You need a mirror copy so that when it checks the checksum it can go and pull the good copy
Raidz will repair as well.











