How to install TrueNAS on a Raspberry Pi
jeffgeerling.com
jeffgeerling.com
I liked FreeNAS for awhile, but after a certain point I kind of just learned how to properly use Samba and NFS and ZFS, and after that I kind of felt like it was just getting in the way.
Nowadays, my "NAS" is one of those little "mini gaming PCs" you an buy on Amazon for around ~$400, and I have three 8-bay USB hard drive enclosures, each filled with 16TB drives all with ZFS. I lose six drives to the RAID, so total storage is about ~288TB, but even though it's USB it's actually pretty fast; fast enough for what I need to for anyway, which is to watch videos off Jellyfin or host a Minecraft server.
I am not 100% sure who TrueNAS is really for, at least in the "install it yourself" sense; if you know enough about how to install something like TrueNAS, you probably don't really need it...
Now I just run Ubuntu/Samba and use KVM and docker for anything that doesn't need access to the underlying hardware.
Similarly while docker's CLI interface is relatively nice, it's even nicer to just take my phone, open a browser, and push "update" or "restart" in a little gui to quickly get things back up & going. Or to add new services. Or whatever else I want. Sure I could SSH in from my phone, but that's awful. I could go get a laptop whenever I need to do something, but if Jellyfin or Plex or whatever is cranky and I'm already sitting on the couch, I don't want to have to get up and go find a laptop. I want to just hit "restart service" without moving.
And that's the point of things like TrueNAS or Unraid or whatever. It makes things nicer to use from more interfaces in more places.
But for probably 90% of users, they just want a UI they can click through and mostly use the defaults, then log into now and then to make sure things are good.
The UI is also especially helpful for focusing on things that matter, like setting up scrubs, notifications, etc. (though even there I think TrueNAS could do better).
It's why Synology persists, despite growing more and more hostile to their NAS owners.
I guess the thing is that I've never done that with [Free|True]NAS :). I've always used some kind of hardware (either retired rack mount servers or thin clients) and installed FreeNAS on there, and then it never really felt like it saved me a lot of time or effort compared to just doing it all manually.
Of course, I'm being a typical "Hacker News Poster" here; I will (partly) acknowledge that I am the weird one, but at the same time, as I said, if you're in the market to install TrueNAS on your own hardware, you are also probably someone who could relatively easily Google your way through doing it manually in roughly the same amount of time, at least with NixOS.
But this is one aspect of my life that I look at as core infrastructure that needs to just work, and I don’t really gain anything from rolling my own in this particular category.
I still run a mini home lab where I do all kinds of tinkering. Just not with storage.
But I also completely understand wanting to do this all manually. I’ve been there. That’s just not where I am today.
Having a playground/homelab at home is one thing, but playing with your family's data and access to it can get annoying really fast.
It's a little different though; TrueNAS still requires a fairly high level of technical competence to install on your own hardware; you still need to understand how to manage partitions and roughly what a Jail is and basic settings for Samba and the like. It's not completely analogous to Dropbox because Dropbox is trivial for pretty much anyone.
The problem with TrueNAS is it fills the niche where it’s targeting people who want to tinker, but don’t want to learn how to tinker. Which is likely a smaller demographic than those who are willing to roll their own and those who just want a fully off-the-shelf experience.
I also think Synology would be closer to the Dropbox experience than TrueNAS.
Sure, but both are free options, one just takes strictly less work to do the task of being a NAS. Why would I pick the one that's just more work for the same result? If I want a NAS, why would I roll my own with NixOS instead of just picking a distro that focuses on being a NAS out of the box? What's the benefits of doing it manually?
If I want to just play around with stuff in a homelab setting, that's what proxmox clustering is for :) But storage / NAS is boring. It just needs to sit there doing basic storage stuff. I want it to do the least amount of things possible because everything else depends on storage being there.
> But for probably 90% of users, they just want a UI they can click through...
I think if you've gone through the effort of setting up Ansible scripts for setting up & maintaining a NAS, you probably are not actually making a NAS anymore. Like maybe you're doing Ceph or Gluster cluster or something, which can be fun to play with. Heck, I did that with a bunch of ODROID-HC2's as well. It was fun to setup a cluster storage system.
It also wasn't practical at all and at no point did it ever seriously compete with replacing my "real" NAS (which currently is Unraid, but I'd absolutely consider switching to TrueNAS in a future upgrade), since the main feature that a NAS needs to provide is uptime.
I agree for the most part, though even vanilla Ubuntu has Cockpit if you need a GUI.
Personally I find that getting it set up with NixOS is pretty straightforward and it's "set and forget", and generally it's not too hard to find configurations you can just copypaste done "correctly". And of course, if you break something you can just reboot and choose a previous generation. Of course restarting still requires SSHing in and `systemctl restart myapp`, so YMMV.
A few big shares is all I really need; I no longer create a share for every single idea/thing I can think of.
It actually has been considerably more reliable than the MediaSonics than they replaced.
I'm looking to retire my power hungry 4790k server and go with a mini pc + external storage.
I wouldn't do USB-connected enclosures for NAS drives - either SATA via SFF, or Thunderbolt.
The cost is very low, but it would have to be a nvme only build.
If you do a vanilla FreeBSD+samba+NFS+ZFS setup, you'll need to edit several files around the file system, which are easy to forget months down the line in case of adjustment or disaster recovery.
?!?
How do you fill 288 TB? Is it mostly media?
>I liked FreeNAS for awhile, but after a certain point I kind of just learned how to properly use Samba and NFS and ZFS, and after that I kind of felt like it was just getting in the way.
I've been a mostly happy TrueNAS user for about four years, but I'm starting to feel this way.
I recently wrote about expanding my 4-disk raidz1 pool to a 6-disk raidz2 pool.[1] I did everything using ZFS command-line tools because what I wanted wasn't possible through the TrueNAS UI.
A developer from iXsystems (the company that maintains TrueNAS) read my post and told me that creating a ZFS pool from the zfs command-line utility is not supported, and so I may hit bugs when I use the pool in TrueNAS.
I was really surprised that TrueNAS can't just accept whatever the state of the ZFS pool is. It feels like an overreach that TrueNAS expects to manage all ZFS interactions.
I'm converting more of my infrastructure to NixOS, and I know a lot of people just manage their NAS with NixOS, which is sounding more and more appealing to me.
[1] https://mtlynch.io/raidz1-to-raidz2/
[2] https://www.reddit.com/r/truenas/comments/1m7b5e0/migrating_...
I kind of purposefully don't fill it up :).
This has been a project of tech hoarding over the last ~8 years, but basically I wanted "infinite storage". I wanted to be able to do pretty much any project and know that no matter how crazy I am, I'll have enough space for it. Thus far, even with all my media and AI models and stock data and whatnot, I'm sitting around ~45TB.
On the off chance that I do start running low on space, there's plenty of stuff I can delete if I need to, but of course I probably won't need to for the foreseeable future.
> I'm converting more of my infrastructure to NixOS, and I know a lot of people just manage their NAS with NixOS, which is sounding more and more appealing to me.
Yeah, that's what I do, I have my NixOS run a Samba on my RAID, and it works fine. It was like fifteen lines of config, version controlled, and I haven't thought about it in months.
older nas was 4x 4tb drives when I retired it and passed it to a friend.
Current nas is a 6-bay synology with ram and nvme upgraded, along with a 5-bay expansion. All with 12tb drives in a 6drive 2-parity and 5-drive 2-parity arrays. I'm just under 40tb used mostly media.
I have one enclosure plugged into a USB 3.0 line, another plugged into a "super speed" line, and one plugged into a Thunderbolt line (shared with my 10GbE Thunderbolt card with a 2x20 gigabit splitter).
This was deliberate, each is actually on a separate USB controller. Assuming I'm bottlenecked by the slowest, I'm limited to 5 gigabits per RAID, but for spinners that's really not that bad.
ETA: It's just a soft RAID with ZFS, I set up each 8-bay enclosure with its own RAIDZ2, and then glued all three together into one big pool mounted at `/tank`. I had to do a bit of systemd chicanery with NixOS to make sure it mounted after the USB stuff started, but that was like ten lines of config I do exactly once, so it wasn't a big deal.
But often after extensive use, `dmesg` would complain about problems when talking to the drives, e.g. drive not responding or other strange error messages (forgot the exact text but very irritating and google-fu didn't help). There were also problems with SMART commands passthrough and drive power management e.g. sleep/standby adjustment which wasn't reliable when talking via bridge chips.
I use only disks directly connected to SATA controllers afterwards and no such issues ever happened again.
Though even flaky adapters just tend to lock up, I think.
ETA: It would certainly take longer with more data. I've not gotten anywhere close to the 288TB
In the last few years I've transitioned to "My family just wants plex to work and I could give a shit about the details". I think I'm more of the target audience. When I had my non-truenas zfs set up I just didn't pay a lot of attention, and when something broke it was like re-learning the whole system over again.
I accidentally gave myself the opportunity to test out my automations when I decided I wanted to rename my k8s nodes (FQDN rather than just hostname). When I did that, everything broke, and I decided it would be easier to simply re-provision than to troubleshoot. I was up and running with completely rebuilt nodes in around an hour.
In my experience, a vanilla install and some daemons sprinkled on top works better than these GUI flavours.
Less breakage, fewer quirks, more secure.
YMMV and I’m not saying you’re wrong - just my experience
Not to jinx it, but I have never had a hard drive failure since around 2008!
i've wanted to get a NAS running again, but while the QNAP form factor is great, the QNAP OS was overkill – difficult to manage (too many knobs and whistles) – and ultimately not reliable.
so, i'm at a junction: 1) no NAS (current state), 2) custom NAS (form factor dominates this discussion – i don't want a gaming tower), or 3) back to an off-the-shelf brand (poor experience previously).
maybe the ideal would be a Mac Mini that i could plug 4 HDDs into, but that setup would be cost-inefficient. so, it's probably a custom build w/ NixOS or an off-the-shelf, but i'm lacking the motivation to get back into the game.
ZFS on macOS sucks really bad, too, so that rules out the obvious alternative.
- External enclosure disconnected frequently (this is more of an issue with the enclosure and its chipset, but I bought a reputable one)
- Many services can't start without being logged in
- If you want to use FileVault, you'll have to input your password when you reboot
Little things went wrong too frequently that needed an attended fix.
If you go off the shelf, I recommend Synology, but make sure you get an Intel model with QSV if you plan to transcode video. You can also install Synology OS to your own hardware using Xpenology - its surprisingly stable, moreso than the mac mini was for me.
How about if you _don't_ plan to transcode video? For example, this years models DS425+ uses the (six year old) Intel Celeron J4125, while the DS925+ uses the (seven year old) AMD Ryzen Embedded V1500B. Why choose one over the other?
They can be had for a bit less than the Mac mini, and I had no issues getting headless Linux working on there. I even have hardware transcoding in Jellyfin working with VAAPI. I think it cost me about $400.
“Have to”, since I no longer suggest virtualizing TrueNAS even with PCI passthru. I will say the same about zfs-over-USB, but you do you. I’ve had too many bad experiences with both (for those not on the weeds here, both are officially very much not supported and recommended, but they _do_ work).
I really like the TrueNAS value prop - it makes something I’m clearly capable of by hand much easier and less tedious. I back up both my primary zfs tank and well as my PBS storage to it, plus cold backups. It does scheduling, alerts, configuration, and shares, and nothing else. I never got the weird K8s mini cluster they ship - seems like a weird thing that clashes with the core philosophy of just offering a NAS OS.
At first, I thought I might just go with TrueNAS. It can manage my containers and my storage. But it's got proprietary bits, and I don't necessarily want to be locked into their way of managing containers.
Then my plan was to run Proxmox with a TrueNAS VM managing a ZFS raidz volume, so I could use whatever I want for container management (I'm going with Podman)
But the more I've researched and planned out this migration, the more I realize that it's pretty easy to do all the stuff I want from TrueNAS, by myself. Setting up ZFS scrubbing and SMART checks, and email alerts when something fishy happens, is pretty easy.
I'm beginning to really understand the UNIX "do one thing and do it well" philosophy.
It's actually not too bad; the main "server" idles at around 14W and the power supply for it only goes to 100W under load. The drive bays go up to 100W (I think) but generally idle around 20W each. All together it idles at around ~70-80W.
Not that impressive BUT it replaced a big rack mount server that idled at about 250W and would go up to a kilowatt under load.
> Should you care about the "USB bottleneck"?
Not really, at least not for what I'm doing. I generally can get pretty decent speeds and I think network is often the bottleneck more than the drives themselves.
> How do you organize this thing so it's not a mess of USB cables?
I don't :). It's a big mess of USB cables that's hidden in a closet. It doesn't look pretty at all.
4k Blu-ray rips peak at over 100 Mbps, but usually average around 80 Mbps. I don't know how much disk I/O a Minecraft server does ... I wouldn't think it would do all that much. USB2 (high-speed) bandwidth should be plenty for that; although filling the array and scrubbing/resilvering would be painful.
That said, I have lots of home movies that just so happen to be at the exact same bitrates as Blu-rays and after the initial setup, I've never really had any issues with them choking or any bandwidth weirdness. Minecraft doesn't use a ton of disk IO, especially since it is rare that anyone plays on my server other than me.
I do occasionally do stuff that requires decent bandwidth though, enough to saturate a WiFi connection at the very least, and the USB3 + USB SS + Thunderbolt never seems to have much of an issue getting to Wifi speeds.
The only reason I can see is "I have one that I don't use". Because otherwise...
Idle power isn't all that much better than a low power Intel N100 or something similar. And it's all downhill from there. Network transfer speeds and disk transfers will all be kneecapped by the (lack of) available PCIe lanes. Available RAM or CPU speeds are even worse...
> I've found numerous times, running modern applications on slower hardware is an excellent way to expose little configuration flaws and misconceptions that lead to learning how to run the applications much better on more capable machines.
It's less about the why, and more about the 'why not?' :)
I explicitly don't recommend running TrueNAS on a Pi currently, at the end (though I don't see a problem with anyone doing it for the fun, or if they need an absolutely tiny build and want to try Arm):
> Because of the current UEFI limitations, I would still recommend running TrueNAS on higher-end Arm hardware (like Ampere servers).
[1] https://www.jeffgeerling.com/blog/2024/radxas-sata-hat-makes...
imo, if the software doesn't work without issue on my Pi, it isn't good enough for prod.
The SATA controller isn't terrible, but it and other hardware areas have had many strange behaviors over the years to the point of compiling the kernel being needed to fiddle with some settings to get a hardware device to do what it's supposed to.
Even if you're using power that is well supported eventually you seem to hit internal limits and get problems. That's when you see people underclocking the chip to move some of this phantom power budget to other chips. Likewise you have to power most everything from a separate source which pushes me even closer to a "regular PC" anyhow.
I just grab an old PC from Facebook for under $100. The current one is a leftover from the DDR3 + Nvidia 1060 gaming era. It's a quad core with HT so I get 8 threads. Granted most of those threads cause the system to go into 90% usage even when running jobs with only 2 threads, probably because the real hardware being used there is something like AVX and it can't be shared between all of the cores at the same time.
The SATA controller has been a bit flaky, but you can pick up 4-port SATA cards for about $10 each.
When my Raspberry Pi fails I need to start looking at configurations and hacks to get the firmware/software stack to work.
When my $100 random PC fails I look at the logs to find out what hardware component failed and replace it.
That's not the right explanation; each physical core has its own vector ALUs for handling SSE and AVX instructions. The chip's power budget is shared between cores, but not the physical transistors doing the vector operations.
Maybe stating the blindingly obvious but seems like there is a gap in the market for a board or full kit with a high efficiency ~1-10W CPU and a bunch of SATA and PCIe ports.
Minisforum sort of working on it, I'd imagine the AMD "AI" processors are pretty low power at idle as they're mobile chips. Obviously has the downsides of other minipcs tho (high cost, low expandability)
I think they generally idle at 5-10W, like modern laptop parts do in general.
I don't know how much each of those SATA disks take up, but probably more than a single Raspberry Pi does.
Likewise it has a few case fans in it that may be pointless. I would prefer it never has a heating issue versus saving a few "whurrr" sounds off in a closet somewhere that nobody cares about.
It's also powering that Nvidia 1060 that I do almost nothing with on the NAS. I don't even bother to enable the Jellyfin GPU transcoding configuration because I prefer to keep my library encoded in h264 right now anyhow as I have not yet made the leap to a newer codec because the different smart TVs have varying support. And sometimes my daughter has a friend come over that has a weird Amazon tablet thing that only does a subset of things correctly.
The 1060 isn't an amazing card really, but it could do some basic Ollama inference if I wanted. I think it has 6GB of memory, which is pretty low, but usable for some small LLMs.
If your build allows the extra money for an LSI or real raid controller is well worth it. The no-name PCI-e sata cards are flakey and very slow. Putting an LSI in my NAS was a literal 10x performance boost, particularly with zfs which tends to have all of the drives active at once.
I get if I was wiring up NVME drives that are going to go 500MB/s and higher all the time.
What I really care about with the SATA and what I mean by flaky is I when I have to reboot a system physically every day because the controller stays on in some way even if it gets a soft `reboot` command and then Linux fills up with IO timeouts because the controller seems to stop working after X amount of time.
Keep in mind if you get a real RAID controller and want to use ZFS, you probably want to ensure it can present the disks to the OS as JBOD. Sometimes this requires flashing a non-RAID firmware.
That said, slightly older LSI cards are quite cheap on eBay, and the two I've bought have worked perfectly for many years.
But at home things are different. While I want to use as little power as possible, the realistic plan for being sustainable at home is to use solar with batteries. That's a plan I actually think can matter and that I am able to participate in for relatively low cost ($10k)
Messing around with a system to save a few watts for me in this context isn't very valuable.
As far as solar goes, where I live it's not worth getting, so I can't comment there. Raspberry Pi costs $5 a year to run. A 60 Watt computer costs $79 a year to run. I agree that if it's very difficult to setup a Pi it might be worth the extra cost for the computer, but the energy savings are pretty incredible.
Used server-grade LSI cards seem to be the way to go. Too bad they're power hungry based on what I've read.
On the other hand, they still have the destructive backspace behaviour, and inefficient recursive implementation, that breaks the boot loader spinners that the NetBSD and other boot loaders display. It's a tiny thing, but if one is used to the boot sequence the absence of a spinner makes the experience ever so slightly jarring.
* https://github.com/NumberOneGit/edk2/blob/master/MdeModulePk...
* https://github.com/tianocore/edk2/blob/master/MdeModulePkg/U...
The TianoCore's unnecessarily recursive implementation of a destructive BS is slow enough, on a Pi 4, and in combination with how the boot loaders themselves emitted their spinners, that I could just, very occasionally, see parts of spinner characters flashing very briefly on the screen when the frame refresh timing was just right; which led me to look into what was going on.
It was a fun project and looked cool but never really worked that well. It was quite unstable and drives seemed to disconnect and reconnect a lot. There are probably better quality connectors out there but I think for a NAS you really want proper SATA connections.
I eventually built my own box and went with OMV again. I like it because it's just userland software you install on Debian. Some of the commenters here who think TrueNAS is overkill might want to check out OMV if they haven't already.
To be honest I still only have a few TB of storage on it, probably not really enough to be worth all the hassle of building and configuring a PC, but it was more about the journey which was fun.
I have two "normal" NAS devices, but I would like to find a compact N100 board with multiple SATA ports, (like some of the stackable HATs for the Pi, some of which will take 4 disks directly on the PCB) to put some older discarded drives to good use.
My go-to solution software-wise is actually to install Proxmox, set up ZFS on it and then drop in a lightweight LXC that exposes the local filesystem via SMB, because I like to tweak the "recycle bin" option and some Mac-specific flags--I've been using that setup for a while, also off Proxmox: https://taoofmac.com/space/notes/2024/11/09/1940#setting-up-...
Over time and lots of reading random information sources, I got notes about disabling several settings that are cool for datacenter-levels of storage, but useless for the kinds of "Raspberry-pi tied with a couple USB disks" that I was interested in.
There are also some decent 6 and 8 bay mini-itx cases.
I went looking for completed systems recently and couldn’t find any integrators that make them. Surprised nobody will take a few hundred bucks to plug in all the motherboard power/reset pin headers for me.
The mini itx boards are also interesting, but more if you're actually building the machine yourself. I'm typing this from a 16 core AMD one.
Based on a Marvell/Armada CN9130 SoC which supports ECC, it has mainline Linux support, and public-but-non-upstream code for uboot. With local serial console and a bit of effort, the QNAP OS can be replaced by Arm Debian/Devuan with ZFS.
Rare combo of low power, small size, fast network, ECC memory and upstream-friendly Linux. QNAP also sell a 10GbE router based on the same SoC, which is a successor to the Armada 388 in Helios4 NAS (RIP), https://kobol.io/helios4/
No UEFI support, so TrueNAS for Arm won't work out of the box.
The audience/needs for TruNAS are probably looking to not have to do much beyond either turning it on or plugging in an update stick.
So I just went "raw" smbd and never looked back, but then again I've been running Samba for almost two decades now and configuring it is almost second nature to me (I started doing it as an alternative to Novell/IPX, and later to replace AppleTalk...)
In practice, I've found that worked well because I very seldom had to do any changes to a file server once it was set up (adding shares was mostly a matter of cloning configs, and what minimal user setup there needed to be done in the enterprise was deferred to SAMBA/CIFS glue).
But that's how 99% of the services I run are set up, just a bunch of configuration files managed by Ansible.
The only servers I run in production with a UI are two NASes at home, and the only reason for that is I want to run off the shelf units so I never have to think about them, since my wife and kids rely on them for Jellyfin and file storage.
I really wish I just used something else or raw Ubuntu server.
The Time Machine backup feature corrupts itself.
You can’t have home assistant and Time Machine backups on at the same time. It just feels like a janky UI that has no polish too.
You also need a dedicated OS Drive for TrueNAS, which is reasonable in principle for critical systems, but doesn't always really meet the needs of home users with limited drive bays.
They moved away from Kubernetes in favor of Docker Compose. You can add your own custom compose script, or use ones from the "app store".
This is for the Scale variant of course, no Docker on Core.
Probably not a great idea given how ZFS is architected for memory utilization and ECC. :-)
With Thunderbolt 4 M.2 NVMe enclosure, you can plug in M.2 SATA adapter to connect 8 SATA III drives. Little Pico PSU to power the HDDs and it makes a really low powered NAS.
My plan is to give TrueNAS a spin with two drives and if it is stable, move everything into it.
GTFO, as you re-key your installations this fall with Microsoft's permission. =3
But most manufacturers try to lock the firmware down, and users only get a small subset of configuration menus. For example, the Gigabyte rtx based laptops require patching a machine specific bios to even gain access to the oem firmware areas.
Mostly the modern builds just created a bunch a problems nobody wanted, and didn't improve anything as Asus, Gigabyte, and Razer recently showed.
https://www.youtube.com/watch?v=4er6kD-pxZs
https://www.youtube.com/watch?v=KhfqhCxqpQ8
If you are running signed code on many machines. YMMV... Raspberry Pi avoided the signed code features built into most Broadcom ARM chips for good reasons. =3
It's a rare case when jack (RPi) of all trades is not at all better than master of one or even other jacks of all trades. Running anything but official distro is pain. Managing official distro is pain. Even you it wasn't it doesn't have enough raw power or I/O to do anything really useful.
It's an amazing "temporary" solution because it's so awful that you will actually replace it with a proper one.