macOS Ventura, /nix is empty after reboot following clean installation

[zmitchell]

What I did

I ran the command to install:

$ curl --proto '=https' --tlsv1.2 -sSf -L https://install.determinate.systems/nix | sh -s -- install

I ran the installation process (I don't remember the exact steps, I didn't realize there was a failure at this point yet). I sourced whatever script it mentioned and was able to run the nix command. That was yesterday.

Symptoms

Today, likely after opening a new terminal, I'm no longer able to run the nix command. I looked in my .zshrc and there's nothing sourcing anything having to do with nix. Further investigation shows that there's nothing in /etc/nix besides nix.conf.

I downloaded the installer directly to see if I could use it to repair my installation:

$ curl -sL -o nix-installer https://install.determinate.systems/nix/nix-installer-aarch64-darwin

then I ran nix-installer plan and immediately ran into this error:

$ ./nix-installer plan
Error:
   0: Planner error
   1: Error executing action
   2: Path exists `/Library/LaunchDaemons/org.nixos.darwin-store.plist`

Location:
   /rustc/69f9c33d71c871fc16ac445211281c6e7a340943/library/core/src/convert/mod.rs:726

Backtrace omitted. Run with RUST_BACKTRACE=1 environment variable to display it.
Run with RUST_BACKTRACE=full to include source snippets.

This tells me that the installation at least put some files in the right place.

Ok, what if I run the uninstall command?

$ ./nix-installer uninstall
`nix-installer` needs to run as `root`, attempting to escalate now via `sudo`...
Error:
   0: Reading receipt
   1: No such file or directory (os error 2)

Location:
   src/cli/subcommand/uninstall.rs:126

Backtrace omitted. Run with RUST_BACKTRACE=1 environment variable to display it.
Run with RUST_BACKTRACE=full to include source snippets.

Sure enough, there's no /nix/receipt.json (makes sense, I said earlier that /etc/nix only contains receipt.json). I thought this was supposed to be created during the installation process?

Questions

• How do I repair the installation?
• How do I verify that the installation is complete?
• What went wrong?

[zmitchell]

I dug through my shell history to find out which file I sourced during the installation procedure:

$ . /nix/var/nix/profiles/default/etc/profile.d/nix-daemon.sh

What's interesting is that /nix/var no longer exists for whatever reason. I certainly never typed in a command to delete the directory, so maybe it was removed by the daemon script or it was only there temporarily?

[zmitchell]

I just went through the uninstallation procedure I found here: https://iohk.zendesk.com/hc/en-us/articles/4415830650265-Uninstall-nix-on-MacOS

Then I tried to install again. What I noticed is that if I tried to run the installer without root, it proceeded and ran into an error, only mentioning that it needed root along the way:

❯ ./nix-installer install darwin-multi
`nix-installer` needs to run as `root`, attempting to escalate now via `sudo`...
Nix install plan (v0.1.0)

Planner: darwin-multi

Planner settings:

* nix_build_group_name: "nixbld"
* volume_encrypt: false
* volume_label: "Nix Store"
* root_disk: null
* case_sensitive: false
* nix_build_user_id_base: 300
* force: false
* extra_conf: []
* nix_package_url: "https://releases.nixos.org/nix/nix-2.12.0/nix-2.12.0-aarch64-darwin.tar.xz"
* nix_build_group_id: 3000
* nix_build_user_count: 32
* channels: ["nixpkgs=https://nixos.org/channels/nixpkgs-unstable"]
* modify_profile: true
* nix_build_user_prefix: "_nixbld"

The following actions will be taken:

* Create an APFS volume `Nix Store` for Nix on `disk3`
* Fetch `https://releases.nixos.org/nix/nix-2.12.0/nix-2.12.0-aarch64-darwin.tar.xz` to `/nix/temp-install-dir`
* Create build users (UID 300-332) and group (GID 3000)
* Create a directory tree in `/nix`
* Move the downloaded Nix into `/nix`
* Setup the default Nix profile
* Configure Nix daemon related settings with systemd
* Place the Nix configuration in `/etc/nix/nix.conf`
* Place channel configuration at `/Users/zmitchell/.nix-channels`
* Configure the shell profiles
* Kickstart the launchctl unit `system/org.nixos.nix-daemon`


Proceed? ([Y]es/[n]o/[e]xplain):

I cleaned up any files that this left behind after hitting the error, deleted the /nix volume, rebooted, and tried the installer again with root:

$ sudo ./nix-installer install darwin-multi

This installation succeeded. Afterwards I was able to run nix commands, there was a receipt.json in /nix. All seemed well. Then I rebooted just to see if it would stick. No dice, we're back to where I started where everything in /nix is missing.

So at least I can reproduce the issue now?

[zmitchell]

More developments! I uninstalled, then ran the official installer and the contents of /nix survived the reboot. Next step is to try running nix-installer again without sudo to double check that sudo wasn't causing some permissions issue with mounting the contents of /nix.

[zmitchell]

Uninstalled and then ran nix-installer without sudo and I was prompted to enter my password, which I did. That's what we expect, so that's good. I took a look at the fstab with sudo vifs and noticed that the official installer references the volume by UUID whereas nix-installer references the volume by NAME. Not sure if that's an issue, but I've written down the UUID as seen in Disk Utility so if /nix is empty I can potentially try referencing via UUID instead of name to see if that will mount it properly.

[zmitchell]

That was it! When I rebooted /nix was again empty. I edited the fstab with sudo vifs and replaced NAME="Nix Store" with UUID=<my UUID>, rebooted, and everything seems to be working.

[elgertam]

I also have experienced this issue. Haven't tried @zmitchell's proposed fix yet, though I'll probably reset my system and do so.

[woodwardjd]

During our brand new M2 (max, MacBook Pro 16) configurations this weekend we decided to give this a try instead of the official install. Ran into the same symptom (/nix empty on reboot). The fstab edit to use the UUID instead of the name worked for us.

Thanks @zmitchell for the thorough report AND reporting the workaround! And thanks to the team behind this tool for the great docs and simple tool. It makes onboarding folks into a dev ecosystem that uses nix a ton easier.

[elgertam]

Can confirm that using vifs to update the entry in /etc/fstab to UUID=<disk UUID> rather than NAME="Nix Store" solved my problem. Nix is now running & loading properly after reboot.

[Hoverbear]

@zmitchell Thanks for reporting this! It seems you found a bug and found the fix!

I'll work on reflecting this fix in the installer and cut a new version this week!

I'm sorry about the issues! I'm not a Mac user myself and while we did test rebooting on the remote ephemeral Macs I suspect they are a bit different than a "real" user machine.

[zmitchell]

No worries, it was fun to debug it :) I'm learning Nix and can volunteer to test things on macOS if you need a real machine to test on in the future.

[Hoverbear]

Out of curiousity, does your system have any other volumes on it called Nix Store?

[woodwardjd]

@Hoverbear mine may or may not have (I don't think it did). My partner's definitely did not.

edit: it definitely didn't. My install failed when I did have that existing volume. I disconnected it before getting the successful install, and I haven't reattached it since (I only had it originally connected to run Migration Assistant from its contents).

[zmitchell]

I didn't. This was my first time installing Nix and while I was debugging I deleted the Nix Store volume during each uninstall.

[Hoverbear]

Okay! Thanks! We had a report from someone else where they'd previously installed Nix and experienced this, I just wanted to see if it was isolated to that particular case.

[Hoverbear]

Okay, #215 should apply this workaround! I've tested it a few times on our ephemeral Macs and they seem quite content with themselves.

If you'd be willing to be a guinea pig for me:

Because you changed the /etc/fstab line it may be safe to revert to starting with NAME="Nix Store" before doing this! I don't think we're that clever quite yet.

/nix/nix-installer uninstall

Then:

curl --proto '=https' --tlsv1.2 -sSf -L https://install.determinate.systems/nix/pr/215 | sh -s -- install

I hope this works. :)

[woodwardjd]

Close! fstab shows /nix as the mount point. After Install+Reboot the volume gets mounted at /Volumes/Nix Store anyway. I removed the quotes in fstab's UUID="xyz-123-...", resulting in UUID=xyz-123..., and after a reboot it automounts at /nix and it's back to smooth sailing.

[Hoverbear]

Ah heck! Ok! Thanks! We'll make that change and a release should happen in the next day or two.