The Nix-provided `nix-daemon.socket` file has a
> ConditionPathIsReadWrite=/nix/var/nix/daemon-socket/socket
line, to skip that unit if /nix/var/nix/daemon-socket/socket is
read-only (which is the case in some nixos-containers with that folder
bind-ro-mounted from the host).
In these cases, the unit was skipped.
Systemd 250 (rightfully) started to also skip in these cases:
> [ 237.187747] systemd[1]: Nix Daemon Socket was skipped because of a failed condition check (ConditionPathIsReadWrite=/nix/var/nix/daemon-socket).
However, systemd < 250 didn't skip if /nix/var/nix/daemon-socket/socket
didn't /exist at all/, and we were relying on this bug in the case for
fresh NixOS systems, to have /nix/var/nix/daemon-socket/socket created
initially.
Move the creation of that folder to systemd-tmpfiles, by shipping an
appropriate file in `${nixPackage}/lib/tmpfiles.d/nix-daemon.conf`
(NixOS/nix#6285).
In the meantime, set a systemd tmpfiles rule manually in NixOS.
This has been tested to still work with read-only bind-mounted
/nix/var/nix/daemon-socket/socket in containers, it'll keep them
read-only ;-)
As `nix-daemon.service` does not make use of `ExecStop`, we prefer
to keep the socket up and available. This is important for machines
that run Nix-based services, such as automated build, test, and deploy
services, that expect the daemon socket to be available at all times.
See committed inline comment for further explanation.
The `substituters` option in `nix.settings` uses the order
of the substituters listed to define priority. Prior to https://github.com/NixOS/nixpkgs/pull/139075,
the corresponding option `binaryCaches` is declared in the `nix` namespace,
which is guaranteed to be merged last. However, the order of merging isn't
guaranteed in submodules. This cause definitions to be appended to the default
value instead of prepended, breaking backwards compatibility as reported in https://github.com/NixOS/nixpkgs/issues/158356.
The way this is addressed in the module system is with order priorities via
`mkOrder` and sorting definitions before merging. This PR restores the previous
behavior by setting a higher priority to the substituters option defined internally,
thus all definitions with default priority will be merged before it. This was chosen because
the `mkRenamedOption` function does not preserve order priority so users using legacy options do not have
precise control on placement.
This change should suffice for simple configuration, but further revision to the module system
is needed for to make various `mk*` functions aware of order priorities.
In #139075, mandatoryFeatures was removed from the generated
supportedFeatures, which breaks backward compatibility and is
different from what the description of supportedFeatures says.
The `nix.*` options, apart from options for setting up the
daemon itself, currently provide a lot of setting mappings
for the Nix daemon configuration. The scope of the mapping yields
convience, but the line where an option is considered essential
is blurry. For instance, the `extra-sandbox-paths` mapping is
provided without its primary consumer, and the corresponding
`sandbox-paths` option is also not mapped.
The current system increases the maintenance burden as maintainers have to
closely follow upstream changes. In this case, there are two state versions
of Nix which have to be maintained collectively, with different options
avaliable.
This commit aims to following the standard outlined in RFC 42[1] to
implement a structural setting pattern. The Nix configuration is encoded
at its core as key-value pairs which maps nicely to attribute sets, making
it feasible to express in the Nix language itself. Some existing options are
kept such as `buildMachines` and `registry` which present a simplified interface
to managing the respective settings. The interface is exposed as `nix.settings`.
Legacy configurations are mapped to their corresponding options under `nix.settings`
for backwards compatibility.
Various options settings in other nixos modules and relevant tests have been
updated to use structural setting for consistency.
The generation and validation of the configration file has been modified to
use `writeTextFile` instead of `runCommand` for clarity. Note that validation
is now mandatory as strict checking of options has been pushed down to the
derivation level due to freeformType consuming unmatched options. Furthermore,
validation can not occur when cross-compiling due to current limitations.
A new option `publicHostKey` was added to the `buildMachines`
submodule corresponding to the base64 encoded public host key settings
exposed in the builder syntax. The build machine generation was subsequently
rewritten to use `concatStringsSep` for better performance by grouping
concatenations.
[1] - https://github.com/NixOS/rfcs/blob/master/rfcs/0042-config-option.md
The nix.daemonNiceLevel options allows for setting the nice level of the
Nix daemon process. On a modern Linux kernel with group scheduling the
nice level only affects threads relative to other threads in the same
task group (see sched(7)). Therefore this option has not the effect one
might expect.
The options daemonCPUSchedPolicy and daemonIOSchedClass are introduced
and the daemonIONiceLevel option renamed to daemonIOSchedPrority for
consistency. These options allow for more effective control over CPU
and I/O scheduling.
Instead of setting daemonNiceLevel to a high value to increase the
responsiveness of an interactive system during builds -- which would not
have the desired effect, as described above -- one could set both
daemonCPUSchedPolicy and daemonIOSchedClass to idle.
Don't worry, it's is true by default. But I think this is important to
have because NixOS indeed shouldn't need Nix at run time when the
installation is not being modified, and now we can verify that.
NixOS images that cannot "self-modify" are a legitamate
use-case that this supports more minimally. One should be able to e.g. do a
sshfs mount and use `nixos-install` to modify them remotely, or just
discard them and build fresh ones if they are run VMs or something.
The next step would be to make generations optional, allowing just
baking `/etc` and friends rather than using activation scripts. But
that's more involved so I'm leaving it out.
Before this commit, the `flake` option was typed with `types.unspecified`.
This type get's merged via [`mergeDefaultOption`](ebb592a04c/lib/options.nix (L119-L128)), which has a line
```nix
else if all isFunction list then x: mergeDefaultOption loc (map (f: f x) list)
```
`lib.isFunction` detects an attrs in the shape of `{__functor = ...}` as
a function and hence this line substitutes such attrs with a function
(f: f x).
If now, a flake input has a `__functor` as it's output, this will
coerce the once attrs to a function. This breaks a lot of things later
in the stack, for example a later `lib.filterAttrs seive <LAMBDA>` will
fail for obious reasons.
According to @infinisil, `types.unspecified` is due to deprecation. In
the meantime this PR provides a specific fix for the specific problem
discovered.
In newer versions of Nix (at least on 2.4pre20201102_550e11f) the
`extra-` prefix for config options received a special meaning and the
option `extra-sandbox-paths` isn't recognized anymore. This commit fixes
it.
It doesn't cause a behavior change when using older versions of Nix but
does cause an extra newline to appear in the config, thus changing the
hash.
As the only consequence of isSystemUser is that if the uid is null then
it's allocated below 500, if a user has uid = something below 500 then
we don't require isSystemUser to be set.
Motivation: https://github.com/NixOS/nixpkgs/issues/112647
Commit 5395397f removed the assertions from the buildMachines to
ensure that either system or systems is set for each buildmachine.
This patch re-implements those assertions.
The symptom is that if both system and systems are omitted, then the
/etc/machines file has the wrong number of columns and any attempt to
run a `nix` operation that has to perform a build will fail with a
`strtoull` exception.
The `platform` field is pointless nesting: it's just stuff that happens
to be defined together, and that should be an implementation detail.
This instead makes `linux-kernel` and `gcc` top level fields in platform
configs. They join `rustc` there [all are optional], which was put there
and not in `platform` in anticipation of a change like this.
`linux-kernel.arch` in particular also becomes `linuxArch`, to match the
other `*Arch`es.
The next step after is this to combine the *specific* machines from
`lib.systems.platforms` with `lib.systems.examples`, keeping just the
"multiplatform" ones for defaulting.
This reverts commit f19b7b03a0, reversing
changes made to 572a864d02.
Sorry. I pushed the wrong staging-next (the one that had my master
merged in). This was not intended.
This reverts commits 9d0de0dc57,
27d2857a99. 'nix ping-store' is an
experimental command so it doesn't work in Nix 2.4 unless you set
'experimental-features = nix-command' in nix.conf.
Instead of making the configuration less portable by hard coding the number of
jobs equal to the cores we can also let nix set the same number at runtime.
This allows you to specify the system-wide flake registry. One use is
to pin 'nixpkgs' to the Nixpkgs version used to build the system:
nix.registry.nixpkgs.flake = nixpkgs;
where 'nixpkgs' is a flake input. This ensures that commands like
$ nix run nixpkgs#hello
pull in a minimum of additional store paths.
You can also use this to redirect flakes, e.g.
nix.registry.nixpkgs.to = {
type = "github";
owner = "my-org";
repo = "my-nixpkgs";
};
A centralized list for these renames is not good because:
- It breaks disabledModules for modules that have a rename defined
- Adding/removing renames for a module means having to find them in the
central file
- Merge conflicts due to multiple people editing the central file
Scott Olson