Control runtime of addonsπ
Addons are run by the joex component as background tasks in an external process. Depending on the machine it is running on, the addon can be run
- inside a docker container
- inside a systemd-nspawn container
- directly on the machine
Addons can be provided as source packages, where the final program may need to be built. They also can depend on other software. In order to not prepare for each addon, it is recommended to install nix with flakes and docker on the machine running joex.
Please also look at addon section in the default configuration for joex.
You need to explicitly enable addons in the restserver config file.
Docspell uses "runners" to execute an addon. This includes building it if necessary. The following runner exist:
docker
: uses docker to build an run the addonnix-flake
: builds vianix build
and runs the executable in$out/bin
trivial
: simply executes a file inside the addon (as specified in the descriptor)
In the joex configuration you can specify which runners your system supports.
Prepare for running addonsπ
Depending on how you want addons to be run, you need to install either
docker and/or systemd-nspawn on the machine running joex.
Additionally, the user running joex must be able to use these tools.
For docker it usually means to add the user to some group. For
systemd-nspawn you most likely want to configure sudo
to run
passwordless the systemd-nspawn
command.
Without this, an addon can only be run "directly" on the machine that hosts joex (which might be perfectly fine). The addon then "sees" all files on the machine and could potentially do harm.
It is recommended to install nix
and docker
, if possible. Addons
may only run with docker or only without, so supporting both leaves
more options.
Prepare for building addonsπ
Addons can be packaged as source or binary packages. For the former, joex will build the addon first. There are two supported ways to do so:
- via
docker build
when the addons provides aDockerfile
(use runnerdocker
) - via
nix build
when the addon provides aflake.nix
file (use runnernix-flake
)
Both build strategies will cache the resulting artifact, so subsequent builds will be (almost) no-ops.
Building addons requires to be connected to the internet! Running them may not require a network connection.
If the addon is packaged as a binary, then usually the trivial
runner (possibly in combination with systemd-nspawn
) can be used.
Runtimeπ
Cache directoryπ
Addons can use a "cache directory" to store data between runs. This directory is not cleaned by docspell. If you have concerns about space, use a cron job or systemd-timer to periodically clean this directory.
"Pure" vs "Impure"π
Addons can talk back to Docspell in these ways: they can use the http api, for example with dsc, or they can return data to instruct Docspell to apply changes.
The former requires the addon to be connected to the network to reach the Docspell restserver. This allows the addon to do arbitrary changes at any time - this is the "impure" variant.
The second approach can be run without network connectivity. When using docker or systemd-nspawn, Docspell will run these addons without any network. Thus they can't do anything really, except return data back to Docspell.
The pure way is much preferred! It allows for more consistent behaviour, because Docspell is in charge for applying any changes. Docspell can apply changes only if the addon returned successfully. Addons can also be retried on error, because no changes happened yet.
It's the decision of the addon author, how the addon will work. It
should document whether it is pure or impure. You can also look into
the descriptor and check for a networking: false
setting. As the
server administrator, you can configure Docspell to only accept pure
addons.
Runnersπ
nix flake runnerπ
For addons providing a flake.nix
this runner can build it and find
the file to execute. With this flake.nix
file addons can declare how
they should be build and what dependencies are required to run them.
The resulting executable can be executed via systemd-nspawn
in a
restricted environment or directly on the machine.
dockerπ
Addons can provide a Dockerfile or an image. If no image is given,
docker build
will be run to build an image from the Dockerfile
.
Then docker run
is used to run the addon.
You need to install docker
to use this runner.
trivialπ
Addons can simply declare a file to execute. Docspell can use
systemd-nspawn
to run it in an restricted environment, or it can be
run directly on the machine. This variant is only useful for very
simple addons, that don't require any special dependencies.
You need to check each addon for its requirements and prepare the machine accordingly.
Choosing runnersπ
The config addons.executor-config.runners
accepts a list of runners.
It specifies the preferred runner first. If an addon can be executed
via docker and nix, Docspell will choose the runner first in the list.
If you don't have nix installed, remove the nix-flake
runner from
this list and same for docker, of course.
systemd-nspawnπ
The systemd-nspawn
can be used to run programs in a lightweight
ad-hoc container. It is available on most linux distributions (it is
part of systemdβ¦). It doesn't require an image to exist first; this
makes it very convenient for running addons in a restricted
environment.
If you enable it in the config file, then all addons are either run
via systemd-nspawn
or docker - and thus always in a restricted
environment, where they can only access their own files and the files
provided by Docspell.
The downside is that systemd-nspawn
needs to be run as root (as far
as I know). Therfore, configure sudo
to allow the user that is
running joex to execute systemd-nspawn
non-interactively.
Install systemd-nspawn
and enable the user running joex to use it
password-less via sudo.
Within Dockerπ
If joex itself is run as a docker container, things get a bit
complicated. The default image for joex does not contain nix
, so the
nix-flake
runner cannot be used out of the box.
In order to use the docker
runner, the container must be configured
to access the hosts docker daemon. On most systems this can be
achieved by bind-mounting the unix socket (usually at
/var/run/docker.sock
) into the container. Here is a snippet from the
provided docker-compose
file:
joex:
image: docspell/joex:latest
# ... left out for brevity
volumes:
- /var/run/docker.sock:/var/run/docker.sock
- /tmp:/tmp
Additionally to /var/run/docker.sock
, it also bind mounts the /tmp
directory. This is necessary, because docker will be invoked with bind
mounts from inside the continer - but these must be available on the
host, because the docker client in the container actually runs the
command on the host.
The addon executor uses the systems temp-directory (which is usually
/tmp
) as a base for creating a working and cache directory. Should
you change this in joex config file (or your system uses a different
default temp-dir), then the bind mount must be adapted as well.
Another variant is to extend the default joex image and add more
programs as needed by addons and then use the trivial
runner.
Summary / tl;drπ
When joex is not inside a container:
- (optional) Install
systemd-nspawn
- it is provided on many GNU/Linux distributions - Configure
sudo
to allow the user running the joex component to executesystemd-nspawn
non-interactively (without requiring a password) - Install docker
- Install nix and enable flakes
- Allow the user who runs the joex component to use docker and nix. If you install nix as multi-user, then this is already done.
- Check the section on addons in the default configuration for joex