Starting Servers with `reStart`

When developing, it's very convenient to use the revolver sbt plugin. Start the sbt console and then run:

sbt:docspell-root> restserver/reStart

This starts a REST server. Once this started up, type:

sbt:docspell-root> joex/reStart

if also a joex component is required. Prefixing the commads with ~, results in recompile+restart once a source file is modified.

It is possible to start both in the root project:

sbt:docspell-root> reStart

Custom config file

The sbt build is setup such that a file dev.conf in the directory local (at root of the source tree) is picked up as config file, if it exists. So you can create a custom config file for development. For example, a custom database for development may be setup this way:

#jdbcurl = "jdbc:h2:///home/dev/workspace/projects/docspell/local/docspell-demo.db;MODE=PostgreSQL;DATABASE_TO_LOWER=TRUE;AUTO_SERVER=TRUE"
jdbcurl = "jdbc:postgresql://localhost:5432/docspelldev"
#jdbcurl = "jdbc:mariadb://localhost:3306/docspelldev"

docspell.server {
  backend {
    jdbc {
      url = ${jdbcurl}
      user = "dev"
      password = "dev"
    }
  }
}

docspell.joex {
  jdbc {
    url = ${jdbcurl}
    user = "dev"
    password = "dev"
  }
  scheduler {
    pool-size = 1
  }
}

Developing Frontend

The frontend is a SPA written in Elm. The UI framework in use is tailwind.

The frontend code is in the sub-project webapp. Running sbt's compile task, compiles elm sources and creates the final CSS file. Whenever the restserver module is build by sbt, the webapp sub-project is built as well and the final files to deliver are updated. So, when in sbt shell, "watch-compile" the project restserver, (via ~ restserver/compile), re-compiles elm-code on change. However, it also re-creates the final css, which is a rather long task.

To speed things up when only developing the frontend, a bash script is provided in project/dev-ui-build.sh. Start the restserver once, using restserver/reStart task as described above. Then run this script in the source root. It will watch elm files and the css file and re-compiles only on change writing the resulting files in the correct locations so they get picked up by the restserver.

Now you can edit elm files and the index.css and then only refresh the page. Elm compilation is very fast, it's difficult to reach the refresh button before it is done compiling :). When editing the CSS, it takes a little longer, but this is hardly necessary, thanks to tailwind.

There is still a problem: the browser caches the js and css files by default, so a page refresh is not enough, you need to clear the cache, too. To avoid this annoyance, set a env variable DOCSPELL_ENV to the value dev. Docspell then adds a response header, preventing the browser to cache these files. This must be done, obviously, before starting the restserver:

$ export DOCSPELL_ENV=dev
$ sbt "restserver/reStart"

Nix Expressions

The directory /nix contains nix expressions to install docspell via the nix package manager and to integrate it into NixOS.

Testing NixOS Modules

The modules can be build by building the configuration-test.nix file together with some nixpkgs version. For example:

nixos-rebuild build-vm -I nixos-config=./configuration-test.nix \
  -I nixpkgs=https://github.com/NixOS/nixpkgs-channels/archive/nixos-19.09.tar.gz

This will build all modules imported in configuration-test.nix and create a virtual machine containing the system. After that completes, the system configuration can be found behind the ./result/system symlink. So it is possible to look at the generated systemd config for example:

cat result/system/etc/systemd/system/docspell-joex.service

And with some more commands (there probably is an easier way…) the config file can be checked:

cat result/system/etc/systemd/system/docspell-joex.service | grep ExecStart | cut -d'=' -f2 | xargs cat | tail -n1 | awk '{print $NF}'| sed 's/.$//' | xargs cat | jq

To see the module in action, the vm can be started (the first line sets more memory for the vm):

export QEMU_OPTS="-m 2048"
export QEMU_NET_OPTS "hostfwd=tcp::7880-:7880"
./result/bin/run-docspelltest-vm

Release

The CI and making a release is done via github actions. The workflow is roughly like this:

each PR is only merged if the sbt ci task returns successfully. This is ensured by the ci.yml workflow that triggers on each pull request
each commit to the master branch is also going through sbt ci and then a prerelease is created. The tag nightly is used to point to the latest commit in master. Note, that this is discouraged by git, but github doesn't allow to create a release without a tag. So this tag moves (and is not really a tag then…). After the prerelease is created, the docker images are built and pushed to docker hub into the docspell organization. The docker images are also tagged with nightly at docker hub. This is all done via the realease-nightly.yml workflow.
A stable release is started by pushing a tag with pattern v* to github. This triggers the release.yml workflow which builds the packages and creates a release in draft mode. The sbt ci task is not run, because it is meant to only release commits already in the master branch. After this completes, the release notes need to be added manually and then the release must be published at github. This then triggers the docker-images.yml workflow, which builds the corresponding docker images and pushes them to docker hub. The docker images are tagged with the exact version and the latest tag is moved to the new images. Another manual step is to set the branch current-docs to its new state and push it to github. This will trigger a build+publish of the website.
Publishing the website happens automatically on each push to the branch current-docs. Changes to the current website must be based on this branch.

Some notes: I wanted a 2/3 step process when doing a stable release, to be able to add release notes manually (I don't want this to be automated right now) and to do some testing with the packages before publishing the release. However, for the nightly releases, this doesn't matter - everything must be automated here obviously. I also wanted the docker images to be built from the exact same artifacts that have been released at github (in contrast to being built again).

Background Info

There is a list of ADRs containing internal/background info for various topics.

6.8 KiB Raw Blame History

Starting Servers with reStart