Suckless/Simple Continous Integration

Jenkins, Travis, GitHub Actions, GitLab CI. The list goes on. This is a minimal tool for fulfilling the CI (Continous Integration) use case.

Brainstorm

If you dont want to congest your CI server. Too bad. Write faster ci suites. (TODO: implement runners)

I would like to try to avoid writing a million REST APIs, as that just results in bloat usually.

Could this also be used as a systest replacement? The value of systest is the drivers and the configuration files. Not really the controllers themselves. But yes, this system could replace the systest runner scripts. But not the drivers or the configuration files themselves...

Using

/etc/sci/conf.d/pipelines.conf

aaltitoad-release   https://github.com/sillydan1/aaltitoad      onpush-main         /home/sci/pipelines/cpp-dev-git.sh
aaltitoad-develop   https://github.com/sillydan1/aaltitoad      onpush-dev          /home/sci/pipelines/cpp-rel-git.sh
schoolnotes         https://gitlab.com/sillydan/schoolnotes     manual-name         "echo hello"
horse               https://gtz.dk/horse"                       manual-name         "curl -X POST https://example.com"
cool-horse          https://gtz.dk/horse"                       manual-name         "docker run --rm -v $SCI_PIPELINE_DIR:$SCI_PIPELINE_DIR -v /home/sci/pipelines:/pipelines alpine /pipelines/your-script.sh"
rad-horse           https://gtz.dk/horse"                       manual-name         "/home/sci/pipelines/dockerized-horse.sh"
<name>              <url>                                       <trigger-filename>  <pipeline-command>

Triggers are just bare files that reside in /etc/sci/triggers/ - this means that multiple pipelines can use the same trigger. When a trigger file is being written to or touched, this triggers the associated pipeline. Trigger files are automatically created by the scid daemon during startup or reconfiguring.

Any webhooks should be managed by shim-APIs / adapter APIs that are very slim (possibly flask) REST APIs that just listen for webhook event triggers and does a simple touch <associated file>. It could even read the same configuration file as scid uses to autogenerate these webhooks. But it is explicity NOT part of the core system, as that would be unneeded bloat.

How do I get information about a running pipeline? Perhaps a /var/tmp/sci/<pipeline-name> is continually updated with runtime information? Then you can cat, or tail -f it to get the information.

scid manages running subprocesses, so it must be implemented in a language that handles that nicely.

Pipelines - Should pipelines have a custom language? NO! God already gave us shell scripts! Why would we reinvent the wheel? Take a page out of Jenkins' book. scid should just call some command on trigger. That command could / should usually be a quick shell script (these calls are executed in the dispatched separate thread, so dont worry about sleeping or whatever you need to do). These scripts are equivalent to the Jenkins pipelines, and should update their current state in /var/tmp/scid/$PIPELINE_NAME by simply writing to the file (when implementing runners, we probably need to rsync these files automagically). Oh, and dont worry about cleaning up. Each pipeline execution will be placed in a /tmp/<pipeline-name>-<uuid>/ directory and will be executed by a gimped user. If you need to interact with the linux stuff and it might break other pipelines or whatever, then your pipeline-command should just start a docker container, volume mount the tmpdir and what else you need and run your command in that.

Need multistep pipelines? Then write a multistep shell script! You're the boss of your own CI server man.

#!/bin/bash
# cpp-dev-git-pipeline,sh
# NOTE: Will be executed from /tmp/$SCI_PIPELINE_NAME-<uuid>/
git clone -b dev "$SCI_PIPELINE_URL" "$SCI_PIPELINE_NAME"
echo "clone success" > "/var/tmp/sci/$SCI_PIPELINE_NAME" # You can have any kind of data in the pipeline status file.
cd "$SCI_PIPELINE_NAME"
cmake -B build
cmake --build build

Secrets

Not planned yet. But they could be as simple as a keystore.

Development

What language should I implement scid in?

• c slow devtime, decent tooling, very minimal and forces you to be minimal also. Maybe
• c++ slow devtime, good tooling, very familiar. Can easily get out of hand. Maybe
• python quick, easy, slow performance, terrible multiprocessing capabilities. Maybe
• java slow devtime, bloated. No.
• rust slow devtime (not familiar), great tooling, too many risks for a first prototype. No.
• shell quick, easy, instant spaghetti. No.

I choose c!

I also choose Makefiles! - Just to force myself to use another build system than CMake! If you want compile_commands.json files, you should use bear as it works well

Progress

• Zeroth things first, let's create a simple CLI application with --verbosity VAL and --help options.
• First things first, let's implement something that reacts when some provided file changes (not poll please).
• Second things second, implement a simple logging system with differing levels of verbosity and configurable output file using cli options.
• Third things third, implement a thing that simultaneously watches two different files (multithreading). it should be cancellable with ctrl+c, but it should just contiuously print event notifications.
• Fourth things fourth, implement a prototype that reads a space-separated file and populates a struct.
• Fifth things fifth, implement a prototype that spawns a new thread that executes a shell command.
• Sixth things sixth, daemonize it!
• Seventh things seventh, package the sucker (arch, debian, alpine, docker)
• Eight things eight, try it out! - maybe even write the python webhook extension.
• Ninth things ninth, fix bugs, see below
• Tenth things tenth, write manpages
• Eleventh things last, release!

Bugs

• command output is being inherited. It should be piped into some random log-file
• pretty sure that ctrl+c / SIGINT is not graceful yet.
• missing license
• I am deliberately not using Restart=on-failure in the scid.service file because we are using Type=exec and not Type=notify (yet) - which would require a sd_notify call of READY=1 (see man systemd.service)

Note Regarding inotify usage

From the manpage:

With careful programming, an application can use inotify to efficiently monitor and cache the state of a set of
filesystem  objects.  However, robust applications should allow for the fact that bugs in the monitoring logic or races
of the kind described below may leave the cache inconsistent with the filesystem state.  It is probably wise to do some
consistency checking, and re‐build the cache when inconsistencies are detected.

i.e., we should also poll the watched files every once in a while (maybe once per minute? idk) to ensure that we catch all events.