[Laminar](http://laminar.ohwg.net) is a lightweight and modular Continuous Integration service for Linux. It is self-hosted and developer-friendly, eschewing a configuration web UI in favor of simple version-controllable configuration files and scripts.
Laminar encourages the use of existing GNU/Linux tools such as `bash` and `cron` instead of reinventing them.
Although the status and progress front-end is very user-friendly, administering a Laminar instance requires writing shell scripts and manually editing configuration files. That being said, there is nothing esoteric here and the tutorial below should be straightforward for anyone with even very basic Linux server administration experience.
Throughout this document, the fixed base path `/var/lib/laminar` is used. This is the default path and can be changed by setting `LAMINAR_HOME` in `/etc/laminar.conf` as desired.
## Terminology
- *job*: a task, identified by a name, comprising of one or more executable scripts.
You can build an image that runs `laminard` by default, and contains `laminarc` for use based on `alpine:edge` using the `Dockerfile` in the `docker/` directory.
Keep in mind that this is meant to be used as a base image to build from, so it contains only the minimum packages required to run laminar. The only shell available by default is sh (so scripts with `#!/bin/bash` will fail to execute) and it does not have `ssh` or `git`. You can use this image to run a basic build server, but it is recommended that you build a custom image from this base to better suit your needs.
The [`-v` flag](https://docs.docker.com/storage/volumes/#choose-the--v-or---mount-flag) is necessary to persist job scripts and artefacts beyond the container lifetime.
The [`--env-file` flag](https://docs.docker.com/engine/reference/commandline/run/#set-environment-variables--e---env---env-file) is necessary to pass configuration from `laminar.conf` to `laminard` because `laminard` does not read `/etc/laminar.conf` directly but expects variables within to be exported by `systemd` or other process supervisor.
Use `systemctl start laminar` to start the laminar system service and `systemctl enable laminar` to launch it automatically on system boot.
After starting the service, an empty laminar dashboard should be available at http://localhost:8080
Laminar's configuration file may be found at `/etc/laminar.conf`. Laminar will start with reasonable defaults if no configuration can be found.
## Running on a different HTTP port or Unix socket
Edit `/etc/laminar.conf` and change `LAMINAR_BIND_HTTP` to `IPADDR:PORT`, `unix:PATH/TO/SOCKET` or `unix-abstract:SOCKETNAME`. `IPADDR` may be `*` to bind on all interfaces. The default is `*:8080`.
Do not attempt to run laminar on port 80. This requires running as `root`, and Laminar will not drop privileges when executing job scripts! For a more complete integrated solution (including SSL), run laminar behind a regular webserver acting as a reverse proxy.
A reverse proxy is required if you want Laminar to share a port with other web services. It is also recommended to improve performance by serving artefacts directly or providing a caching layer for static assets.
If you use [artefacts](#Archiving-artefacts), note that Laminar is not designed as a file server, and better performance will be achieved by allowing the frontend web server to serve the archive directory directly (e.g. using a `Location` directive).
Laminar uses Server Sent Events to provide a responsive, auto-updating display without polling. Most frontend webservers should handle this without any extra configuration.
If you use a reverse proxy to host Laminar at a subfolder instead of a subdomain root, the `<base href>` needs to be updated to ensure all links point to their proper targets. This can be done by setting `LAMINAR_BASE_URL` in `/etc/laminar.conf`.
See the [reference section](#Service-configuration-file)
---
# Defining a job
To create a job that downloads and compiles [GNU Hello](https://www.gnu.org/software/hello/), create the file `/var/lib/laminar/cfg/jobs/hello.run` with the following content:
Laminar uses your script's exit code to determine whether to mark the run as successful or failed. If your script is written in bash, the [`-e` option](http://tldp.org/LDP/abs/html/options.html) is helpful for this. See also [Exit and Exit Status](http://tldp.org/LDP/abs/html/exit-status.html).
In this case, `laminarc` returns immediately, with its error code indicating whether adding the job to the queue was sucessful. The run number will be printed to standard output.
If the server is busy, a run may wait in the queue for some time. To have `laminarc` instead block until the run leaves the queue and starts executing, use
In this case, `laminarc` blocks until the job starts executing, or returns immediately if queueing failed. The run number will be printed to standard output.
In all cases, a started run means the `/var/lib/laminar/cfg/jobs/hello.run` script will be executed, with a working directory of `/var/lib/laminar/run/hello/1` (or current run number)
This is against the design principles of Laminar and was deliberately excluded. Laminar's web UI is strictly read-only, making it simple to deploy in mixed-permission or public environments without an authentication layer. Furthermore, Laminar tries to encourage ideal continuous integration, where manual triggering is an anti-pattern. Want to make a release? Push a git tag and implement a post-receive hook. Want to re-run a build due to sporadic failure/flaky tests? Fix the tests locally and push a patch. Experience shows that a manual trigger such as a "Build Now" button is often used as a crutch to avoid doing the correct thing, negatively impacting traceability and quality.
to `laminar`'s crontab. For more information about `cron`, see `man crontab`.
`LAMINAR_REASON` is an optional human-readable string that will be displayed in the web UI as the cause of the build.
## Triggering on a git commit
This is what [git hooks](https://git-scm.com/book/gr/v2/Customizing-Git-Git-Hooks) are for. To create a hook that triggers the `example-build` job when a push is made to the `example` repository, create the file `hooks/post-receive` in the `example.git` bare repository.
For a more advanced example, see [examples/git-post-receive-hook-notes](https://github.com/ohwgiles/laminar/blob/master/examples/git-post-receive-hook-notes)
What if your git server is not the same machine as the laminar instance?
## Triggering on a remote laminar instance
`laminarc` and `laminard` communicate by default over an [abstract unix socket](http://man7.org/linux/man-pages/man7/unix.7.html). This means that any user **on the same machine** can send commands to the laminar service.
On a trusted network, you might want `laminard` to listen for commands on a TCP port instead. To achieve this, in `/etc/laminar.conf`, set
```
LAMINAR_BIND_RPC=*:9997
```
or any interface/port combination you like. This option uses the same syntax as `LAMINAR_BIND_HTTP`.
Then, point `laminarc` to the new location using an environment variable:
If you need more flexibility, consider running the communication channel as a regular unix socket and applying user and group permissions to the file. To achieve this, set
```
LAMINAR_BIND_RPC=unix:/var/run/laminar.sock
```
or similar path in `/etc/laminar.conf`.
This can be securely and flexibly combined with remote triggering using `ssh`. There is no need to allow the client full shell access to the server machine, the ssh server can restrict certain users to certain commands (in this case `laminarc`). See [the authorized_keys section of the sshd man page](https://man.openbsd.org/sshd#AUTHORIZED_KEYS_FILE_FORMAT) for further information.
## Triggering on a push to GitHub
Consider using [webhook](https://github.com/adnanh/webhook) or a similar application to call `laminarc`.
A job's console output can be viewed on the Web UI at http://localhost:8080/jobs/$NAME/$NUMBER.
Additionally, the raw log output may be fetched over a plain HTTP request to http://localhost:8080/log/$NAME/$NUMBER. The response will be chunked, allowing this mechanism to also be used for in-progress jobs. Furthermore, the special endpoint http://localhost:8080/log/$NAME/latest will redirect to the most recent log output. Be aware that the use of this endpoint may be subject to races when new jobs start.
A typical pipeline may involve several steps, such as build, test and deploy. Depending on the project, these may be broken up into separate laminar jobs for maximal flexibility.
The preferred way to accomplish this in Laminar is to use the same method as [regular run triggering](#Triggering-a-run), that is, calling `laminarc` directly in your `example.run` scripts.
```bash
#!/bin/bash -xe
# simultaneously starts example-test-qemu and example-test-target
# and returns a non-zero error code if either of them fail
laminarc run example-test-qemu example-test-target
```
An advantage to using this `laminarc` approach from bash or other scripting language is that it enables highly dynamic pipelines, since you can execute commands like
`laminarc` reads the `$JOB` and `$RUN` variables set by `laminard` and passes them as part of the queue/start/run request so the dependency chain can always be traced back.
If the script `/var/lib/laminar/cfg/jobs/example.before` exists, it will be executed as part of the `example` job, before the primary `/var/lib/laminar/cfg/jobs/example.run` script.
Similarly, if the script `/var/lib/laminar/cfg/jobs/example.after` script exists, it will be executed as part of the `example` job, after the primary `var/lib/laminar/cfg/jobs/example.run` script. In this script, the `$RESULT` variable will be `success`, `failed`, or `aborted` according to the result of `example.run`.
See also [script execution order](#Script-execution-order)
## Conditionally trigger a downstream job
Often, you may wish to only trigger the `example-test` job if the `example-build` job completed successfully. `example-build.after` might look like this:
Any script can set environment variables that will stay exposed for subsequent scripts of the same run using `laminarc set`. In `example.before`:
```bash
#!/bin/bash
laminarc set foo=bar
```
Then in `example.run`
```bash
#!/bin/bash
echo $foo # prints "bar"
```
---
# Archiving artefacts
Laminar's default behaviour is to remove the run directory `/var/lib/laminar/run/JOB/RUN` after its completion. This prevents the typical CI disk usage explosion and encourages the user to judiciously select artefacts for archive.
Laminar provides an archive directory `/var/lib/laminar/archive/JOB/RUN` and exposes its path in `$ARCHIVE`. `example-build.after` might look like this:
```bash
#!/bin/bash -xe
cp example.out $ARCHIVE/
```
This folder structure has been chosen to make it easy for system administrators to host the archive on a separate partition or network drive.
## Accessing artefacts from an upstream build
Rather than implementing a separate mechanism for this, the path of the upstream's archive should be passed to the downstream run as a parameter. See [Parameterized runs](#Parameterized-runs).
---
# Email and IM Notifications
As well as per-job `.after` scripts, a common use case is to send a notification for every job completion. If the global `after` script at `/var/lib/laminar/cfg/after` exists, it will be executed after every job. One way to use this might be:
```bash
#!/bin/bash -xe
if [ "$RESULT" != "$LAST_RESULT" ]; then
sendmail -t <<EOF
To: engineering@company.com
Subject: Laminar $JOB #$RUN: $RESULT
From: laminar-ci@company.com
Laminar $JOB #$RUN: $RESULT
EOF
fi
```
Of course, you can make this as pretty as you like. A [helper script](#Helper-scripts) can be a good choice here.
If you want to send to different addresses depending on the job, replace `engineering@company.com` above with a variable, e.g. `$RECIPIENTS`, and set `RECIPIENTS=nora@company.com,joe@company.com` in `/var/lib/laminar/cfg/jobs/JOB.env`. See [Environment variables](#Environment-variables).
You could also update the `$RECIPIENTS` variable dynamically based on the build itself. For example, if your run script accepts a parameter `$rev` which is a git commit id, as part of your job's `.after` script you could do the following:
See [examples/notify-email-pretty](https://github.com/ohwgiles/laminar/blob/master/examples/notify-email-pretty) and [examples/notify-email-text-log](https://github.com/ohwgiles/laminar/blob/master/examples/notify-email-text-log).
The directory `/var/lib/laminar/cfg/scripts` is automatically prepended to the `PATH` of all runs. It is a convenient place to drop executables or scripts to help keep individual job scripts clean and concise. A simple example might be `/var/lib/laminar/cfg/scripts/success_trigger`:
With this in place, any `.after` script can conditionally trigger a downstream job more succinctly:
```bash
success_trigger example-test
```
Another excellent candidate for helper scripts is automatically sending notifications on job status change.
---
# Data sharing and Workspaces
Often, a job will require a (relatively) large block of (relatively) unchanging data. Examples are a git repository with a long history, or static asset files. Instead of fetching everything from scratch for every run, a job may make use a *workspace*, a per-job folder that is reused between builds.
For example, the following script creates a tarball containing both compiled output and some static asset files from the workspace:
```bash
#!/bin/bash -ex
git clone /path/to/sources .
make
# Use a hardlink so the arguments to tar will be relative to the CWD
ln $WORKSPACE/StaticAsset.bin ./
tar zc a.out StaticAsset.bin > MyProject.tar.gz
# Archive the artefact (consider moving this to the .after script)
mv MyProject.tar.gz $ARCHIVE/
```
For a project with a large git history, it can be more efficient to store the sources in the workspace:
Laminar will automatically create the workspace for a job if it doesn't exist when a job is executed. In this case, the `/var/lib/laminar/cfg/jobs/JOBNAME.init` will be executed if it exists. This is an excellent place to prepare the workspace to a state where subsequent builds can rely on its content:
**CAUTION**: By default, laminar permits multiple simultaneous runs of the same job. If a job can **modify** the workspace, this might result in inconsistent builds when simultaneous runs access the same content. This is unlikely to be an issue for nightly builds, but for SCM-triggered builds it will be. To solve this, use [contexts](#Contexts) to restrict simultaneous execution of jobs, or consider [flock](https://linux.die.net/man/1/flock).
The following example uses [flock](https://linux.die.net/man/1/flock) to efficiently share a git repository workspace between multiple simultaneous builds:
```bash
#!/bin/bash -xe
# This script expects to be passed the parameter 'rev' which
# should refer to a specific git commit in its source repository.
# The commit ids could have been read from a server-side
# post-commit git hook, where many commits could have been pushed
# at once, but we want to check them all individually. This means
# this job can be executed several times (with different values
# for $rev) simultaneously.
# Locked subshell for modifying the workspace
(
flock 200
cd $WORKSPACE
# Download all the latest commits
git fetch
git checkout $rev
cd -
# Fast copy (hard-link) the source from the specific checkout
# to the build dir. This relies on the fact that git unlinks
# during checkout, effectively implementing copy-on-write.
In Laminar, each run of a job is associated with a context. The context defines an integer number of *executors*, which is the amount of runs which the context will accept simultaneously. A context may also provide additional environment variables.
Uses for this feature include limiting the amount of concurrent CPU-intensive jobs (such as compilation); and controlling access to jobs [executed remotely](#Remote-jobs).
If no contexts are defined, Laminar will behave as if there is a single context named "default", with `6` executors. This is a reasonable default that allows simple setups to work without any consideration of contexts.
## Defining a context
To create a context named "my-env" which only allows a single run at once, create `/var/lib/laminar/cfg/contexts/my-env.conf` with the content:
When trying to start a job, laminar will wait until the job can be matched to a context which has at least one free executor. There are two ways to associate jobs and contexts. You can specify a comma-separated list of patterns `JOBS` in the context configuration file `/var/lib/laminar/cfg/contexts/CONTEXT.conf`:
```
JOBS=amd64-target-*,usage-monitor
```
This approach is often preferred when you have many jobs that need to share limited resources.
in `/var/lib/laminar/cfg/jobs/JOB.conf`. This approach is often preferred when you have a small number of jobs that require exclusive access to an environment and you can supply alternative environments (e.g. target devices), because new contexts can be added without modifying the job configuration.
In both cases, Laminar will iterate over the known contexts and associate the run with the first matching context with free executors. Patterns are [glob expressions](http://man7.org/linux/man-pages/man7/glob.7.html).
This environment will then be available the run script of jobs associated with this context. Note that these definitions are not expanded by a shell, so `FOO="bar"` would result in a variable `FOO` whose contents *include* double-quotes.
Laminar provides no specific support, `bash`, `ssh` and possibly NFS are all you need. For example, consider two identical target devices on which test jobs can be run in parallel. You might create a [context](#Contexts) for each, `/var/lib/laminar/cfg/contexts/target{1,2}.conf`:
Laminar's frontend supports ANSI colours using the [ansi-up library](https://github.com/drudru/ansi_up). Unfortunately, there is no standard way of convincing applications to output colours when not connected to a tty. It is recommended to set [CLICOLOR_FORCE=1](https://bixense.com/clicolors/) in Laminar's [global environment file](#Environment-variables), plus any of the following environment variables that may be relevant (please submit more):
*Groups* may be used to organise the "Jobs" page into tabs. Edit `/var/lib/laminar/cfg/groups.conf` and define the matched jobs as a [javascript regular expression](https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Regular_Expressions), for example:
```
Builds=compile-\w+
My Fav Jobs=^(target-foo-(build|deploy)|run-benchmarks)$
All=.*
```
Changes to this file are detected immediately and will be visible on next page refresh.
If it exists, the file `/var/lib/laminar/custom/index.html` will be served by laminar instead of the default markup that is bundled into the Laminar binary. This file can be used to change any aspect of Laminar's WebUI, for example adding menu links or adding a custom stylesheet. Any required assets will need to be served directly from your [HTTP reverse proxy](#Service-configuration) or other HTTP server.
Laminar will serve a job's current status as a pretty badge at the url `/badge/JOBNAME.svg`. This can be used as a link to your server instance from your Github README.md file or cat blog:
`laminard` reads the following variables from the environment, which are expected to be sourced by `systemd` from `/etc/laminar.conf`:
-`LAMINAR_HOME`: The directory in which `laminard` should find job configuration and create run directories. Default `/var/lib/laminar`
-`LAMINAR_BIND_HTTP`: The interface/port or unix socket on which `laminard` should listen for incoming connections to the web frontend. Default `*:8080`
-`LAMINAR_BIND_RPC`: The interface/port or unix socket on which `laminard` should listen for incoming commands such as build triggers. Default `unix-abstract:laminar`
-`LAMINAR_TITLE`: The page title to show in the web frontend.
-`LAMINAR_KEEP_RUNDIRS`: Set to an integer defining how many rundirs to keep per job. The lowest-numbered ones will be deleted. The default is 0, meaning all run dirs will be immediately deleted.
-`LAMINAR_ARCHIVE_URL`: If set, the web frontend served by `laminard` will use this URL to form links to artefacts archived jobs. Must be synchronized with web server configuration.
Note that definitions in these files are not expanded by a shell, so `FOO="bar"` would result in a variable `FOO` whose contents *include* double-quotes.
Finally, variables supplied on the command-line call to `laminarc queue`, `laminarc start` or `laminarc run` will be available. See [parameterized runs](#Parameterized-runs)
-`set [VARIABLE=VALUE]...` sets one or more variables to be exported in subsequent scripts for the run identified by the `$JOB` and `$RUN` environment variables
`laminarc` connects to `laminard` using the address supplied by the `LAMINAR_HOST` environment variable. If it is not set, `laminarc` will first attempt to use `LAMINAR_BIND_RPC`, which will be available if `laminarc` is executed from a script within `laminard`. If neither `LAMINAR_HOST` nor `LAMINAR_BIND_RPC` is set, `laminarc` will assume a default host of `unix-abstract:laminar`.
All commands return zero on success or a non-zero code if the command could not be executed. `laminarc run` will return a non-zero exit status if any executed job failed.
