ohwgiles_laminar/README.md

# Laminar CI

[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.

## Getting Started

### Building from source

First install development packages for `capnproto (git)`, `rapidjson`, `websocketpp`, `sqlite` and `boost-filesystem` from your distribution's repository or other source. Then:

```bash
git clone https://github.com/ohwgiles/laminar.git
cd laminar
cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/
make -j4
sudo make install
```

`make install` includes a systemd unit file. If you intend to use it, consider creating a new user `laminar` or modifying the user specified in the unit file.

### Installation from binaries

Alternatively to the source-based approach shown above, precompiled packages are supplied for x86_64 Debian Stable and CentOS 7.

Under Debian:

```bash
wget https://github.com/ohwgiles/laminar/releases/download/0.4/laminar-0.4-1-amd64.deb
sudo dpkg -i laminar-0.4-1-amd64.deb
```

Under CentOS:

```bash
wget https://github.com/ohwgiles/laminar/releases/download/0.4/laminar-0.4-1.x86_64.rpm
sudo rpm -i laminar-0.4-1.x86_64.rpm
```

Both install packages will create a new `laminar` user and install (but not activate) a systemd service for launching the laminar daemon.

### Service Configuration

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), simply run laminar as a reverse proxy behind a regular webserver.

#### Running behind a reverse proxy

Laminar relies on WebSockets to provide a responsive, auto-updating display without polling. This may require extra support from your frontend webserver.

For nginx, see [NGINX Reverse Proxy](https://www.nginx.com/resources/admin-guide/reverse-proxy/) and [WebSocket proxying](http://nginx.org/en/docs/http/websocket.html).

For Apache, see [Apache Reverse Proxy](https://httpd.apache.org/docs/2.4/howto/reverse_proxy.html) and [mod_proxy_wstunnel](https://httpd.apache.org/docs/2.4/mod/mod_proxy_wstunnel.html).

#### Set the page title

Change `LAMINAR_TITLE` in `/etc/laminar.conf` to your preferred page title.

#### More configuration options

See the [reference section](#reference)

## User Guide

### Terminology

- *job*: a task, identified by a name, comprising of one or more executable scripts.
- *run*: a numbered execution of a *job*

Throughout this document, the fixed path `/var/lib/laminar` is used. This is simply a default value and can be changed by setting `LAMINAR_HOME` in `/etc/laminar.conf` as desired.

### Creating 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:

```bash
#!/bin/bash -ex
wget ftp://ftp.gnu.org/gnu/hello/hello-2.10.tar.gz
tar xzf hello-2.10.tar.gz
cd hello-2.10
./configure
make
```

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).

Don't forget to mark the script executable:

```bash
chmod +x /var/lib/laminar/cfg/hello.run
```

### Triggering a run

To trigger a run of the `hello` job, execute

```bash
laminarc trigger hello
```

This causes the `/var/lib/laminar/cfg/hello.run` script to be executed, with a working directory of `/var/lib/laminar/run/hello/1`

The result and log output should be visible in the Web UI at http://localhost:8080/jobs/hello/1

#### Isn't there a "Build Now" button I can click?

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.

#### Triggering a job at a certain time

This is what `cron` is for. To trigger a build of `hello` every day at 0300, add

```
0 3 * * * LAMINAR_REASON="Nightly build" laminarc trigger hello
```

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.

```bash
#!/bin/bash
LAMINAR_REASON="Push to git repository" laminarc trigger example-build
```

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:

```bash
LAMINAR_HOST=192.168.1.1:9997 laminarc trigger example
```

##### Access control

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.

### Job chains

A typical pipeline may involve several steps, such as build, test and deploy. Depending on the project, these may be broken up into seperate 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.

In addition to `laminarc trigger`, `laminar start` triggers a job run, but waits for its completion and returns a non-zero exit code if the run failed. Furthermore, both `trigger` and `start` will accept multiple jobs in a single invocation:

```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 start 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

```bash
if [ ... ]; then
  laminarc start example-downstream-special
else
  laminarc start example-downstream-regular
fi

laminarc start example-test-$TARGET_PLATFORM
```

`laminarc` reads the `$JOB` and `$RUN` variables set by `laminard` and passes them as part of the trigger/start request so the dependency chain can always be traced back.

### Parameterized runs

Any argument passed to `laminarc` of the form `var=value` will be exposed as an environment variable in the corresponding build scripts. For example:

```bash
laminarc trigger example foo=bar
```

In `/var/lib/laminar/cfg/jobs/example.run`:

```bash
#!/bin/bash
if [ "$foo" == "bar" ]; then
   ...
else
   ...
fi
```

### Pre-build actions

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.

See also [script execution order](#script-execution-order)

#### Passing variables between run scripts

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"
```

This works because laminarc reads `$JOB` and `$NUM` and passes them to the laminar daemon as part of the `set` request. (It is thus possible to set environment variables on other jobs by overriding these variables, but this is not very useful).

### Post-build actions

Analagously to [Pre-build actions](#pre-build-actions), if the script `example.after` exists, it will be executed after the primary `example.run` script.

The `$RESULT` environment variable will contain the run result. See also [Environment variables](#environment-variables).

#### 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.


#### 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:

```bash
#!/bin/bash -xe
if [ "$RESULT" == "success" ]; then
  laminarc trigger example-test
fi
```

#### Accessing artifacts 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 dependending 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:

```bash
author_email=$(git show -s --format='%ae' $rev)
laminarc set RECIPIENTS $author_email
```

### Helper scripts

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`:

```bash
#!/bin/bash -e
if [ "$RESULT" == "success" ]; then
  laminarc trigger "$@"
fi
```

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. See [Example scripts](#appendix-example-scripts) for more useful starting points.

### 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 artifact (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:

```bash
#!/bin/bash -ex
cd $WORKSPACE/myproject
git pull
cd -

cmake $WORKSPACE/myproject
make -j4
```

**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 the job has multiple simultaneous runs. This is unlikely to be an issue for nightly builds, but for SCM-triggered builds it will be. To solve this, use [nodes](#nodes-and-tags) to restrict simultaneous execution of jobs, or [locks](#locks) to temporarily take exclusive control of a resource.

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.

### Nodes and Tags

In Laminar, a *node* is an abstract concept allowing more fine-grained control over job execution scheduling. Each node can be defined to support an integer number of *executors*, which defines how many runs can be executed simultaneously.

A typical example would be to allow only a few concurrent CPU-intensive jobs (such as compilation), while simultaneously allowing many more less-intensive jobs (such as monitoring or remote jobs). To create a node named `build` with 3 executors, create the file `/var/lib/laminar/cfg/nodes/build.conf` with the following content:

```
EXECUTORS=3
```

To associate jobs with nodes, laminar uses *tags*. Tags may be applied to nodes and jobs. If a node has tags, only jobs with a matching tag will be executed on it. If a node has no tags, it will accept any job. To tag a node, add them to `/var/lib/laminar/cfg/nodes/NODENAME.conf`:

```
EXECUTORS=3
TAGS=tag1,tag2
```

To add a tag to a job, add the following to `/var/lib/laminar/cfg/jobs/JOBNAME.conf`:

```
TAGS=tag2
```

If Laminar cannot find any node configuration, it will assume a single node with 6 executors and no tags.

#### Grouping jobs with tags

Tags are also used to group jobs in the web UI. Each tag will presented as a tab in the "Jobs" page.

#### Node scripts

If `/var/lib/laminar/cfg/nodes/NODENAME.before` exists, it will be executed before the run script of a job scheduled to that node. Similarly, if `/var/lib/laminar/cfg/nodes/NODENAME.after` exists, it will be executed after the run script of a job scheduled to that node.

#### Node environment

If `/var/lib/laminar/cfg/nodes/NODENAME.env` exists and can be parsed as a list of `KEY=VALUE` pairs, these variables will be exposed as part of the run's environment.

### Remote jobs

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 [node](#nodes-and-tags) for each, `/var/lib/laminar/cfg/nodes/target{1,2}.conf` with a common tag:

```
EXECUTORS=1
TAGS=remote-target
```

In each node's `.env` file, set the individual device's IP address:

```
TARGET_IP=192.168.0.123
```

And tag the job accordingly in `/var/lib/laminar/cfg/jobs/myproject-test.conf`:

```
TAGS=remote-target
```

This means the job script `/var/lib/laminar/cfg/jobs/myproject-test.run` can be generic:

```bash
#!/bin/bash -e

ssh root@$TARGET_IP /bin/bash -xe <<"EOF"
  uname -a
  ...
EOF
scp root@$TARGET_IP:result.xml "$ARCHIVE/"
```

Don't forget to add the `laminar` user's public ssh key to the remote's `authorized_keys`.

### Docker container jobs

Laminar provides no specific support, but just like [remote jobs](#remote-jobs) these are easily implementable in plain bash:

```bash
#!/bin/bash

docker run --rm -ti -v $PWD:/root ubuntu /bin/bash -xe <<EOF
  git clone http://...
  ...
EOF
```

### Locks

*Locks* are a simple way to control access to shared resources. Any string may be used as a lock name. The command `laminarc lock mylock` locks `mylock`. Subsequent calls to `laminarc lock mylock` will block until `laminarc release mylock` is called a corresponding number of times.

**CAUTION**: Locks are independent of any other job control mechanism in laminar, and will not be released automatically. Making sure calls to lock and release are symmetric is the administrator's responsibility.

An example use builds on the situation described in [Data sharing and Workspaces](#data-sharing-and-workspaces), where a large git repository is stored in the workspace. Conisder this run script:

```bash
#!/bin/bash -x

# 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 commit ids)
# at once.

cd $WORKSPACE
# Acquire a lock for modifying the workspace
laminarc lock $JOB-workspace
# Download all the latest commits
git fetch
git checkout $rev
cd -
# Fast copy (hard-link) the specific checkout to the build dir
cp -al $WORKSPACE/src src
# Release the lock to allow other jobs to do the same
laminarc release $JOB-workspace

# run the (much longer) build process
set -e
cmake src
make
```


## Reference

### Server options

`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_RUNDIR`: If set, do not delete run directories after completion
- `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.

### Script execution order

When `$JOB` is triggered on `$NODE`, the following scripts (relative to `$LAMINAR_HOME/cfg`) may be triggered:

- `jobs/$JOB.init` if the [workspace](#data-sharing-and-workspaces) did not exist
- `before`
- `nodes/$NODE.before`
- `jobs/$JOB.before`
- `jobs/$JOB.run`
- `jobs/$JOB.after`
- `nodes/$NODE.after`
- `after`

### Environment variables

The following variables are available in run scripts:

- `RUN` integer number of this *run*
- `JOB` string name of this *job*
- `RESULT` string run status: "success", "failed", etc.
- `LAST_RESULT` string previous run status
- `WORKSPACE` path to this job's workspace
- `ARCHIVE` path to this run's archive

In addition, `$LAMINAR_HOME/cfg/scripts` is prepended to `$PATH`. See [helper scripts](#helper-scripts).

Laminar will also export variables in the form `KEY=VALUE` found in these files:

- `env`
- `nodes/$NODE.env`
- `jobs/$JOB.env`

Finally, variables supplied on the command-line call to `laminarc start` or `laminarc trigger` will be available. See [parameterized builds](#parameterized-builds)

### `laminarc`

`laminarc` commands are:

- `trigger [JOB [PARAMS...]]...` triggers one or more jobs with optional parameters, returning immediately.
- `start [JOB [PARAMS...]]...` triggers one or more jobs with optional parameters and waits for the completion of all jobs. Returns a non-zero error code if any job failed.
- `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
- `lock [NAME]` acquires the [lock](#locks) named `NAME`
- `release [NAME]` releases the lock named `NAME`

`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`.

## Appendix: Example scripts