Extending Pipeline with Shared Libraries

As Pipeline is adopted for more and more projects in an organization, common patterns are likely to emerge. Oftentimes it is useful to share parts of Pipelines between various projects to reduce redundancies and keep code "DRY" ^[1].

Pipeline has support for creating "Shared Libraries" which can be defined in external source control repositories and loaded into existing Pipelines.

Defining Shared Libraries

An Shared Library is defined with a name, a source code retrieval method such as by SCM, and optionally a default version. The name should be a short identifier as it will be used in scripts.

The version could be anything understood by that SCM; for example, branches, tags, and commit hashes all work for Git. You may also declare whether scripts need to explicitly request that library (detailed below), or if it is present by default. Furthermore, if you specify a version in Jenkins configuration, you can block scripts from selecting a different version.

The best way to specify the SCM is using an SCM plugin which has been specifically updated to support a new API for checking out an arbitrary named version (Modern SCM option). As of this writing, the latest versions of the Git and Subversion plugins support this mode; others should follow.

If your SCM plugin has not been integrated, you may select Legacy SCM and pick anything offered. In this case, you need to include ${library.yourLibName.version} somewhere in the configuration of the SCM, so that during checkout the plugin will expand this variable to select the desired version. For example, for Subversion, you can set the Repository URL to https://svnserver/project/${library.yourLibName.version} and then use versions such as trunk or branches/dev or tags/1.0.

Directory structure

The directory structure of a Shared Library repository is as follows:

(root)
+- src                     # Groovy source files
|   +- org
|       +- foo
|           +- Bar.groovy  # for org.foo.Bar class
+- vars
|   +- foo.groovy          # for global 'foo' variable
|   +- foo.txt             # help for 'foo' variable
+- resources               # resource files (external libraries only)
|   +- org
|       +- foo
|           +- bar.json    # static helper data for org.foo.Bar

The src directory should look like standard Java source directory structure. This directory is added to the classpath when executing Pipelines.

The vars directory hosts scripts that define global variables accessible from Pipeline. The basename of each *.groovy file should be a Groovy (~ Java) identifier, conventionally camelCased. The matching *.txt, if present, can contain documentation, processed through the system’s configured markup formatter (so may really be HTML, Markdown, etc., though the txt extension is required).

The Groovy source files in these directories get the same “CPS transformation” as in Scripted Pipeline.

A resources directory allows the libraryResource step to be used from an external library to load associated non-Groovy files. Currently this feature is not supported for internal libraries.

Other directories under the root are reserved for future enhancements.

Global Shared Libraries

There are several places where Shared Libraries can be defined, depending on the use-case. Manage Jenkins » Configure System » Global Pipeline Libraries as many libraries as necessary can be configured.

Since these libraries will be globally usable, any Pipeline in the system can utilize functionality implemented in these libraries.

These libraries are considered "trusted:" they can run any methods in Java, Groovy, Jenkins internal APIs, Jenkins plugins, or third-party libraries. This allows you to define libraries which encapsulate individually unsafe APIs in a higher-level wrapper safe for use from any job. Beware that anyone able to push commits to this SCM repository could obtain unlimited access to Jenkins. You need the Overall/RunScripts permission to configure these libraries (normally this will be granted to Jenkins administrators).

Folder-level Shared Libraries

Any Folder created can have Shared Libraries associated with it. This mechanism allows scoping of specific libraries to all the Pipelines inside of the folder or subfolder.

Folder-based libraries are not considered "trusted:" they run in the Groovy sandbox just like typical Pipelines.

Automatic Shared Libraries

Other plugins may add ways of defining libraries on the fly. For example, the GitHub Organization Folder plugin allows a script to use an untrusted library such as github.com/someorg/somerepo without any additional configuration. In this case, the specified GitHub repository would be loaded, from the master branch, using an anonymous checkout.

Using libraries

Pipelines can access shared libraries marked Load implicitly, They may immediately use classes or global variables defined by any such libraries (details below).

To access other shared libraries, a script needs to use the @Library annotation, specifying the library’s name:

@Library('somelib')
/* Using a version specifier, such as branch, tag, etc */
@Library('somelib@1.0')
/* Accessing multiple libraries with one statement */
@Library(['somelib', 'otherlib@abc1234'])

The annotation can be anywhere in the script where an annotation is permitted by Groovy. When referring to class libraries (with src/ directories), conventionally the annotation goes on an import statement:

@Library('somelib')
import com.mycorp.pipeline.somelib.UsefulClass

It is legal, though unnecessary, to import a global variable (or method) defined in the vars/ directory:

Note that libraries are resolved and loaded during compilation of the script, before it starts executing. This allows the Groovy compiler to understand the meaning of symbols used in static type checking, and permits them to be used in type declarations in the script, for example:

@Library('somelib')
import com.mycorp.pipeline.somelib.Helper

int useSomeLib(Helper helper) {
    helper.prepare()
    return helper.count()
}

echo useSomeLib(new Helper('some text'))

Global Variables however, are resolved at runtime.

Overriding versions

A @Library annotation may override a default version given in the library’s definition, if the definition permits this. In particular, a shared library marked for implicit use can still be loaded in a different version using the annotation (unless the definition specifically forbids this).

Writing libraries

At the base level, any valid Groovy code is okay for use. Different data structures, utility methods, etc, such as:

// src/org/foo/Point.groovy
package org.foo;

// point in 3D space
class Point {
  float x,y,z;
}

Accessing steps

Library classes cannot directly call steps such as sh or git. They can however implement methods, outside of the scope of an enclosing class, which in turn invoke Pipeline steps, for example:

// src/org/foo/Zot.groovy
package org.foo;

def checkOutFrom(repo) {
  git url: "git@github.com:jenkinsci/${repo}"
}

Which can then be called from a Scripted Pipeline:

def z = new org.foo.Zot()
z.checkOutFrom(repo)

This approach has limitations; for example, it prevents the declaration of a superclass.

Alternately, a set of steps can be passed explicitly to a library class, in a constructor, or just one method:

package org.foo
class Utilities implements Serializable {
  def steps
  Utilities(steps) {this.steps = steps}
  def mvn(args) {
    steps.sh "${steps.tool 'Maven'}/bin/mvn -o ${args}"
  }
}

When saving state on classes, such as above, the class must implement the Serializable interface. This ensures that a Pipeline using the class, as seen in the example below, can properly suspend and resume in Jenkins.

@Library('utils') import org.foo.Utilities
def utils = new Utilities(steps)
node {
  utils.mvn 'clean package'
}

If the library needs to access global variables, such as env, those should be explicitly passed into the library classes, or methods, in a similar manner.

Instead of passing numerous variables from the Scripted Pipeline into a library,

package org.foo
class Utilities {
  static def mvn(script, args) {
    script.sh "${script.tool 'Maven'}/bin/mvn -s ${script.env.HOME}/jenkins.xml -o ${args}"
  }
}

The above example shows the script being passed in to one static method, invoked from a Scripted Pipeline as follows:

@Library('utils') import static org.foo.Utilities.*
node {
  mvn this, 'clean package'
}

Defining global variables

Internally, scripts in the vars directory are instantiated on-demand as singletons. This allows multiple methods or properties to be defined in a single .groovy file which interact with each other, for example:

// vars/acme.groovy
def setName(value) {
    this.name = value;
}
def getName() {
    return this.name;
}
def caution(message) {
    echo "Hello, ${this.name}! CAUTION: ${message}"
}

The Pipeline can then invoke these methods which will be defined on the acme object:

acme.name = 'Alice'
echo acme.name /* prints: 'Alice' */
acme.caution 'The queen is angry!' /* prints: 'Hello, Alice. CAUTION: The queen is angry!' */

A variable defined in a shared library will only show up in Global Variables Reference (under Pipeline Syntax) after Jenkins loads and uses that library as part of a successful Pipeline run.

Defining steps

Shared Libraries can also define global variables which behave similarly to built-in steps, such as sh or git. Global variables defined in Shared Libraries must be named with all lower-case or "camelCased" in order to be loaded properly by Pipeline. ^[2]

For example, to define sayHello, the file vars/sayHello.groovy should be created and should implement a call method. The call method allows the global variable to be invoked in a manner similar to a step:

// vars/sayHello.groovy
def call(String name = 'human') {
    // Any valid steps can be called from this code, just like in other
    // Scripted Pipeline
    echo "Hello, ${name}."
}

The Pipeline would then be able to reference and invoke this variable:

sayHello 'Joe'
sayHello() /* invoke with default arguments */

If called with a block, the call method will receive a Closure. The type should be defined explicitly to clarify the intent of the step, for example:

// vars/windows.groovy
def call(Closure body) {
    node('windows') {
        body()
    }
}

The Pipeline can then use this variable like any built-in step which accepts a block:

windows {
    bat "cmd /?"
}

Defining a more structured DSL

If you have a lot of Pipeline jobs that are mostly similar, the global variable mechanism gives you a handy tool to build a higher-level DSL that captures the similarity. For example, all Jenkins plugins are built and tested in the same way, so we might write a step named buildPlugin:

// vars/buildPlugin.groovy
def call(body) {
    // evaluate the body block, and collect configuration into the object
    def config = [:]
    body.resolveStrategy = Closure.DELEGATE_FIRST
    body.delegate = config
    body()

    // now build, based on the configuration provided
    node {
        git url: "https://github.com/jenkinsci/${config.name}-plugin.git"
        sh "mvn install"
        mail to: "...", subject: "${config.name} plugin build", body: "..."
    }
}

Assuming the script has either been loaded as a Global Shared Library or as a Folder-level Shared Library the resulting Jenkinsfile will be dramatically simpler:

Jenkinsfile (Scripted Pipeline)

buildPlugin {
    name = 'git'
}

Using third-party libraries

It is possible to use third-party Java libraries, typically found in Maven Central, from trusted library code using the @Grab annotation. Refer to the Grape documentation for details, but simply put:

@Grab('org.apache.commons:commons-math3:3.4.1')
import org.apache.commons.math3.primes.Primes
void parallelize(int count) {
  if (!Primes.isPrime(count)) {
    error "${count} was not prime"
  }
  // …
}

Third-party libraries are cached by default in ~/.groovy/grapes/ on the Jenkins master.

Loading resources

External libraries may load adjunct files from a resources/ directory using the libraryResource step. The argument is a relative pathname, akin to Java resource loading:

def request = libraryResource 'com/mycorp/pipeline/somelib/request.json'

The file is loaded as a string, suitable for passing to certain APIs or saving to a workspace using writeFile.

It is advisable to use an unique package structure so you do not accidentally conflict with another library.

Pretesting library changes

If you notice a mistake in a build using an untrusted library, simply click the Replay link to try editing one or more of its source files, and see if the resulting build behaves as expected. Once you are satisfied with the result, follow the diff link from the build’s status page, and apply the diff to the library repository and commit.

(Even if the version requested for the library was a branch, rather than a fixed version like a tag, replayed builds will use the exact same revision as the original build: library sources will not be checked out again.)

Replay is not currently supported for trusted libraries. Modifying resource files is also not currently supported during Replay.

Pipeline Guided Tour

Pipeline Resources

User Handbook

Developers

Writing plugins

JavaDoc