Skip to content

ibmruntimes/ci-jenkins-pipelines

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Jenkins pipeline files for building OpenJDK

Eclipse Adoptium makes use of these scripts to build binaries on the build farm at https://ci.adoptium.net

Repository contents

This repository contains several useful scripts in order to build OpenJDK personally or at build farm scale via jenkins. For the Temurin project at Adoptium, this is done with the jenkins instance at ci.adoptium.net

  1. The docs folder contains images and utility scripts to produce up to date documentation.
  2. The pipelines folder contains the Groovy pipeline scripts for Jenkins (e.g. build | test | checksum | release).
  3. The tools folder contains pipelines/ analysis scripts that deliever success/failure trends and build scripts for code-tool dependancies for the build and test process (e.g. asmtools | jcov | jtharness | jtreg | sigtest).

For those who prefer diagrams, there is an overview of the information below including it's interactions with the scripts in other repositories in our [docs/ARCHITECTURE.md] file and specific ones on the pipeline types in [docs/diagram.md]. If you want to set up these pipelines in your own jenkins instance, see this guide.

Overview of pipeline types

The starting point on the jenkins instance from the perspective of the overall build pipelines is the build-scripts folder. This contains the high level pipelines which are used to run the different types of build. In the names in this document XX is the JDK version number e.g. 8, 17, 21 and so on. There is one of these for each JDK version which we support.

When talking about the different types of pipelines, the ones named "*openjdkXX-pipeline" are referred to as the "top level versioned pipelines" and the subjobs later on are the "platform specific pipelines"

openjdkXX-pipeline

These were historically used for regular builds of each of our release platforms using the current state of the master branch of the codebase - which is it's default behaviour - but is now run each time there is a new tag in the upstream openjdk codebase. These are triggered by the betaTrigger_XXea jobs in the build-scripts/utils folder. Note that JDK8 for, which comes from a separate codebase and therefore is tagged separately, is triggered via a separate betaTrigger_8ea_arm32Linux job.

The betaTrigger_XXea jobs use trigger_beta_build.groovy to determine when to run a build. This contains a trap for the expected GA release times to prevent triggering so that machine time is not used up while we are performing release build and test cycles.

Once complete, the openjdkXX-pipelines which will, by default, invoke the separate (refactor_openjdk_release_tool) job which will publish them as an ea-beta-suffixed release in github under e.g. temurin-21-binaries.

release-openjdkXX-pipeline

These are not publicly visible but are used to build the fully tested production binaries on a quarterly basis. Similar to the openjdkXX-pipeline jobs these are automatically triggered by the releaseTrigger_jdkXX jobs in build-scripts/utils every time a new -ga suffixed tag is detected.

releaseTrigger_jdkXX runs once an hour between the 10th and 25th of release months (Jan, Mar, Apr, Jul, Sep, Oct) and check for a new -ga tag. It runs triggerReleasePipeline.sh (from the mirror-scripts repository). That script has a loop that checks 5 times with a ten minute gap between them so that the overall trigger is checked every ten minutes during the days when it is active based on checking for the "expected" tag from releasePlan.cfg using the readExpectedGATag function in common.sh

A couple of points to note on the release configurations:

  • jdk8u_release.groovy, jdk11u_release.groovy and jdk17u_release.groovy do not automatically run the win32 (x32Windows) builds - they get triggered manually during a release cycle in order to prioritise the x64Windows builds during the release cycle on the machines.
  • Similarly the jdk8u_release.groovy does not include arm32, since that is built from a separate codebase and is tagged separately so cannot generally be triggered alongside the main builds.

evaluation-openjdkXX-pipeline

These are similar to the openjdkXX-pipeline jobs, and are triggered from the same betaTrigger_XXea jobs. The evaluation pipelines are for platforms which the Adoptium team are looking to potentially release at some point, but they are not yet reliable or sufficiently tested. which are not in the release.

weekly-openjdkXX-pipeline / weekly-evaluation-openjdkXX-pipeline

These are no longer used. These were triggered over the weekend with an extended set of tests, but since the regular openjdkXX-pipeline jobs are now running approximately once a week (the usual cadence of new tags appearing in the upstream codebases) we are running the full AQA suite in those pipelines. These were triggered by timer and then invoked the openjdkXX-pipeline jobs with the appropriate parameters.

trestle-openjdkXX-pipeline

Trestle is the name of the experimental project to allow upstream openjdk committers to run pipelines on our infrastructure in order to test code changes in openjdk on the full set of platforms which Temurin supports. They are triggered on demand from a subset of authorized users.

PR tester

In addition to the main pipelines we have "PR tester" jobs that are run on PRs to the pipelines repository in order to ensure they do not have any unintended side effects before they are merged. These are triggered when specific comments from authorized users are added into the PR. In that folder in jenkins there are separate versions of all of the openjdkXX-pipelines that can be used to run against PRs and will not "pollute" the history of the main pipelines.

More documentation on the PR tester process can be found in the prTester documentation.

Subjobs of the top level versioned pipelines (i.e. "platform specific pipelines")

Each of the top level versioned pipelines described above invoke lower level jobs to run the platform-specific builds. The jenkins folders containing these scripts for each of the above top level versioned pipelines are as follows:

Top level versioned pipeline Platform-specific pipeline folder (TODO: Name these!)
openjdkXX-pipeline jobs/jdkXX
evaluation-openjdkXX-pipeline jobs/evaluation/jdkXX [†]
weekly-openjdkXX-pipeline jobs/jdkXX (Shared with openjdkXX-pipeline)
release-openjdkXX-pipeline jobs/release/jobs/jdkXX (Restricted access)
PR testers build-test/jobs/jdkXX

[†] - The release jobs here are restricted access. The release folder here should also not be confused with the build-scripts/release folder which contains jobs used for the final publishing of the builds (early access or GA) to github

Note: jdkXX is generally related to the name of the upstream codebase, which will often have a u suffix. At the moment we have a separate set of jobs for non-u and u versions when the upstream codebase changes. TODO: Add note about the new branch process for jdk23+

Inside the jdkXX folders there are pipelines which perform a build of one variant (e.g. Temurin) for on JDK version on one platform, for example jdk21u-linux-aarch64-temurin which are reponsible for running the build using kick_off_build.groovy and initiating the tests and other jobs against the completed build if successful. A "Smoke Test" job such as jdk21u-linux-aarch64-temurin-SmokeTests which is similar to our test jobs but runs the tests from the temurin-build buildAndPackage directory which is initiated after the build perfoms some basic tests against the build artefacts and acts as a gate to kicking off the subsequent steps. Once complete, the openjdkXX-pipelines which run the early access builds will generally invoke the jobs to publish them as a release in github (e.g. temurin-21-binaries.

Job generation

As you can see from the above sections, there are a lot of separate jobs in jenkins which are used during the build process. Since there are so many of them, these are not created manually, but are autogenerated using an automatic generator process.

The top level build-pipeline-generator job uses build_pipeline_generator.groovy to generate the pipelines. It will generate the top level versioned openjdkXX-pipeline jobs. Similarly there are pipeline_jobs_generator_jdkXX jobs which use build_job_generator.groovy to generate the subjobs for each platform/variant combination. Both of these pipelines are triggered on a change (PR merge) to the ci-jenkins-pipelines repository. They will pause themselves if a pipeline is not running so as to avoid interfering with a currently executing pipeline. T Similarly there is an evaluation-pipeline-generator and evaluation-pipeline_jobs_generator_jdkXX for generating the evaluation jobs, a trestle-pipeline-generator for those jobs, plus release-pipeline-generator andand release_pipeline_jobs_generator_jdkXX for release jobs (the release generators are not triggered automatically but are re-run manually at certain points during each release cycle

The following is a brief summary of how the generation jobs work but more details can be found in the generator documentation

The generators make use of the following files in pipelines/jobs/configurations. The README file in that directory contains more details of the configuration format:

  • The jdkXX.groovy, jdkXX_evaluation.groovy, jdkXX_release.groovy to determine which platforms to configure and generate for each version.
  • The individual platform configurations, such as jenkins labels, are defined by jdkXX_pipeline_config.groovy files.

For more details on the regeneration process overall see the regeneration documentation

Metadata files generated with each build

Information about the metadata file generated alongside the build

Alongside the built assets a metadata file will be created with info about the build. This will be a JSON document of the form:

{
    "vendor": "Eclipse Adoptium",
    "os": "mac",
    "arch": "x64",
    "variant": "openj9",
    "version": {
        "minor": 0,
        "patch": null,
        "msi_product_version": "11.0.18.6",
        "security": 0,
        "pre": null,
        "adopt_build_number": 0,
        "major": 15,
        "version": "15+29-202007070926",
        "semver": "15.0.0+29.0.202007070926",
        "build": 29,
        "opt": "202007070926"
    },
    "scmRef": "<output of git describe OR buildConfig.SCM_REF>",
    "buildRef": "<build-repo-name/build-commit-sha>",
    "version_data": "jdk15",
    "binary_type": "debugimage",
    "sha256": "<shasum>",
    "full_version_output": "<output of java --version>",
    "makejdk_any_platform_args": "<output of configure to makejdk-any-platform.sh>",
    "configure_arguments": "<output of bash configure>"
}

The Metadata class is contained in the Metadata.groovy file and the Json is constructed and written in the openjdk_build_pipeline.groovy file.

It is worth noting the additional tags on the SemVer is the Adoptium build number.

Below are all of the keys contained in the metadata file and some example values that can be present.


  • vendor: Example values: [Eclipse Adoptium, Alibaba, Huawei]

This tag is used to identify the vendor of the JDK being built, this value is set in the build.sh file and defaults to "Temurin".


  • os: Example values: [windows, mac, linux, aix, solaris, alpine-linux]

This tag identifies the operating system the JDK has been built on (and should be used on).


  • arch: Example values: [aarch64, ppc64, s390x, x64, x86-32, arm]

This tag identifies the architecture the JDK has been built on and it intended to run on.


  • variant: Example values: [hotspot, temurin, openj9, corretto, dragonwell, bisheng, fast_startup]

This tag identifies the JVM being used by the JDK, "dragonwell" itself is not a JVM but is currently considered a variant in its own right.


  • variant_version:

This tag is used to identify a version number of the variant being built, it currently is exclusively used by OpenJ9 and has the following keys:

  • major: Example values: [0, 1]

  • minor: Example values: [22, 23, 24]

  • security: Example values: [0, 1]

  • tags: Example values: [m1, m2]


  • version:

This tag contains the full version information of the JDK built, it uses the VersionInfo.groovy class and the ParseVersion.groovy class.

It contains the following keys:

  • minor: Example values: [0]

  • security: Example Values: [0, 9, 252 272]

  • pre: Example values: [null]

  • adopt_build_number: Example values: [0] If the ADOPT_BUILD_NUMBER parameter is used to build te JDK that value will appear here, otherwise a default value of 0 appears.

  • major: Example values: [8, 11, 15, 16]

  • version: Example values: [1.8.0_272-202010111709-b09, 11.0.9+10-202010122348, 14.0.2+11-202007272039, 16+19-202010120348]

  • semver: Example values: [8.0.202+8.0.202008210941, 11.0.9+10.0.202010122348, 14.0.2+11.0.202007272039, 16.0.0+19.0.202010120339] Formed from the major, minor, security, and build number by the formSemver() function.

  • build: Example values: [6, 9, 18] The OpenJDK build number for the JDK being built.

  • opt: Example values: [202008210941, 202010120348, 202007272039]


  • scmRef: Example values: [dragonwell-8.4.4_jdk8u262-b10, jdk-16+19_adopt-61198-g59e3baa94ac, jdk-11.0.9+10_adopt-197-g11f44f68c5, 23f997ca1]

A reference the the base JDK repository being build, usually including a GitHub commit reference, i.e. jdk-16+19_adopt-61198-g59e3baa94ac links to https://github.com/adoptium/jdk/commit/59e3baa94ac via the commit SHA 59e3baa94ac.

Values that only contain a commit reference such as 23f997ca1 are OpenJ9 commits on their respective JDK repositories, for example 23f997ca1 links to the commit https://github.com/ibmruntimes/openj9-openjdk-jdk14/commit/23f997ca1.


  • buildRef: Example values: [temurin-build/fe0f2dba, temurin-build/f412a523] A reference to the build tools repository used to create the JDK, uses the format repository-name/commit-SHA.

  • version_data: Example values: [jdk8u, jdk11u, jdk14u, jdk]

  • binary_type: Example values: [jdk, jre, debugimage, testimage]

  • sha256: Example values: [20278aa9459e7636f6237e85fcd68deec1f42fa90c6c541a2dfa127f4156d3e2, 2f9700bd75a807614d6d525fbd8d016c609a9ea71bf1ffd5d4839f3c1c8e4b8e] A SHA to verify the contents of the JDK.

  • full_version_output: Example values:
openjdk version \"1.8.0_252\"\nOpenJDK Runtime Environment (Alibaba Dragonwell 8.4.4) (build 1.8.0_252-202010111720-b06)\nOpenJDK 64-Bit Server VM (Alibaba Dragonwell 8.4.4) (build 25.252-b06, mixed mode)\n`

The full output of the command java -version for the JDK.


  • configure_arguments: The full output generated by make/autoconf/configure from the JDK built.

Tag driven "beta" EA builds

pipelines/build/common/trigger_beta_build.groovy

Jenkins pipeline to automate triggering of "beta" EA builds from the publication of upstream build tags. Eclipse Adoptium no longer runs scheduled nightly/weekend builds, instead pipeline builds are triggered by this job.

The one exception to this is Oracle managed STS versions, whose builds are managed internal to Oracle and not published until the GA day. For these a triggerSchedule_weekly is required to build the upstream HEAD commits on a regular basis.

pipelines/build/common/trigger_beta_build.groovy job parameters:

  • String: JDK_VERSION - JDK version to trigger. (Numerical version number, 8, 11, 17, ...)

  • String: MIRROR_REPO - github repository where source mirror is located for the given JDK_VERSION

  • String: BINARIES_REPO - github organisation/repo template for where binaries are published for jdk-NN, "_NN_" gets replaced by the version

  • CheckBox: FORCE_MAIN - Force the trigger of the "main" pipeline build for the current latest build tag, even if it is already published

  • CheckBox: FORCE_EVALUATION - Force the trigger of the "evaluation" pipeline build for the current latest build tag, even if it is already published

  • Multi-line Text: OVERRIDE_MAIN_TARGET_CONFIGURATIONS - Override targetConfigurations for FORCE_MAIN, eg: { "x64Linux": [ "temurin" ], "x64Mac": [ "temurin" ] }

  • Multi-line Text: OVERRIDE_EVALUATION_TARGET_CONFIGURATIONS - Override targetConfigurations for FORCE_EVALUATION, eg: { "aarch64AlpineLinux": [ "temurin" ] }

Build status

Table generated with generateBuildMatrix.sh

Platform Java 8 Java 11 Java 17 Java 21 Java HEAD
aix-ppc64-temurin Build Status Build Status Build Status Build Status Build Status
alpine-linux-aarch64-temurin Build Status Build Status Build Status N/A Build Status
alpine-linux-x64-temurin Build Status Build Status Build Status Build Status Build Status
linux-aarch64-temurin Build Status Build Status Build Status Build Status Build Status
linux-arm-temurin Build Status Build Status Build Status Build Status Build Status
linux-ppc64le-temurin Build Status Build Status Build Status Build Status Build Status
linux-riscv64-temurin N/A N/A N/A N/A Build Status
linux-riscv64-temurin-cross N/A N/A N/A N/A Build Status
linux-s390x-temurin Build Status Build Status Build Status Build Status Build Status
linux-x64-temurin Build Status Build Status Build Status Build Status Build Status
mac-aarch64-temurin N/A Build Status Build Status Build Status Build Status
mac-x64-temurin Build Status Build Status Build Status Build Status Build Status
solaris-sparcv9-temurin Build Status N/A N/A N/A N/A
solaris-x64-temurin Build Status N/A N/A N/A N/A
windows-aarch64-temurin N/A Build Status Build Status N/A Build Status
windows-x64-temurin Build Status Build Status Build Status Build Status Build Status
windows-x64-temurin_reproduce_compare N/A N/A N/A N/A N/A
windows-x86-32-temurin Build Status Build Status Build Status N/A N/A

Packages

No packages published

Languages

  • Groovy 96.0%
  • Shell 3.8%
  • Dockerfile 0.2%