This page is prepared for Hadoop Core committers. You need committer rights to create a new Hadoop Core release.

These instructions have been updated to use dev-support/bin/create-release. Earlier versions of this document are at HowToReleaseWithSvnAndAnt and HowToReleasePostMavenization and HowToReleasePreDSBCR. For releasing from the 2.6.x or the 2.7.x line, you'll need to consult HowToReleasePreDSBCR to find applicable steps.



  1. If you have not already done so, append your code signing key to the KEYS file. Once you commit your changes, they will automatically be propagated to the website. Also upload your key to a public key server if you haven't. End users use the KEYS file (along with the web of trust) to validate that releases were done by an Apache committer. For more details on signing releases, see Signing Releases and Step-By-Step Guide to Mirroring Releases.

  2. Bulk update JIRA to unassign from this release all issues that are open non-blockers. This is involved since you can only bulk change issues within the same project, so minimally requires four bulk changes for each of HADOOP, HDFS, MAPREDUCE, and YARN. Editing the "Target Version/s" field is also a blind write, so you need to be careful not to lose any other fix versions that are set. For updating 3.0.0-beta1 to 3.0.0, the process looked like this:
    1. Start with this query:
      project in (HADOOP, HDFS, YARN, MAPREDUCE) AND "Target Version/s" = 3.0.0-beta1 and statusCategory != Done
    2. Filter this list down until it's only issues with a Target Version of just "3.0.0-beta1". My query ended up looking like:
      project in (HADOOP, HDFS, YARN, MAPREDUCE) AND "Target Version/s" = 3.0.0-beta1 and "Target Versions/" not in (2.9.0, 2.8.3, 2.8.2) AND statusCategory != Done
    3. Do the bulk update for each project individually to set the target version to 3.0.0.
    4. Check the query for the next most common set of target versions and again filter it down:
      project in (HADOOP, HDFS, YARN, MAPREDUCE) AND "Target Version/s" = 3.0.0-beta1 and "Target Version/s" = 2.9.0 and statusCategory != Done
      project in (HADOOP, HDFS, YARN, MAPREDUCE) AND "Target Version/s" = 3.0.0-beta1 and "Target Version/s" = 2.9.0 and "Target Version/s" not in (2.8.2, 2.8.3) and statusCategory != Done
    5. Do the bulk update for each project individually to set the target version field to (3.0.0, 2.9.0).
    6. Return to the original query. If there aren't too many, update the remaining straggler issues by hand (faster than doing the bulk edits):
      project in (HADOOP, HDFS, YARN, MAPREDUCE) AND "Target Version/s" = 3.0.0-beta1 and statusCategory != Done
  3. Send follow-up notification to the developer list that this was done.
  4. To deploy artifacts to the Apache Maven repository create ~/.m2/settings.xml:

    <settings xmlns=""
         <username>Apache username</username>
         <password>Apache password</password>


When releasing Hadoop X.Y.Z, the following branching changes are required. Note that a release can match more than one of the following if-conditions. For a major release, one needs to make the changes for minor and point releases as well. Similarly, a new minor release is also a new point release.

  1. If this is a new major release (i.e., Y = 0 and Z = 0)
    1. Create a new branch (branch-X) for all releases in this major release.
    2. Update the version on trunk to (X+1).0.0-SNAPSHOT
      mvn versions:set -DnewVersion=(X+1).0.0-SNAPSHOT
    3. Commit the version change to trunk.
      git commit -a -m "Preparing for (X+1).0.0 development"
  2. If this is a new minor release (i.e., Z = 0)
    1. Create a new branch (branch-X.Y) for all releases in this minor release.
    2. Update the version on branch-X to X.(Y+1).0-SNAPSHOT
      mvn versions:set -DnewVersion=X.(Y+1).0-SNAPSHOT
    3. Commit the version change to branch-X.
      git commit -a -m "Preparing for X.(Y+1).0 development"
  3. If this is a new point release (i.e., always)
    1. Create a new branch (branch-X.Y.Z) for this release.
    2. Update the version on branch-X.Y to X.Y.(Z+1)-SNAPSHOT
      mvn versions:set -DnewVersion=X.Y.(Z+1)-SNAPSHOT
    3. Commit the version change to branch-X.Y.
      git commit -a -m "Preparing for X.Y.(Z+1) development"
  4. Release branch (branch-X.Y.Z) updates:
    1. Update hadoop-project/src/site/markdown/ to reflect the right versions, new features and big improvements.

    2. Update the version on branch-X.Y.Z TO X.Y.Z
      mvn versions:set -DnewVersion=X.Y.Z

Now, for any branches in {trunk, branch-X, branch-X.Y, branch-X.Y.Z} that have changed, push them to the remote repo taking care of any conflicts.

git push <remote> <branch>

Creating the release candidate (X.Y.Z-RC<N>)

These steps need to be performed to create the Nth RC for X.Y.Z, where N starts from 0.

  1. Check if the release year for Web UI footer is updated (the property <release-year> in hadoop-project/pom.xml). If not, create a JIRA to update the property value to the right year, and propagate the fix from trunk to all necessary branches. Consider the voting time needed before publishing, it's better to use the year of (current time + voting time) here, to be consistent with the publishing time.

  2. In JIRA, ensure that only issues in the "Fixed" state have a "Fix Version" set to release X.Y.Z.
  3. Verify that $HOME/.gpg defaults to the key listed in the KEYS file.
  4. For the Apache release, a machine capable of running Docker- and Internet- capable, build the release candidate with create-release. Unless the --logdir is given, logs will be in the patchprocess/ directory. Artifacts will be in the target/artifacts NOTE: This will take quite a while, since it will download and build the entire source tree, including documentation and native components, from scratch to avoid maven repository caching issues hiding issues with the source release.

     dev-support/bin/create-release --asfrelease --docker --dockercache
  5. While it should fail create-release if there are issues, doublecheck the rat log to find and fix any potential licensing issues.

     grep 'Rat check' patchprocess/mvn_apache_rat.log
  6. Check that release files look ok - e.g. install it somewhere fresh and run examples from tutorial, do a fresh build, read the release notes looking for WARNINGs, etc.
  7. Set environment variable version for later steps. export version=X.Y.Z-RCN

  8. Tag the release candidate:
     git tag -s release-$version -m "Release candidate - $version"
  9. Push branch-X.Y.Z and the newly created tag to the remote repo.
  10. Deploy the maven artifacts, on your personal computer. Please be sure you have completed the prerequisite step of preparing the settings.xml file before the deployment. You might want to do this in private and clear your history file as your gpg-passphrase is in clear text.

    mvn deploy -Psign -DskipTests -DskipShade
  11. Copy release files to a public place and ensure they are readable. Note that only supports SFTP, so this may be easier with a graphical SFTP client like Nautilus, Konqueror, etc.

    > cd public_html
    > mkdir hadoop-${version}
    > put -r /home/hadoop/hadoop-${version}
    > bye
  12. Log into Nexus, select "Staging Repositories" from the left navigation pane, select the check-box against the specific hadoop repository, and close the release.

  13. Call a release vote on common-dev at It's usually a good idea to start the release vote on Monday so that people will have a chance to verify the release candidate during the week. Example

  14. If the release candidate contains a serious issue, withdraw the vote, make necessary changes, and repeat this process.
  15. If non-trivial changes are committed to the release branch, ensure the commits are present in the upstream branches.


In 5 days if the release vote passes, the release may be published.

  1. In JIRA, "release" the version, setting the date to the end-of-vote date. Visit the "Administer Project" page, then the "Manage versions" page. You need to have the "Admin" role in HADOOP, HDFS, MAPREDUCE, and YARN.
  2. Set environment variable version for later steps. export version=X.Y.Z

  3. Tag the release. Do it from the release branch and push the created tag to the remote repository:
    git tag -s rel/release-${version} -m "Hadoop ${version} release"
    git push origin rel/release-${version}
  4. Copy release files to the distribution directory
    1. Check out the corresponding svn repo if need be
      svn co hadoop-dist
    2. Copy the release files to hadoop-dist/hadoop-${version}
    3. Update the symlinks to current2 and stable2. The release directory usually contains just two releases, the most recent from two branches.
    4. Commit the changes (it requires a PMC privilege)
      svn ci -m "Publishing the bits for release ${version}"
  5. Update upstream branches to make them aware of this new release:
    1. Copy and commit the and
       cp target/artifacts/ hadoop-common-project/hadoop-common/src/site/markdown/release/${version}/RELEASENOTES.${version}.md
       cp target/artifacts/ hadoop-common-project/hadoop-common/src/site/markdown/release/${version}/CHANGES.${version}.md
    2. Copy the jdiff xml files for this version to their appropriate directory.
        cp hadoop-hdfs-project/hadoop-hdfs/target/site/jdiff/xml/Apache_Hadoop_HDFS_${version}.xml hadoop-hdfs-project/hadoop-hdfs/dev-support/jdiff
    3. Update hadoop-project-dist/pom.xml

  6. In Nexus

    1. effect the release of artifacts by selecting the staged repository and then clicking Release

    2. If there were multiple RCs, simply drop the staging repositories corresponding to failed RCs.
  7. Wait 24 hours for release to propagate to mirrors.
  8. Edit the website.
    1. Checkout the website if you haven't already
      svn co hadoop-common-site
    2. Update the documentation links in author/src/documentation/content/xdocs/site.xml.

    3. Update the release news in author/src/documentation/content/xdocs/releases.xml.

    4. Update the news on the home page author/src/documentation/content/xdocs/index.xml.

    5. Copy the new release docs to svn and update the docs/current link, by doing the following:

      cd publish/docs
      tar xvf /path/to/hadoop-${version}-site.tar.gz
      # Update current2, current, stable and stable2 as needed.
      # For example
      rm current2 current
      ln -s r${version} current2
      ln -s current2 current
    6. Similarly update the symlinks for stable if need be.
    7. Add the documentation changes.
      svn add publish/docs/r${version}
    8. Regenerate the site, review it, then commit it per the instructions in HowToCommit.

      <run Forrest>
      <preview changes>
      svn commit -m "Updated site for release X.Y.Z."
  9. Send announcements to the user and developer lists once the site changes are visible.
  10. In JIRA, close issues resolved in the release. Disable mail notifications for this bulk change. Recommend not closing, since it prevents JIRAs from being edited and makes it more difficult to track backports.

See Also

HowToRelease (last edited 2017-10-04 00:33:49 by AndrewWang)