Back to blog

SCM API 2.0 Release Take 2

Stephen Connolly
February 6, 2017

In January we announced the release of SCM API 2.0. After the original release was published we identified four new high-impact issues. We decided to remove the new versions of the plugins from the update center until those issues could be resolved. The issues have now been resolved and the plugins are now available from the update center.

Summary for busy Jenkins Administrators

Upgrading should make multi-branch projects much better. When you are ready to upgrade you must ensure that you upgrade all the required plugins. If you miss some, just upgrade them and restart to fix the issue. And of course, it’s always a good idea to take a backup of your JENKINS_HOME before upgrading any plugins.

In the list below, version numbers in bold indicate a change from the original version in the original announcement

Folders Plugin

5.17 or newer

SCM API Plugin

2.0.2 or newer

Branch API Plugin

2.0.2 or newer

Git Plugin

This depends on the exact release line of the Git plugin that you are using.

  • Following the 2.6.x release line: 2.6.4 or newer

  • Following the 3.0.x release line (recommended): 3.0.4 or newer

Mercurial Plugin

1.58 or newer

GitHub Branch Source Plugin

2.0.1 or newer

BitBucket Branch Source Plugin

2.0.2 or newer

GitHub Organization Folders Plugin


Pipeline Multibranch Plugin

2.12 or newer

If you are using the Blue Ocean plugin

Blue Ocean Plugin

1.0.0-b22 or newer

Other plugins that may require updating:

GitHub API Plugin

1.84 or newer

GitHub Plugin

1.25.0 or newer

If you upgrade to Branch API 2.0.x and you have either the GitHub Branch Source or the BitBucket Branch Source plugins and you do not upgrade those instances to the 2.0.x line then your Jenkins instance will fail to start-up correctly.

The solution is just to upgrade the GitHub Branch Source or the BitBucket Branch Source plugin (as appropriate) to the 2.0.x line.

After an upgrade you will see the data migration warning (see the screenshot in JENKINS-41608 for an example) this is normal and expected. The unreadable data will be removed by the next scan / index or can be removed manually using the Discard Unreadable Data button. The warning will disappear on the next restart after the unreadable data has been removed.

Please update to the versions listed above. If you want to know more about the issues and how they were resolved, see the next section.

Analysis of the issues

The issues described below are resolved with these plugin releases:

  • Folders Plugin: 5.17

  • SCM API Plugin: 2.0.2

  • Branch API Plugin: 2.0.2

  • Git Plugin: Either 2.6.4 or 3.0.4

  • GitHub Branch Source Plugin: 2.0.1

  • BitBucket Branch Source Plugin: 2.0.2

  • Pipeline Multibranch Plugin: 2.12

Migration of GitHub branches from 1.x to 2.x resulted in a change of the implementation class used to identify branches. Some other other bugs in Branch API had been fixed and the combined effect resulted in a rebuild of all GitHub Branches (not PRs) after an upgrade to GitHub Branch Source Plugin 2.0.0. This rebuild was referred to as a "build storm".


  • The SCM API plugin was enhanced to add an extension point that allows for a second round of data migration when upgrading.

  • The second round of data migration allows plugins implementing the SCM API contract to fix implementation class issues in context.

  • The Branch API plugin was enhanced to use this new extension point.

  • The GitHub Branch Source plugin was enhanced to provide an implementation of this extension point.

The GitHub Branch Source and BitBucket Branch Source plugins in 1.x were not assigning consistent IDs to multi-branch projects discovered in an Organization Folder. Both plugins were fixed in 2.0.0 to assign consistent IDs as a change of ID would result in a rebuild of all projects. What was missed is that the very first scan of an Organization Folder after an upgrade will change the randomly assigned ID assigned by the 1.x plugins into the consistent ID assigned by the 2.0.0 plugins and consequently trigger a rebuild of all branches. This rebuild was referred to as a "build storm".


The Branch API plugin was enhanced to detect the case where a branch source has been changed but the change is only changing the ID. When such changes are identified, the downstream references of the ID are all updated which will prevent a build storm.

The BitBucket Branch Source 1.x did not store all the information about PRs that is required by the SCM API 2.0.x model. This could well have resulted in subtle effects when manually triggering a rebuild of a merge PR if the PR’s target branch has been modified after the PR branch was first detected by Jenkins. Consequently, as the information is required, BitBucket Branch Source plugin 2.0.0 populated the information with dummy values which would force the correct information to be retrieved. The side-effect is that all PR branches would be rebuilt.


  • The changes in SCM API 2.0.2 introduced to resolve JENKINS-41121 provided a path to resolve this issue without causing a rebuild of all PR branches.

  • The BitBucket Branch Source plugin was enhanced to provide an implementation of the new SCM API extension point that connects to BitBucket and retrieves the missing information.

During initial testing of the Branch API 2.0.0 release an issue was identified with how Organization Folders handled unusual names. None of the existing implementations of the SCMNavigator API could generate such unusual names due to form validation on GitHub / BitBucket replacing unusual characters with - when creating a repository.

It would be irresponsible to rely on external services sanitizing their input data for the correct operation of Organization Folders. Consequently, in Branch API 2.0.0 the names were all transformed into URL safe names, with the original URLs still resolving to the original projects so that any existing saved links would remain functional.

Quite a number of people objected to this change of URL scheme.


  • There has been a convention in Jenkins that the on-disk storage structure for jobs mirrors the URL structure. This is only a convention and there is nothing specific in the code that mandates following the convention.

  • The Folders Plugin was enhanced to allow for computed folders (where the item names are provided by an external source) to provide a strategy to use when generating the on-disk storage names as well as the URL component names for the folder’s child items.

  • The Branch API plugin was enhanced to use this new strategy for name transformation.

  • The net effect of this change is that the URLs remain the same as for 1.x but the on-disk storage uses transformed names that are future proofed against any new SCMNavigator implementations where the backing service allows names that are problematic to use as filesystem directory names.


  • The Branch API 2.0.0 approach handled the transformation of names by renaming the items using the Jenkins Item rename API.

  • The Branch API 2.0.2 approach does not rename the child items as it is only the on-disk storage location that is moved.

This means that the Jenkins Item rename API cannot be used.

At this time, the only known side-effect is in the Job Configuration History plugin. The configuration history of each child item will still be tracked going forward after the upgrade. The pre-upgrade configuration history is also retained. Because the Jenkins Item rename API cannot be used to flag the configuration file location change, there is no association between the pre-upgrade history chain and the post-upgrade history chain.

About the author

Stephen Connolly

This author has no biography defined. See social media links referenced below.