[test case] -> this test case is similar as test case from JENKINS-12077
// settings
1. create matrix project
2. create axis1 [a b], axis2 [c d]
3. create LOCK-$EXPORTED_AXIS1
4. Inject environment variables to the build process EXPORTED_AXIS1=$axis1
5. add critical-block-start
6. add windows batch command e.g. dir d:\ /s /b | find "lock"
7. add critical-block-end
// running
8. run build -> #1
9. run build and stop it -> #2
10. run build it -> it causes 'deadlock' there is unreleased resource from build #2

I write (locally) simple patch in IdAllocationManager.java which can works. It release resource from completed build whom downstream projects are also completed.

new method:

private void releaseDeadlockedResource(String id, BuildListener buildListener) throws InterruptedException, IOException {
PrintStream logger = buildListener.getLogger();

// check if 'lockable resource' exists
AbstractBuild resourceOwner = ids.get(id);

if (resourceOwner != null && !resourceOwner.isBuilding()) { // build was completed
List<AbstractProject> downstreamProjects = resourceOwner.getProject().getDownstreamProjects();
boolean canRelease = true;

for (Iterator<AbstractProject> it = downstreamProjects.iterator(); it.hasNext() {
AbstractProject proj = it.next();
if (proj.isBuilding())

{ canRelease = false; break; }

}
if (canRelease)

{ logger.println("Release resource from completed build: " + resourceOwner.toString()); ids.remove(id); }

}
}

and changes in allocate() method:

public synchronized String allocate(AbstractBuild owner, String id, BuildListener buildListener) throws InterruptedException, IOException {

PrintStream logger = buildListener.getLogger();

boolean printed = false;
while (ids.get(id) != null) {
if (printed == false)

{ logger.println("Waiting ressource : " + id + " currently use by : " + ids.get(id).toString()); printed = true; }

wait(1000);

// periodically detect if any locked resource belongs to the completed build
releaseDeadlockedResource(id, buildListener);
}

ids.put(id, owner);
return id;
}