Fix node leak when skipping child jobs

Zuul has the possibility to skip child jobs using zuul_return. However
zuul still requests and locks nodes for these skipped jobs and never
returns them. If many jobs are skipped this can lead to a completely
wedged system. The only way to recover from this is to restart the
zuul-scheduler. This can be fixed by filtering out skipped jobs in
findJobsToRequest.

Change-Id: I69f7dd1fe142c5b4d8530407c3bca37fdd976597
Co-Authored-By: Tobias Henkel <tobias.henkel@bmw.de>

tests/fixtures/config/data-return/git/common-config/zuul.yaml

@@ -37,9 +37,17 @@
     success-url: docs/index.html
     run: playbooks/data-return-relative.yaml
 
+# This child job will be skipped in the test case test_data_return_child_jobs.
+# In order to verify that this doesn't lead to node leaks attach a nodeset to
+# it. Each test case automatically verifies that there are no open node
+# requests and no locked nodes left behind.
 - job:
     name: child
     run: playbooks/child.yaml
+    nodeset:
+      nodes:
+        - name: node
+          label: test
 
 - job:
     name: several-zuul-return-parent

zuul/model.py

@@ -2178,6 +2178,8 @@ class QueueItem(object):
             build = build_set.getBuild(job.name)
             if build and (build.result == 'SUCCESS' or build.paused):
                 successful_job_names.add(job.name)
+            elif build and build.result == 'SKIPPED':
+                pass
             else:
                 nodeset = build_set.getJobNodeSet(job.name)
                 if nodeset is None: