Fix duplicate setResult calls in deduplicated builds

We call item.setResult after a build is complete so that the queue
item can do any internal processing necessary (for example, prepare
data structures for child jobs, or move the build to the retry_builds
list).

In the case of deduplicated builds, we should do that for every queue
item the build participates in since each item may have a different
job graph.

We were not correctly identifying other builds of deduplicated jobs
and so in the case of a dependency cycle we would call setResult on
jobs of the same name in that cycle regardless of whether they were
deduplicated.

This corrects the issue and adds a test to detect that case.

Change-Id: I4c47beb2709a77c21c11c97f1d1a8f743d4bf5eb

1 files changed, 6 insertions, 3 deletions

diff --git a/zuul/manager/__init__.py b/zuul/manager/__init__.py
index 642aededd..417de9acd 100644
--- a/zuul/manager/__init__.py
+++ b/zuul/manager/__init__.py
@@ -1768,9 +1768,12 @@ class PipelineManager(metaclass=ABCMeta):
         build_in_items = [item]
         if item.bundle:
             for other_item in item.bundle.items:
-                if other_item not in build_in_items:
-                    if other_item.current_build_set.getBuild(build.job.name):
-                        build_in_items.append(other_item)
+                if other_item in build_in_items:
+                    continue
+                other_build = other_item.current_build_set.getBuild(
+                    build.job.name)
+                if other_build is not None and other_build is build:
+                    build_in_items.append(other_item)
         for item in build_in_items:
             # We don't care about some actions below if this build
             # isn't in the current buildset, so determine that before