We moved this to pipeline completion so that we have the
total count of failures across all builds. We need this count
to check if the limit is reached.