Actually I rewrote the example to write to text rather than BigQuery and in the process changed both the nature and mechanism of its output, along with a new matcher for it.
So essentially all of the code involved is fairly new and I would find it easy to believe there was some source of flakiness in there. Let's see about multi-run deflaking to repro, and some good eyes on the example code might turn something up too. On Thu, Jan 12, 2017 at 7:29 PM, Jason Kuster < [email protected]> wrote: > This is a weird failure -- both WindowedWordCount tests failed with the > same "expected didn't match actual" error, both having the same "actual". > Mark, you wrote the matcher for this test; do you have more insight into > what's going on? > > On Thu, Jan 12, 2017 at 12:11 PM, Apache Jenkins Server < > [email protected]> wrote: > > > See <https://builds.apache.org/job/beam_PostCommit_Java_ > > MavenInstall/2341/changes> > > > > > > > -- > ------- > Jason Kuster > Apache Beam (Incubating) / Google Cloud Dataflow >
