Looking at these logs, transaction service is definitely running and I don't see errors from it. If it can't be discovered in ZooKeeper, another reason may be that the ZK connection was lost due to a timeout. How is the ZK configured for this test case? Is it possible that the test encounters a GC pause long enough for the ZK to lose its connection? By the way, I see other errors in that log, too, due to HDFS connection refused. Those don't seem related to Tephra, but may have the same root cause.
-Andreas On Wed, Oct 18, 2017 at 4:23 PM, James Taylor <[email protected]> wrote: > Here's a pointer to the failure in the test run: > https://builds.apache.org/job/Phoenix-master/1840/ > testReport/junit/org.apache.phoenix.tx/TxCheckpointIT/ > testUpsertSelectDoesntSeeUpsertedData_TxCheckpointIT_ > localIndex_false_mutable_false_columnEncoded_false_/ > > On Wed, Oct 18, 2017 at 1:39 PM, Andreas Neumann <[email protected]> wrote: > > > Hi James, > > > > do you have logs from that test run? There can be different reasons: > > - the test started before Tephra was initialized, in that case maybe wait > > until it is discoverable? > > - Transaction manager crashed for some reason - logs would show that > > > > -Andreas > > > > On Tue, Oct 17, 2017 at 8:04 PM, James Taylor <[email protected]> > > wrote: > > > > > During Phoenix test runs, we sometimes get this exception: > > > > > > java.lang.RuntimeException: org.apache.thrift.TException: Unable to > > > discover transaction service. > > > > > > Is this something that we should retry the test after receiving? What > > would > > > cause this? > > > > > > Thanks, > > > James > > > > > >
