Uploaded image for project: 'Blazegraph (by SYSTAP)'
  1. Blazegraph (by SYSTAP)
  2. BLZG-331

Stochastic failure in com.bigdata.service.ndx.pipeline.TestMasterTask.test_stress_startWriteStop2

    Details

      Description

      Occasional stochastic failure. This is a failure of the unit test itself. The javadoc for doStartWriteStop2Test() states:

      {
      There is still one fence post in some of the unit tests. The
      symptom is that the unit test fails due to an unexpected value for
      either masterStats.subtaskStartCount or masterStats.subtaskEndCount
      or for an unexpected value for either subtaskStats.chunksOut or
      subtaskStats.chunksIn. I suspect that the underlying issue is the
      expectation of the tests with respect to when a sink is retired by
      the master and whether or not a new subtaskStats object is
      allocated or an old one reused (e.g., because the subtask was not
      closed by an idle timeout or the like).

      This test failure can be cleared up if you uncomment the [tmp]
      list and its use in
      {@link AbstractMasterStats#getSubtaskStats(Object)}. This forces
      the internal map to hold a hard reference to the subtask statistics
      objects. This demonstrates that the problem is not with the
      termination conditions for the {@link AbstractMasterTask}. However,
      the unit test are still broken until I track down the underlying
      assumption within them which is being violated.
      }

        Activity

        Hide
        bryanthompson bryanthompson added a comment -

        Pending resolution of this issue, the test has been modified to log @ ERROR rather than failing. The log message includes the #of trials which would have failed and the total #of trials. The individual trial failures at logged @ WARN.

        Show
        bryanthompson bryanthompson added a comment - Pending resolution of this issue, the test has been modified to log @ ERROR rather than failing. The log message includes the #of trials which would have failed and the total #of trials. The individual trial failures at logged @ WARN.
        Hide
        bryanthompson bryanthompson added a comment -

        This test has not failed for quite some time.

        Show
        bryanthompson bryanthompson added a comment - This test has not failed for quite some time.

          People

          • Assignee:
            Unassigned
            Reporter:
            bryanthompson bryanthompson
          • Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

            Dates

            • Created:
              Updated:
              Resolved: