I took a look at TestSpillStress and found a bunch of problems with it. There's enough here that it merits a separate bug:
https://issues.cloudera.org/browse/IMPALA-4914 On Wed, Feb 8, 2017 at 3:45 PM, Tim Armstrong <[email protected]> wrote: > For what it's worth there's a similar problem with TestSpillStress that I > came across too. I was surprised when I realised too. > > On Wed, Feb 8, 2017 at 3:38 PM, Michael Brown <[email protected]> wrote: > >> Hello, >> >> Background: I am conducting an audit of metadata tests. >> >> A particular manifestation of IMPALA-3947 [0] is that the test module >> test_ddl_stress.py isn't runnable via any buildall.sh option. I expect >> this means test_ddl_stress.py isn't being run regularly except by >> conscientious developers running it locally. Note that this test was >> not disabled as part of the commit "IMPALA-2605: Omit the sort and >> mini stress tests". >> >> I'm interested to know any history around this. >> >> Could any community members who have been around for several years, >> especially those who work on the Catalog, shed light on this? Were you >> assuming test_ddl_stress.py was being run regularly? Or do you know to >> run it by hand? Do you know it to be reliable and useful, and it >> should be included in regular runs, or do you know it to be flaky and >> needing work? >> >> It's fine if no one knows any such answers, but they would lend >> context to getting test_ddl_stress running regularly. >> >> Thanks. >> >> [0] https://issues.cloudera.org/browse/IMPALA-3947 >>
