The HiveExternalCatalogVersionsSuite is hard to make robust as it downloads several huge Spark archives. It does try several mirrors and fall back to archive.apache, but, still, plenty of scope for occasional errors. We need to keep this restricted to only testing a few recent Spark versions.
On Sun, Feb 10, 2019 at 2:01 PM Felix Cheung <felixcheun...@hotmail.com> wrote: > > +1 > See note > > Tested build from source and running tests. > Also tested SparkR basic - ran more tests in RC1 and checked there was no > change in R since. So I’m ok with that. > > Note: > 1. Opened https://issues.apache.org/jira/browse/SPARK-26855 on the > SparkSubmitSuite failure - (thanks to Sean’s tip) I don’t think it’s blocker. > > 2. Ran into a failure in HiveExternalCatalogVersionsSuite. But passed the 2nd > ran (How reliable is archive.apache? It failed for > me before) > WARN org.apache.spark.sql.hive.HiveExternalCatalogVersionsSuite: Failed to > download Spark 2.3.2 from > https://archive.apache.org/dist/spark/spark-2.3.2/spark-2.3.2-bin-hadoop2.7.tgz: > Socket closed > org.apache.spark.sql.hive.HiveExternalCatalogVersionsSuite *** ABORTED *** > Exception encountered when invoking run on a nested suite - Unable to > download Spark 2.3.2 (HiveExternalCatalogVersionsSuite.scala:97) > > 3. There are a fair bit of changes in Python and SQL - someone should test > that > > 4. Last time k8s integration tests is broken before it isn’t built by > default. Could someone test with -Pkubernetes -Pkubernetes-integration-tests > > SPARK-26482 broke the integration tests > --------------------------------------------------------------------- To unsubscribe e-mail: dev-unsubscr...@spark.apache.org
