[
https://issues.apache.org/jira/browse/BIGTOP-966?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13807144#comment-13807144
]
jay vyas commented on BIGTOP-966:
---------------------------------
Hmmm... its a good idea to simplify the way the tests run , but for those of us
not running oozie etc, it would be a problem.
How about, instead, follow the sqoop model, where integration tests are
separate. If your integration tests are a super set of the low level tests,
you can simply update the maven file to only run your integration tests, while
others only run lower level hive specific tests.
https://github.com/apache/bigtop/tree/master/bigtop-tests/test-artifacts/sqoop/src/main/groovy/org/apache/bigtop/itest/integration/sqoop
That way the smokes arent assuming anything about underlying deployment.
I think in general that the BEST bigtop tests are the mapreduce and mahout
ones, becuase they are very simple to run and dont expect any higher level
services to be running.
> Consider changing Hive tests to use Thrift Java client, JDBC client or Oozie
> workflows
> --------------------------------------------------------------------------------------
>
> Key: BIGTOP-966
> URL: https://issues.apache.org/jira/browse/BIGTOP-966
> Project: Bigtop
> Issue Type: Improvement
> Components: Tests
> Affects Versions: 0.6.0
> Reporter: Mark Grover
> Assignee: Anatoli Fomenko
> Priority: Blocker
> Fix For: 0.8.0
>
>
> Presently, we run one of our Hive integration tests via the JDBC driver and
> others via the Hive CLI. While JDBC is great, the CLI tests can be finicky
> since they are linked to the representation of the results.
> We should consider changing Hive tests to use the Thrift Java client or the
> JDBC client or even better Oozie works for interacting with Hive driver. That
> way the output is deterministic and same regardless of presentation.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
