[
https://issues.apache.org/jira/browse/PHOENIX-674?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13979840#comment-13979840
]
Gabriel Reid commented on PHOENIX-674:
--------------------------------------
Cool, I've just been playing around with this and it seems to be a big
improvement.
The current setting of 0.5C for forkCount seems very conservative to me. The
integration tests are not at all CPU bound as far as I can see, so I don't
think that we need to take the number of CPUs into account. Unfortunately, that
seems to be the only heuristic available in failsafe, so I would be more in
favour of just using a property with a reasonable default (maybe 2 or 4?),
which would then allow overriding it with a -D option for people who have a
resource-starved setup.
The overhead of running the unit tests is so small I almost wonder if this is
worth doing there -- I think that the biggest resource hog there is maven
itself.
Once additional thing that will be needed before enabling this is setting the
hbase.master.info.port property to -1 wherever an HBaseTestingUtility is used
(or in an hbase-site.xml on the classpath). As it is, concurrent tests (such as
under phoenix-pig) clash on the default master info port and then fail.
> Make unit test suite run faster
> -------------------------------
>
> Key: PHOENIX-674
> URL: https://issues.apache.org/jira/browse/PHOENIX-674
> Project: Phoenix
> Issue Type: Task
> Reporter: James Taylor
> Assignee: Samarth Jain
> Attachments: fasttests_v1.patch
>
>
> The main reason for the slowness is the time it takes to spin up and tear
> down a mini cluster for each test suite. We do this to guarantee that there's
> no left over state from previous tests.
> A few things we can do to improve things:
> * consolidate test suites where it makes sense, if there are just a few
> strangler tests that can put combined with other test suites.
> * spin up a single mini cluster and before a test suite runs, disable and
> drop any hbase tables that exist.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
