[
https://issues.apache.org/jira/browse/PHOENIX-6107?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17185630#comment-17185630
]
Lars Hofhansl edited comment on PHOENIX-6107 at 8/27/20, 6:22 AM:
------------------------------------------------------------------
The entire suite finished in less then 2h. It's definitely worth approaching
this.
I'd really like to discuss whether we can reduce the coverage this way. What
we'd lose is all query tests with an index with DESC sorting and some covering,
but we have a lot of index tests to cover those.
On top of this we can put classes together and even save more time.
Update...just saw this:
{quote}Most of the tests modify the table contents in some way{quote}
Meh, shame. :(
was (Author: lhofhansl):
The entire suite finished in less then 2h. It's definitely worth approaching
this.
I'd really like to discuss whether we can reduce the coverage this way. What
we'd lose is all query tests with an index with DESC sorting and some covering,
but we have a lot of index tests to cover those.
On top of this we can put classes together and even save more time.
> Discuss speed up of BaseQueryIT
> -------------------------------
>
> Key: PHOENIX-6107
> URL: https://issues.apache.org/jira/browse/PHOENIX-6107
> Project: Phoenix
> Issue Type: Wish
> Reporter: Lars Hofhansl
> Priority: Major
> Attachments: 6107-master.txt, 6107-proposal.txt
>
>
> All 14 tests derived from BaseQueryIT are some of the slowest we have.
> I noticed that all these tests run 7 times and each time create a table and 6
> indexes.
> So just in terms of setup there are 14*7 = 98 tables created and 14*7*6 = 588
> indexes created.
> It's not clear to me that that the runtime is justified, especially since we
> have so many other index ITs.
> I think we can reduce this to run with one global index and one local index,
> for a repeat of only 3 times, instead of 7. That would benefit all derived
> test and shave of probably around 50% of the overall Phoenix test runtime.
> I.e. 14*3 = 42 tables, and 14*3*2 = 84 indexes.
> Could even go as far and test with no indexes here.
> Yes, it would potentially reduce coverage. Hence a discussion.
> Thoughts?
> (Marked as "Wish" so that we can discuss)
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
