Fwd: Joins issues

[Viacheslav Inozemtsev <vi...@visual-meta.com>]

Hello,

at the Visual Meta company we are trying to use Phoenix right now. And we
are getting problems with joins. We have a cluster of 6 nodes. The versions
we are using are:

Hadoop 2.7.1
HBase 0.98
Phoenix 4.5

So, the problem is with the regular join, that has post-condition on the
columns of type *date*. If we do it without any hint, then hash join is
being executed, and we need enough cache memory for that, otherwise it just
fails with exception *MaxServerCacheSizeExceededException*. We increased
cache size and it worked after that. If we specify a hint to use merge sort
join, it fails right away with the exception *IndexOutOfBoundsException*.
Could you please explain us, what's the problem? How can we make merge sort
join work? Why is hash join used by default? Can we use merge sort join by
default, and hash join only if there's enough cache memory?

I attached three files to this email with our example schema, hash join
query and merge sort join query.

We would appreciate very much any help.

Thank you in advance!

Best regards,
Viacheslav Inozemtsev

-- 

Viacheslav Inozemtsev
Backend Developer

Visual Meta GmbH, Schützenstraße 25, 10117 Berlin
Ein Unternehmen der Axel Springer SE
www.visual-meta.com

E-Mail: viacheslav.inozemtsev@visual-meta.com

Geschäftsführer: Robert M. Maier, Johannes Schaback, Johannes Kotte
Registernummer: HRB 115795 B
Registergericht: Amtsgericht Charlottenburg

Re: Joins issues

[Ayola Jayamaha <ra...@gmail.com>]

Hi,

I too have noticed that inner-join fails to give results.

On Wed, Sep 16, 2015 at 3:42 PM, Viacheslav Inozemtsev <
viacheslav.inozemtsev@visual-meta.com> wrote:

> Hello,
>
> at the Visual Meta company we are trying to use Phoenix right now. And we
> are getting problems with joins. We have a cluster of 6 nodes. The versions
> we are using are:
>
> Hadoop 2.7.1
> HBase 0.98
> Phoenix 4.5
>
> So, the problem is with the regular join, that has post-condition on the
> columns of type *date*. If we do it without any hint, then hash join is
> being executed, and we need enough cache memory for that, otherwise it just
> fails with exception *MaxServerCacheSizeExceededException*. We increased
> cache size and it worked after that. If we specify a hint to use merge sort
> join, it fails right away with the exception *IndexOutOfBoundsException*.
> Could you please explain us, what's the problem? How can we make merge sort
> join work? Why is hash join used by default? Can we use merge sort join by
> default, and hash join only if there's enough cache memory?
>
> I attached three files to this email with our example schema, hash join
> query and merge sort join query.
>
> We would appreciate very much any help.
>
> Thank you in advance!
>
> Best regards,
> Viacheslav Inozemtsev
>
> --
>
> Viacheslav Inozemtsev
> Backend Developer
>
> Visual Meta GmbH, Schützenstraße 25, 10117 Berlin
> Ein Unternehmen der Axel Springer SE
> www.visual-meta.com
>
> E-Mail: viacheslav.inozemtsev@visual-meta.com
>
> Geschäftsführer: Robert M. Maier, Johannes Schaback, Johannes Kotte
> Registernummer: HRB 115795 B
> Registergericht: Amtsgericht Charlottenburg
>
>


-- 
Best Regards,
Nishani Jayamaha
http://ayolajayamaha.blogspot.com/