Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Drill, mail # dev - Comments on pipelined Merge Join approach


Copy link to this message
-
Re: Comments on pipelined Merge Join approach
Jason Altekruse 2013-08-12, 22:55
I'm not sure which join type is being implemented, but I think that some
output records might be missing.

In the JoinTemplate.doJoin method, the case form left and right matching
prompts a copy and advance of both pointers. Do we want to create a new
record for each of the rights that matches a particular left?

[image: Inline image 1]
On Fri, Aug 9, 2013 at 12:55 PM, Jacques Nadeau <[EMAIL PROTECTED]> wrote:

> yes
>
> On Fri, Aug 9, 2013 at 12:49 PM, Timothy Chen <[EMAIL PROTECTED]> wrote:
> > Might be speaking the obvious, but both sides are sorted on the joined
> > columns right?
> >
> > Tim
> >
> >
> > On Fri, Aug 9, 2013 at 12:26 PM, Jacques Nadeau <[EMAIL PROTECTED]>
> wrote:
> >
> >> I'm working on the first implementation of a merge join for the full
> >> exec engine.  My WIP branch is on my personal repo.  Would love logic
> >> buffs to give me feedback on the flow if you have any time today.
> >> I'll be working on pulling this together through the weekend.
> >>
> >> https://github.com/jacques-n/incubator-drill/tree/join-wip
> >>
> >> Thanks,
> >> Jacques
> >>
>