hawq-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Feng Tian <ft...@vitessedata.com>
Subject Re: hawq init?
Date Mon, 05 Oct 2015 15:12:37 GMT
Thank you, very helpful.
On Oct 5, 2015 8:04 AM, "Caleb Welton" <cwelton@pivotal.io> wrote:

> Replying to user@hawq.incubator.apache.org, moving dev list to bcc.
>
> Feng,
>
>   In GPHOME there is a file Edit $GPHOME/etc/slaves. This file should have
> all segment host names, for example: sdw1 sdw2 sdw3.  Once populated if
> your run `hawq init cluster` it will install hawq segments on each of these
> nodes.
>
>   You can also add additional segments after initial cluster initialization
> via `hawq init segment`.  This operation can be done online and the segment
> will register itself with the master automatically.
>
>   It is recommended that after adding initializing a new segment that you
> also add this segment to the  $GPHOME/etc/slaves file so that the segment
> will automatically be started again on if you `hawq stop cluster`/`hawq
> start cluster`.
>
> Lei & the dev team,
>
>   The HAWQ 2.0 quickstart documentation has been publicly posted yet... we
> should probably fix that so that there are clear instructions for new users
> trying out the project for the first time.
>
> Regards,
>   Caleb
>
> On Sun, Oct 4, 2015 at 7:48 AM, Feng Tian <ftian@vitessedata.com> wrote:
>
> > sorry, so, one seg per node.   does the node get defined in
> > hawq-site.xml?    just add nodes there?
> > On Oct 3, 2015 10:46 PM, "Feng Tian" <ftian@vitessedata.com> wrote:
> >
> > > interesting.   what if i have two physical nodes?
> > >
> > > and, how can i control the number of forked processes?   do i have to
> > > believe planner will pickup the right number?   any pointer to design
> > > and/or implementation?
> > >
> > > thanks,
> > > On Oct 3, 2015 8:02 PM, "Zhanwei Wang" <zwang@pivotal.io> wrote:
> > >
> > >> Hi Feng
> > >>
> > >> Unlike Greenplum Database and HAWQ 1.x. Apache HAWQ will init only one
> > >> segment on each node.
> > >>
> > >> But Apache HAWQ can fork many work processes on one segment to execute
> > the
> > >> plan. The number of work processes depends on the query and plan.
> > >>
> > >> Best Regards
> > >> Zhanwei Wang
> > >>
> > >>
> > >> On Sun, Oct 4, 2015 at 12:39 AM, Feng Tian <ftian@vitessedata.com>
> > wrote:
> > >>
> > >> > Hi,
> > >> >
> > >> > Hawq init cluster on my box works great, but it only creates one
> > >> segment.
> > >> > How can I init a cluster of say, two segments?  Which config file
> > >> should I
> > >> > edit?
> > >> >
> > >> > Tried gpinitsystem (does not work), tried google but cannot find
> much
> > >> info.
> > >> >
> > >> > Thanks,
> > >> >
> > >>
> > >>
> > >>
> > >> --
> > >> Best Regards
> > >> ----------
> > >>
> > >> Zhanwei Wang
> > >>
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message