accumulo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dylan Hutchison <dyhutchi...@gmail.com>
Subject Re: LocalityGroupDeleter
Date Fri, 15 Jul 2016 21:21:02 GMT
I see. Pass table name via IteratorOptions? Or just pass locality group
column families?

On Jul 15, 2016 3:03 PM, "Russ Weeks" <rweeks@newbrightidea.com> wrote:

> Hey, Dylan,
>
> Yeah, I thought about that but locality group info is kept at the table
> level (makes sense) and I don't see any way for an iterator to know what
> table it's operating on (which seems weird but...)
>
> On Fri, Jul 15, 2016 at 2:01 PM Dylan Hutchison <
> dhutchis@cs.washington.edu>
> wrote:
>
> > For accessing locality group column family information inside an Accumulo
> > iterator, what about querying the Accumulo table property table.group.*
> > <
> >
> https://accumulo.apache.org/1.7/accumulo_user_manual#TABLE_LOCALITY_GROUP_PREFIX
> > >
> > from IteratorEnvironment.getConfig() in the iterator's init()?  I haven't
> > tested this, but it seems like it would work.
> >
> > On Fri, Jul 15, 2016 at 1:54 PM, Russ Weeks <rweeks@newbrightidea.com>
> > wrote:
> >
> > > Hey, folks,
> > >
> > > Yesterday I asked on IRC about whether there was an optimized way to
> > delete
> > > all key/value pairs in a locality group. Keith had the great suggestion
> > > that, since compactions are done separately for each locality group, I
> > > could write a custom iterator that returned false from hasTop if the
> > > iterator was seeked with the right column families.
> > >
> > > If it's useful to anyone else, the code is in this gist:
> > > https://gist.github.com/rweeks/bedb96e20fcb8d2d2f27c555494181c5 and it
> > > works pretty well, thanks Keith!
> > >
> > > It would be more convenient for the end-user to configure the iterator
> > with
> > > a specific locality group, instead of a set of column families that
> must
> > > match a locality group for the iterator to work correctly.
> Unfortunately
> > > the iterator doesn't have the necessary context to map a locality group
> > > name to the CFs in that group (as far as I can tell).
> > >
> > > -Russ
> > >
> >
>

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