hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrew Purtell <apurt...@apache.org>
Subject Re: State of protobufs in hbase-0.96
Date Tue, 27 Nov 2012 23:26:00 GMT
KeyValue is a special case. They're not really Writables either, in that
readFields() and writeFields() are not used half the time for
marshalling/unmarshalling.


On Tue, Nov 27, 2012 at 1:46 PM, Jimmy Xiang <jxiang@cloudera.com> wrote:

> We can't completely purge Writables since we are still using KV in HFile.
>
> Are we going to convert the file format too?  If so, we need some
> migration strategy.
>
> Thanks,
> Jimmy
>
> On Tue, Nov 27, 2012 at 1:41 PM, Gary Helmling <ghelmling@gmail.com>
> wrote:
> > Yep, the initial conversion work (HBASE-5448) only deprecated
> > CoprocessorProtocol, but in subsequent discussion I think the universal
> > consensus was to remove it completely for 0.96 and go PB-only.  Otherwise
> > we can't completely purge Writables for 0.96.
> >
> >
> >
> > On Tue, Nov 27, 2012 at 1:37 PM, Andrew Purtell <apurtell@apache.org>
> wrote:
> >
> >> We agreed to remove Writables from coprocessors and go PB-only for 0.96.
> >> This way, to avoid building up any transitional cruft.
> >>
> >>
> >> On Tue, Nov 27, 2012 at 1:29 PM, lars hofhansl <lhofhansl@yahoo.com>
> >> wrote:
> >>
> >> > I thought for coprocessors we had decided to only deprecate the old
> >> > protocol. I'm +1 removing it, though, because it let's us potentially
> get
> >> > rid of all Writables.
> >> >
> >> >
> >> >
> >> > ________________________________
> >> >  From: Gary Helmling <ghelmling@gmail.com>
> >> > To: "dev@hbase.apache.org" <dev@hbase.apache.org>; lars hofhansl
<
> >> > lhofhansl@yahoo.com>
> >> > Sent: Tuesday, November 27, 2012 11:45 AM
> >> > Subject: Re: State of protobufs in hbase-0.96
> >> >
> >> >
> >> > For coprocessor endpoints, we have
> >> > https://issues.apache.org/jira/browse/HBASE-6895 as an umbrella issue
> >> for
> >> > removing the Writable based CoprocessorProtocol support.
> >> >
> >> > There are still a few subtasks to work through to get there, but once
> >> done
> >> > we can pull out CoprocessorProtocol, Exec, ExecResult,
> >> > HTable.coprocessorProxy(), HTable.coprocessorExec() and any other
> >> > supporting code.
> >> >
> >> >
> >> >
> >> > On Tue, Nov 27, 2012 at 10:54 AM, lars hofhansl <lhofhansl@yahoo.com>
> >> > wrote:
> >> >
> >> > HBASE-7215 has a rant about the current state of the protobuf work in
> >> > 0.96... Continuing here...
> >> > >
> >> > >HBASE-7215 came about because I simply wanted to add a field to
> Mutation
> >> > (Put, Delete, etc) for HBASE-5954.
> >> > >
> >> > >Should be easy now with protobufs, right? Nope!
> >> > >
> >> > >- Both Put and Delete (and Result, Action, MultiAction,
> MultiResponse,
> >> > MultiPut, MultiResponse, etc, etc, etc) are still implementing
> Writable
> >> and
> >> > are still used that way.
> >> > >- After I dug in I found that Writable is literally all over the
> place
> >> > still.
> >> > >- In some cases we even serialize an Object as Writable inside a
> >> protobuf
> >> > message.
> >> > >- HBaseObjectWritable is still around and still used all over the
> place
> >> > >
> >> > >So what we have now has Writables and Protobuf code, worse than it
> was
> >> > before (lots of extra code, two ways to serialize stuff, and still no
> >> wire
> >> > compatibility).
> >> > >
> >> > >HBASE-7215 will fix some of the stuff, but there's a lot more to do.
> >> > >
> >> > >-- Lars
> >> > >
> >>
> >>
> >>
> >>
> >> --
> >> Best regards,
> >>
> >>    - Andy
> >>
> >> Problems worthy of attack prove their worth by hitting back. - Piet Hein
> >> (via Tom White)
> >>
>



-- 
Best regards,

   - Andy

Problems worthy of attack prove their worth by hitting back. - Piet Hein
(via Tom White)

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