Since 0.96 is a singularity, is it a time to deprecate thrift finally?

Or we can use thrift2 by default, and keep thrift around more?

What's the pros and cons to switch to thrift2?

Thanks,
Jimmy

On Thu, Dec 6, 2012 at 3:24 PM, Jimmy Xiang <[EMAIL PROTECTED]> wrote:

> Since 0.96 is a singularity, is it a time to deprecate thrift finally?
>
> Or we can use thrift2 by default, and keep thrift around more?
>
> What's the pros and cons to switch to thrift2?
>

Interesting notion.  Could move thrift2 to thrift altogether.  Would need
to make sure thrift2 has all from thrift1.
St.Ack

According to the thrift2 docs, it is the 'right' way to access HBase - the
thrift package is based on an older HBase interface that no longer exists.
Seems like a reasonable case for deprecation in the next release and
dropping it all together after.
-------------------
Jesse Yates
@jesse_yates
jyates.github.com
On Thu, Dec 6, 2012 at 4:48 PM, Stack <[EMAIL PROTECTED]> wrote:

> On Thu, Dec 6, 2012 at 3:24 PM, Jimmy Xiang <[EMAIL PROTECTED]> wrote:
>
> > Since 0.96 is a singularity, is it a time to deprecate thrift finally?
> >
> > Or we can use thrift2 by default, and keep thrift around more?
> >
> > What's the pros and cons to switch to thrift2?
> >
>
> Interesting notion.  Could move thrift2 to thrift altogether.  Would need
> to make sure thrift2 has all from thrift1.
> St.Ack
>

+1 for deprecation. Will bring any users out of the woodwork who might not
see this discussion at least.

On Friday, December 7, 2012, Jesse Yates wrote:

> According to the thrift2 docs, it is the 'right' way to access HBase - the
> thrift package is based on an older HBase interface that no longer exists.
> Seems like a reasonable case for deprecation in the next release and
> dropping it all together after.
> -------------------
> Jesse Yates
> @jesse_yates
> jyates.github.com
>
>
> On Thu, Dec 6, 2012 at 4:48 PM, Stack <[EMAIL PROTECTED] <javascript:;>>
> wrote:
>
> > On Thu, Dec 6, 2012 at 3:24 PM, Jimmy Xiang <[EMAIL PROTECTED]<javascript:;>>
> wrote:
> >
> > > Since 0.96 is a singularity, is it a time to deprecate thrift finally?
> > >
> > > Or we can use thrift2 by default, and keep thrift around more?
> > >
> > > What's the pros and cons to switch to thrift2?
> > >
> >
> > Interesting notion.  Could move thrift2 to thrift altogether.  Would need
> > to make sure thrift2 has all from thrift1.
> > St.Ack
> >
>
--
Best regards,

   - Andy

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

The embeded HRegionThriftServer should be ported to thrift2 before we
deprecate thrift1. FB's (and potentially others) c++ client requires
HRegionThriftServer.

__Luke
On Thu, Dec 6, 2012 at 7:21 PM, Andrew Purtell <[EMAIL PROTECTED]> wrote:

> +1 for deprecation. Will bring any users out of the woodwork who might not
> see this discussion at least.
>
> On Friday, December 7, 2012, Jesse Yates wrote:
>
> > According to the thrift2 docs, it is the 'right' way to access HBase -
> the
> > thrift package is based on an older HBase interface that no longer
> exists.
> > Seems like a reasonable case for deprecation in the next release and
> > dropping it all together after.
> > -------------------
> > Jesse Yates
> > @jesse_yates
> > jyates.github.com
> >
> >
> > On Thu, Dec 6, 2012 at 4:48 PM, Stack <[EMAIL PROTECTED] <javascript:;>>
> > wrote:
> >
> > > On Thu, Dec 6, 2012 at 3:24 PM, Jimmy Xiang <[EMAIL PROTECTED]
> <javascript:;>>
> > wrote:
> > >
> > > > Since 0.96 is a singularity, is it a time to deprecate thrift
> finally?
> > > >
> > > > Or we can use thrift2 by default, and keep thrift around more?
> > > >
> > > > What's the pros and cons to switch to thrift2?
> > > >
> > >
> > > Interesting notion.  Could move thrift2 to thrift altogether.  Would
> need
> > > to make sure thrift2 has all from thrift1.
> > > St.Ack
> > >
> >
>
>
> --
> Best regards,
>
>    - Andy
>
> Problems worthy of attack prove their worth by hitting back. - Piet Hein
> (via Tom White)
>

On Fri, Dec 7, 2012 at 1:56 PM, Luke Lu <[EMAIL PROTECTED]> wrote:

> The embeded HRegionThriftServer should be ported to thrift2 before we
> deprecate thrift1. FB's (and potentially others) c++ client requires
> HRegionThriftServer.
>

HRTS is an abandoned experiment as I understand it so I closed HBASE-7302,
the issue that would perpetuate it in thrift2.
St.Ack

It's quite a useful experiment, though, standing up separate thrift servers adds more complexity to a deployment.

If it is really abandoned we should just rip it out, no? Otherwise I'd think at least in principle that is useful (also considering potential future enhancement that we do knowing the thrift server is in process with the RegionServer).

-- Lars

________________________________
 From: Stack <[EMAIL PROTECTED]>
To: HBase Dev List <[EMAIL PROTECTED]>
Sent: Friday, December 7, 2012 3:16 PM
Subject: Re: thrift vs thrift2
 
On Fri, Dec 7, 2012 at 1:56 PM, Luke Lu <[EMAIL PROTECTED]> wrote:

> The embeded HRegionThriftServer should be ported to thrift2 before we
> deprecate thrift1. FB's (and potentially others) c++ client requires
> HRegionThriftServer.
>

HRTS is an abandoned experiment as I understand it so I closed HBASE-7302,
the issue that would perpetuate it in thrift2.
St.Ack

IIRC the FB guys concluded the reverse, that it's better to provision
Thrift endpoints as a tier separate from RegionServers, so there can be
caching there, so the thrift servers in the RS don't bottleneck, etc.

On Saturday, December 8, 2012, lars hofhansl wrote:

> It's quite a useful experiment, though, standing up separate thrift
> servers adds more complexity to a deployment.
>
> If it is really abandoned we should just rip it out, no? Otherwise I'd
> think at least in principle that is useful (also considering potential
> future enhancement that we do knowing the thrift server is in process with
> the RegionServer).
>
> -- Lars
>
>
>
> ________________________________
>  From: Stack <[EMAIL PROTECTED] <javascript:;>>
> To: HBase Dev List <[EMAIL PROTECTED] <javascript:;>>
> Sent: Friday, December 7, 2012 3:16 PM
> Subject: Re: thrift vs thrift2
>
> On Fri, Dec 7, 2012 at 1:56 PM, Luke Lu <[EMAIL PROTECTED] <javascript:;>>
> wrote:
>
> > The embeded HRegionThriftServer should be ported to thrift2 before we
> > deprecate thrift1. FB's (and potentially others) c++ client requires
> > HRegionThriftServer.
> >
>
> HRTS is an abandoned experiment as I understand it so I closed HBASE-7302,
> the issue that would perpetuate it in thrift2.
> St.Ack

--
Best regards,

   - Andy

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

Hmm... Interesting.Maybe the implementation is bad, but principally I don't see that. Why cache data twice when the in process thrift server could just access the data locally from the region server cache anyway?

Anyway, either we should fix it or remove it. (just to state the obvious :) )
I am not a fan of keeping non-functional code around in the hope that it might eventually get fixed.
-- Lars
________________________________
 From: Andrew Purtell <[EMAIL PROTECTED]>
To: "[EMAIL PROTECTED]" <[EMAIL PROTECTED]>; lars hofhansl <[EMAIL PROTECTED]>
Sent: Friday, December 7, 2012 9:06 PM
Subject: Re: thrift vs thrift2
 
IIRC the FB guys concluded the reverse, that it's better to provision
Thrift endpoints as a tier separate from RegionServers, so there can be
caching there, so the thrift servers in the RS don't bottleneck, etc.

On Saturday, December 8, 2012, lars hofhansl wrote:

> It's quite a useful experiment, though, standing up separate thrift
> servers adds more complexity to a deployment.
>
> If it is really abandoned we should just rip it out, no? Otherwise I'd
> think at least in principle that is useful (also considering potential
> future enhancement that we do knowing the thrift server is in process with
> the RegionServer).
>
> -- Lars
>
>
>
> ________________________________
>  From: Stack <[EMAIL PROTECTED] <javascript:;>>
> To: HBase Dev List <[EMAIL PROTECTED] <javascript:;>>
> Sent: Friday, December 7, 2012 3:16 PM
> Subject: Re: thrift vs thrift2
>
> On Fri, Dec 7, 2012 at 1:56 PM, Luke Lu <[EMAIL PROTECTED] <javascript:;>>
> wrote:
>
> > The embeded HRegionThriftServer should be ported to thrift2 before we
> > deprecate thrift1. FB's (and potentially others) c++ client requires
> > HRegionThriftServer.
> >
>
> HRTS is an abandoned experiment as I understand it so I closed HBASE-7302,
> the issue that would perpetuate it in thrift2.
> St.Ack

--
Best regards,

   - Andy

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

Or it's easier to add more external thrift servers than tune up more
internal thrift server threads in a running RS? Speculation.

On Saturday, December 8, 2012, lars hofhansl wrote:

> Hmm... Interesting.Maybe the implementation is bad, but principally I
> don't see that. Why cache data twice when the in process thrift server
> could just access the data locally from the region server cache anyway?
>
> Anyway, either we should fix it or remove it. (just to state the obvious
> :) )
> I am not a fan of keeping non-functional code around in the hope that it
> might eventually get fixed.
>
>
> -- Lars
>
>
> ________________________________
>  From: Andrew Purtell <[EMAIL PROTECTED] <javascript:;>>
> To: "[EMAIL PROTECTED] <javascript:;>" <[EMAIL PROTECTED]<javascript:;>>;
> lars hofhansl <[EMAIL PROTECTED] <javascript:;>>
> Sent: Friday, December 7, 2012 9:06 PM
> Subject: Re: thrift vs thrift2
>
> IIRC the FB guys concluded the reverse, that it's better to provision
> Thrift endpoints as a tier separate from RegionServers, so there can be
> caching there, so the thrift servers in the RS don't bottleneck, etc.
>
> On Saturday, December 8, 2012, lars hofhansl wrote:
>
> > It's quite a useful experiment, though, standing up separate thrift
> > servers adds more complexity to a deployment.
> >
> > If it is really abandoned we should just rip it out, no? Otherwise I'd
> > think at least in principle that is useful (also considering potential
> > future enhancement that we do knowing the thrift server is in process
> with
> > the RegionServer).
> >
> > -- Lars
> >
> >
> >
> > ________________________________
> >  From: Stack <[EMAIL PROTECTED] <javascript:;> <javascript:;>>
> > To: HBase Dev List <[EMAIL PROTECTED] <javascript:;> <javascript:;>>
> > Sent: Friday, December 7, 2012 3:16 PM
> > Subject: Re: thrift vs thrift2
> >
> > On Fri, Dec 7, 2012 at 1:56 PM, Luke Lu <[EMAIL PROTECTED] <javascript:;><javascript:;>>
> > wrote:
> >
> > > The embeded HRegionThriftServer should be ported to thrift2 before we
> > > deprecate thrift1. FB's (and potentially others) c++ client requires
> > > HRegionThriftServer.
> > >
> >
> > HRTS is an abandoned experiment as I understand it so I closed
> HBASE-7302,
> > the issue that would perpetuate it in thrift2.
> > St.Ack
>
>
>
> --
> 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)

Looks like there were activities on HRegionThriftServer.java in 0.89-fb
branch till recently:

r1391221 | mbautin | 2012-09-27 14:07:52 -0700 (Thu, 27 Sep 2012) | 15 lines

[jira] [HBASE-6882] [89-fb] Thrift IOError should include exception class

Author: mbautin

Summary: Return exception class as part of IOError thrown from the Thrift
proxy or the embedded Thrift server in the regionserver.

Test Plan:
Unit tests
Test through C++ HBase client

Reviewers: liyintang, kranganathan, aaiyer, chip

Reviewed By: liyintang

Differential Revision: https://reviews.facebook.net/D5679
------------------------------------------------------------------------
r1388183 | mbautin | 2012-09-20 12:47:23 -0700 (Thu, 20 Sep 2012) | 15 lines

[HBASE-6836] [89-fb] Parallel deletes in HBase Thrift server

Author: mbautin

Summary: We need to expose server-side parallel batch deletes through the
Thrift server.

Test Plan:
Unit tests
C++ client test

Reviewers: aaiyer, nzhang, kranganathan

Reviewed By: aaiyer

Differential Revision: https://reviews.facebook.net/D5541
------------------------------------------------------------------------
r1387433 | mbautin | 2012-09-18 18:52:58 -0700 (Tue, 18 Sep 2012) | 15 lines

[jira] [HBASE-6808] [89-fb] Add the ability to query region flush time and
current time to regionserver Thrift API

On Fri, Dec 7, 2012 at 9:06 PM, Andrew Purtell <[EMAIL PROTECTED]> wrote:

> IIRC the FB guys concluded the reverse, that it's better to provision
> Thrift endpoints as a tier separate from RegionServers, so there can be
> caching there, so the thrift servers in the RS don't bottleneck, etc.
>
> On Saturday, December 8, 2012, lars hofhansl wrote:
>
> > It's quite a useful experiment, though, standing up separate thrift
> > servers adds more complexity to a deployment.
> >
> > If it is really abandoned we should just rip it out, no? Otherwise I'd
> > think at least in principle that is useful (also considering potential
> > future enhancement that we do knowing the thrift server is in process
> with
> > the RegionServer).
> >
> > -- Lars
> >
> >
> >
> > ________________________________
> >  From: Stack <[EMAIL PROTECTED] <javascript:;>>
> > To: HBase Dev List <[EMAIL PROTECTED] <javascript:;>>
> > Sent: Friday, December 7, 2012 3:16 PM
> > Subject: Re: thrift vs thrift2
> >
> > On Fri, Dec 7, 2012 at 1:56 PM, Luke Lu <[EMAIL PROTECTED] <javascript:;>>
> > wrote:
> >
> > > The embeded HRegionThriftServer should be ported to thrift2 before we
> > > deprecate thrift1. FB's (and potentially others) c++ client requires
> > > HRegionThriftServer.
> > >
> >
> > HRTS is an abandoned experiment as I understand it so I closed
> HBASE-7302,
> > the issue that would perpetuate it in thrift2.
> > St.Ack
>
>
>
> --
> Best regards,
>
>    - Andy
>
> Problems worthy of attack prove their worth by hitting back. - Piet Hein
> (via Tom White)
>

On Fri, Dec 7, 2012 at 9:12 PM, Ted Yu <[EMAIL PROTECTED]> wrote:

> Looks like there were activities on HRegionThriftServer.java in 0.89-fb
> branch till recently:
>
Here is what Karthik had to say a while back
http://search-hadoop.com/m/GAFt6B9CRZ/joe+pallas+karthik&subj=Re+Thrift2+interface

As I read it, we should purge the code as a failed approach unless someone
wants to take on this effort.

St.Ack

With rpc moving to protobuf, we should get rid of thrift from HBase core
completely :)
On Fri, Dec 7, 2012 at 11:31 PM, Stack <[EMAIL PROTECTED]> wrote:

> On Fri, Dec 7, 2012 at 9:12 PM, Ted Yu <[EMAIL PROTECTED]> wrote:
>
> > Looks like there were activities on HRegionThriftServer.java in 0.89-fb
> > branch till recently:
> >
>
>
> Here is what Karthik had to say a while back
>
> http://search-hadoop.com/m/GAFt6B9CRZ/joe+pallas+karthik&subj=Re+Thrift2+interface
>
> As I read it, we should purge the code as a failed approach unless someone
> wants to take on this effort.
>
> St.Ack
>

+1
On Fri, Dec 7, 2012 at 11:50 PM, Luke Lu <[EMAIL PROTECTED]> wrote:
> With rpc moving to protobuf, we should get rid of thrift from HBase core
> completely :)
>
>
> On Fri, Dec 7, 2012 at 11:31 PM, Stack <[EMAIL PROTECTED]> wrote:
>
>> On Fri, Dec 7, 2012 at 9:12 PM, Ted Yu <[EMAIL PROTECTED]> wrote:
>>
>> > Looks like there were activities on HRegionThriftServer.java in 0.89-fb
>> > branch till recently:
>> >
>>
>>
>> Here is what Karthik had to say a while back
>>
>> http://search-hadoop.com/m/GAFt6B9CRZ/joe+pallas+karthik&subj=Re+Thrift2+interface
>>
>> As I read it, we should purge the code as a failed approach unless someone
>> wants to take on this effort.
>>
>> St.Ack
>>

Depends on what is your definition of "core".

As a separate maven module in the distro, +1.

As ejected to unmaintained status, -1.

I have to imagine the number of users who use thrift to talk to HBase is
nonzero given we've had this option forever. But let's tag everything
Thrift as deprecated and see.

On Sunday, December 9, 2012, Jimmy Xiang wrote:

> +1
>
>
> On Fri, Dec 7, 2012 at 11:50 PM, Luke Lu <[EMAIL PROTECTED]> wrote:
> > With rpc moving to protobuf, we should get rid of thrift from HBase core
> > completely :)
> >
> >
> > On Fri, Dec 7, 2012 at 11:31 PM, Stack <[EMAIL PROTECTED]> wrote:
> >
> >> On Fri, Dec 7, 2012 at 9:12 PM, Ted Yu <[EMAIL PROTECTED]> wrote:
> >>
> >> > Looks like there were activities on HRegionThriftServer.java in
> 0.89-fb
> >> > branch till recently:
> >> >
> >>
> >>
> >> Here is what Karthik had to say a while back
> >>
> >>
> http://search-hadoop.com/m/GAFt6B9CRZ/joe+pallas+karthik&subj=Re+Thrift2+interface
> >>
> >> As I read it, we should purge the code as a failed approach unless
> someone
> >> wants to take on this effort.
> >>
> >> St.Ack
> >>
>
--
Best regards,

   - Andy

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

-1 for removing all together

If removing from core I agree that it should still be offered as a
separate maven module. If there is need for a maintainer of that module
I would be glad to help

-Jake

> Andrew Purtell <mailto:[EMAIL PROTECTED]>
> December 8, 2012 9:15 PM
> Depends on what is your definition of "core".
>
> As a separate maven module in the distro, +1.
>
> As ejected to unmaintained status, -1.
>
> I have to imagine the number of users who use thrift to talk to HBase is
> nonzero given we've had this option forever. But let's tag everything
> Thrift as deprecated and see.
>
>
>
> Jimmy Xiang <mailto:[EMAIL PROTECTED]>
> December 8, 2012 11:59 AM
> +1
>
> Luke Lu <mailto:[EMAIL PROTECTED]>
> December 8, 2012 2:50 AM
> With rpc moving to protobuf, we should get rid of thrift from HBase core
> completely :)
>
>
>
> Stack <mailto:[EMAIL PROTECTED]>
> December 8, 2012 2:31 AM
>
>
> Here is what Karthik had to say a while back
> http://search-hadoop.com/m/GAFt6B9CRZ/joe+pallas+karthik&subj=Re+Thrift2+interface
>
> As I read it, we should purge the code as a failed approach unless someone
> wants to take on this effort.
>
> St.Ack
>
> Ted Yu <mailto:[EMAIL PROTECTED]>
> December 8, 2012 12:12 AM
> Looks like there were activities on HRegionThriftServer.java in 0.89-fb
> branch till recently:
>
> r1391221 | mbautin | 2012-09-27 14:07:52 -0700 (Thu, 27 Sep 2012) | 15
> lines
>
> [jira] [HBASE-6882] [89-fb] Thrift IOError should include exception class
>
> Author: mbautin
>
> Summary: Return exception class as part of IOError thrown from the Thrift
> proxy or the embedded Thrift server in the regionserver.
>
> Test Plan:
> Unit tests
> Test through C++ HBase client
>
> Reviewers: liyintang, kranganathan, aaiyer, chip
>
> Reviewed By: liyintang
>
> Differential Revision: https://reviews.facebook.net/D5679
> ------------------------------------------------------------------------
> r1388183 | mbautin | 2012-09-20 12:47:23 -0700 (Thu, 20 Sep 2012) | 15
> lines
>
> [HBASE-6836] [89-fb] Parallel deletes in HBase Thrift server
>
> Author: mbautin
>
> Summary: We need to expose server-side parallel batch deletes through the
> Thrift server.
>
> Test Plan:
> Unit tests
> C++ client test
>
> Reviewers: aaiyer, nzhang, kranganathan
>
> Reviewed By: aaiyer
>
> Differential Revision: https://reviews.facebook.net/D5541
> ------------------------------------------------------------------------
> r1387433 | mbautin | 2012-09-18 18:52:58 -0700 (Tue, 18 Sep 2012) | 15
> lines
>
> [jira] [HBASE-6808] [89-fb] Add the ability to query region flush time and
> current time to regionserver Thrift API
>
>

I agree with what Jake said.
I can also spend some time maintaining the new maven module.

On Mon, Dec 10, 2012 at 12:36 PM, Jake Farrell <[EMAIL PROTECTED]> wrote:

> -1 for removing all together
>
> If removing from core I agree that it should still be offered as a
> separate maven module. If there is need for a maintainer of that module I
> would be glad to help
>
> -Jake
>
>   Andrew Purtell <[EMAIL PROTECTED]>
>  December 8, 2012 9:15 PM
> Depends on what is your definition of "core".
>
> As a separate maven module in the distro, +1.
>
> As ejected to unmaintained status, -1.
>
> I have to imagine the number of users who use thrift to talk to HBase is
> nonzero given we've had this option forever. But let's tag everything
> Thrift as deprecated and see.
>
>
>
>   Jimmy Xiang <[EMAIL PROTECTED]>
>  December 8, 2012 11:59 AM
> +1
>
>   Luke Lu <[EMAIL PROTECTED]>
>  December 8, 2012 2:50 AM
> With rpc moving to protobuf, we should get rid of thrift from HBase core
> completely :)
>
>
>
>   Stack <[EMAIL PROTECTED]>
>  December 8, 2012 2:31 AM
>
>
> Here is what Karthik had to say a while back
>
> http://search-hadoop.com/m/GAFt6B9CRZ/joe+pallas+karthik&subj=Re+Thrift2+interface
>
> As I read it, we should purge the code as a failed approach unless someone
> wants to take on this effort.
>
> St.Ack
>
>   Ted Yu <[EMAIL PROTECTED]>
>  December 8, 2012 12:12 AM
> Looks like there were activities on HRegionThriftServer.java in 0.89-fb
> branch till recently:
>
> r1391221 | mbautin | 2012-09-27 14:07:52 -0700 (Thu, 27 Sep 2012) | 15
> lines
>
> [jira] [HBASE-6882] [89-fb] Thrift IOError should include exception class
>
> Author: mbautin
>
> Summary: Return exception class as part of IOError thrown from the Thrift
> proxy or the embedded Thrift server in the regionserver.
>
> Test Plan:
> Unit tests
> Test through C++ HBase client
>
> Reviewers: liyintang, kranganathan, aaiyer, chip
>
> Reviewed By: liyintang
>
> Differential Revision: https://reviews.facebook.net/D5679
> ------------------------------------------------------------------------
> r1388183 | mbautin | 2012-09-20 12:47:23 -0700 (Thu, 20 Sep 2012) | 15
> lines
>
> [HBASE-6836] [89-fb] Parallel deletes in HBase Thrift server
>
> Author: mbautin
>
> Summary: We need to expose server-side parallel batch deletes through the
> Thrift server.
>
> Test Plan:
> Unit tests
> C++ client test
>
> Reviewers: aaiyer, nzhang, kranganathan
>
> Reviewed By: aaiyer
>
> Differential Revision: https://reviews.facebook.net/D5541
> ------------------------------------------------------------------------
> r1387433 | mbautin | 2012-09-18 18:52:58 -0700 (Tue, 18 Sep 2012) | 15
> lines
>
> [jira] [HBASE-6808] [89-fb] Add the ability to query region flush time and
> current time to regionserver Thrift API
>
>
>

That post seems to argue for removing all thrift code. Even the standalone thrift server.

________________________________
 From: Stack <[EMAIL PROTECTED]>
To: HBase Dev List <[EMAIL PROTECTED]>
Sent: Friday, December 7, 2012 11:31 PM
Subject: Re: thrift vs thrift2
 
On Fri, Dec 7, 2012 at 9:12 PM, Ted Yu <[EMAIL PROTECTED]> wrote:

> Looks like there were activities on HRegionThriftServer.java in 0.89-fb
> branch till recently:
>
Here is what Karthik had to say a while back
http://search-hadoop.com/m/GAFt6B9CRZ/joe+pallas+karthik&subj=Re+Thrift2+interface

As I read it, we should purge the code as a failed approach unless someone
wants to take on this effort.

St.Ack

Logged https://issues.apache.org/jira/browse/HBASE-7302 for this.

On Fri, Dec 7, 2012 at 1:56 PM, Luke Lu <[EMAIL PROTECTED]> wrote:

> The embeded HRegionThriftServer should be ported to thrift2 before we
> deprecate thrift1. FB's (and potentially others) c++ client requires
> HRegionThriftServer.
>
> __Luke
>
>
> On Thu, Dec 6, 2012 at 7:21 PM, Andrew Purtell <[EMAIL PROTECTED]>
> wrote:
>
> > +1 for deprecation. Will bring any users out of the woodwork who might
> not
> > see this discussion at least.
> >
> > On Friday, December 7, 2012, Jesse Yates wrote:
> >
> > > According to the thrift2 docs, it is the 'right' way to access HBase -
> > the
> > > thrift package is based on an older HBase interface that no longer
> > exists.
> > > Seems like a reasonable case for deprecation in the next release and
> > > dropping it all together after.
> > > -------------------
> > > Jesse Yates
> > > @jesse_yates
> > > jyates.github.com
> > >
> > >
> > > On Thu, Dec 6, 2012 at 4:48 PM, Stack <[EMAIL PROTECTED]<javascript:;>>
> > > wrote:
> > >
> > > > On Thu, Dec 6, 2012 at 3:24 PM, Jimmy Xiang <[EMAIL PROTECTED]
> > <javascript:;>>
> > > wrote:
> > > >
> > > > > Since 0.96 is a singularity, is it a time to deprecate thrift
> > finally?
> > > > >
> > > > > Or we can use thrift2 by default, and keep thrift around more?
> > > > >
> > > > > What's the pros and cons to switch to thrift2?
> > > > >
> > > >
> > > > Interesting notion.  Could move thrift2 to thrift altogether.  Would
> > need
> > > > to make sure thrift2 has all from thrift1.
> > > > St.Ack
> > > >
> > >
> >
> >
> > --
> > Best regards,
> >
> >    - Andy
> >
> > Problems worthy of attack prove their worth by hitting back. - Piet Hein
> > (via Tom White)
> >
>

Since thrift has many users while thrift2 doesn't, does it make more
sense to deprecate and remove thrift2 instead?
People are more used to the existing thrift interface.  Although it is
different from the java API, it seems to not have
any performance hit.

Thanks,
Jimmy

On Fri, Dec 7, 2012 at 2:01 PM, Ted Yu <[EMAIL PROTECTED]> wrote:
> Logged https://issues.apache.org/jira/browse/HBASE-7302 for this.
>
> On Fri, Dec 7, 2012 at 1:56 PM, Luke Lu <[EMAIL PROTECTED]> wrote:
>
>> The embeded HRegionThriftServer should be ported to thrift2 before we
>> deprecate thrift1. FB's (and potentially others) c++ client requires
>> HRegionThriftServer.
>>
>> __Luke
>>
>>
>> On Thu, Dec 6, 2012 at 7:21 PM, Andrew Purtell <[EMAIL PROTECTED]>
>> wrote:
>>
>> > +1 for deprecation. Will bring any users out of the woodwork who might
>> not
>> > see this discussion at least.
>> >
>> > On Friday, December 7, 2012, Jesse Yates wrote:
>> >
>> > > According to the thrift2 docs, it is the 'right' way to access HBase -
>> > the
>> > > thrift package is based on an older HBase interface that no longer
>> > exists.
>> > > Seems like a reasonable case for deprecation in the next release and
>> > > dropping it all together after.
>> > > -------------------
>> > > Jesse Yates
>> > > @jesse_yates
>> > > jyates.github.com
>> > >
>> > >
>> > > On Thu, Dec 6, 2012 at 4:48 PM, Stack <[EMAIL PROTECTED]<javascript:;>>
>> > > wrote:
>> > >
>> > > > On Thu, Dec 6, 2012 at 3:24 PM, Jimmy Xiang <[EMAIL PROTECTED]
>> > <javascript:;>>
>> > > wrote:
>> > > >
>> > > > > Since 0.96 is a singularity, is it a time to deprecate thrift
>> > finally?
>> > > > >
>> > > > > Or we can use thrift2 by default, and keep thrift around more?
>> > > > >
>> > > > > What's the pros and cons to switch to thrift2?
>> > > > >
>> > > >
>> > > > Interesting notion.  Could move thrift2 to thrift altogether.  Would
>> > need
>> > > > to make sure thrift2 has all from thrift1.
>> > > > St.Ack
>> > > >
>> > >
>> >
>> >
>> > --
>> > Best regards,
>> >
>> >    - Andy
>> >
>> > Problems worthy of attack prove their worth by hitting back. - Piet Hein
>> > (via Tom White)
>> >
>>

Hey,

I don't think my opinion should be taken too seriously on this as I
don't have any need or access to a HBase cluster anymore.

But.. I think it would be a mistake to deprecate or remove thrift2 and
keep the original. It's just bad design to have completely different
looking thrift api to the java and rest apis. That said the thrift2
api is definitely incomplete compared to the thrift1 API as it
intentionally didn't implement the admin methods and people would be
annoyed if thrift1 suddenly disappeared.
I think a good design would be one in which the java/rest and thrift
apis were simply implementing the same interface.
A better one might be that the thrift and rest packages are stand
alone java projects that build separately and simply depend on HBase,
but that would be a hassle for deployment.

The thing is, if I was using hbase today, I wouldn't use the hbase
thrift or thrift2 api, I'd implement my own thrift api (in java) that
was domain specific so that the application didn't know it was talking
to hbase.

~Tim
On 7 December 2012 22:38, Ted Yu <[EMAIL PROTECTED]> wrote:
> Tim:
> Do you have comment for the proposal below ?
>
>
> ---------- Forwarded message ----------
> From: Jimmy Xiang <[EMAIL PROTECTED]>
> Date: Fri, Dec 7, 2012 at 2:20 PM
> Subject: Re: thrift vs thrift2
> To: [EMAIL PROTECTED]
>
>
> Since thrift has many users while thrift2 doesn't, does it make more
> sense to deprecate and remove thrift2 instead?
> People are more used to the existing thrift interface.  Although it is
> different from the java API, it seems to not have
> any performance hit.
>
> Thanks,
> Jimmy
>
> On Fri, Dec 7, 2012 at 2:01 PM, Ted Yu <[EMAIL PROTECTED]> wrote:
>> Logged https://issues.apache.org/jira/browse/HBASE-7302 for this.
>>
>> On Fri, Dec 7, 2012 at 1:56 PM, Luke Lu <[EMAIL PROTECTED]> wrote:
>>
>>> The embeded HRegionThriftServer should be ported to thrift2 before we
>>> deprecate thrift1. FB's (and potentially others) c++ client requires
>>> HRegionThriftServer.
>>>
>>> __Luke
>>>
>>>
>>> On Thu, Dec 6, 2012 at 7:21 PM, Andrew Purtell <[EMAIL PROTECTED]>
>>> wrote:
>>>
>>> > +1 for deprecation. Will bring any users out of the woodwork who might
>>> not
>>> > see this discussion at least.
>>> >
>>> > On Friday, December 7, 2012, Jesse Yates wrote:
>>> >
>>> > > According to the thrift2 docs, it is the 'right' way to access HBase
>>> > > -
>>> > the
>>> > > thrift package is based on an older HBase interface that no longer
>>> > exists.
>>> > > Seems like a reasonable case for deprecation in the next release and
>>> > > dropping it all together after.
>>> > > -------------------
>>> > > Jesse Yates
>>> > > @jesse_yates
>>> > > jyates.github.com
>>> > >
>>> > >
>>> > > On Thu, Dec 6, 2012 at 4:48 PM, Stack
>>> > > <[EMAIL PROTECTED]<javascript:;>>
>>> > > wrote:
>>> > >
>>> > > > On Thu, Dec 6, 2012 at 3:24 PM, Jimmy Xiang <[EMAIL PROTECTED]
>>> > <javascript:;>>
>>> > > wrote:
>>> > > >
>>> > > > > Since 0.96 is a singularity, is it a time to deprecate thrift
>>> > finally?
>>> > > > >
>>> > > > > Or we can use thrift2 by default, and keep thrift around more?
>>> > > > >
>>> > > > > What's the pros and cons to switch to thrift2?
>>> > > > >
>>> > > >
>>> > > > Interesting notion.  Could move thrift2 to thrift altogether.
>>> > > > Would
>>> > need
>>> > > > to make sure thrift2 has all from thrift1.
>>> > > > St.Ack
>>> > > >
>>> > >
>>> >
>>> >
>>> > --
>>> > Best regards,
>>> >
>>> >    - Andy
>>> >
>>> > Problems worthy of attack prove their worth by hitting back. - Piet
>>> > Hein
>>> > (via Tom White)
>>> >
>>>
>

On Fri, Dec 7, 2012 at 3:13 PM, Tim Sell <[EMAIL PROTECTED]> wrote:

> Hey,
>
>

Hey Mighty Tim.
I think a good design would be one in which the java/rest and thrift
> apis were simply implementing the same interface.
>

Yes.

I've not looked at the two idls.  Could we put them together and have
thrift package do both apis (you can all slap me if that a dumb idea).

> A better one might be that the thrift and rest packages are stand
> alone java projects that build separately and simply depend on HBase,
> but that would be a hassle for deployment.
>
>
Yes.  Could make them separate projects if someone wanted to own them.

> The thing is, if I was using hbase today, I wouldn't use the hbase
> thrift or thrift2 api, I'd implement my own thrift api (in java) that
> was domain specific so that the application didn't know it was talking
> to hbase.
>

That makes sense.  I'd imagine though that you wouldn't mind having an
existing IDL for inspiration.

St.Ack

My understanding is the FB in regionserver thrift server was an experiment
that they've moved beyond, or at least would do something different now.
Maybe one of the FB guys could comment further if they're around here.
On Saturday, December 8, 2012, Stack wrote:

> On Fri, Dec 7, 2012 at 3:13 PM, Tim Sell <[EMAIL PROTECTED] <javascript:;>>
> wrote:
>
> > Hey,
> >
> >
>
> Hey Mighty Tim.
>
>
> I think a good design would be one in which the java/rest and thrift
> > apis were simply implementing the same interface.
> >
>
> Yes.
>
> I've not looked at the two idls.  Could we put them together and have
> thrift package do both apis (you can all slap me if that a dumb idea).
>
>
>
> > A better one might be that the thrift and rest packages are stand
> > alone java projects that build separately and simply depend on HBase,
> > but that would be a hassle for deployment.
> >
> >
> Yes.  Could make them separate projects if someone wanted to own them.
>
>
>
> > The thing is, if I was using hbase today, I wouldn't use the hbase
> > thrift or thrift2 api, I'd implement my own thrift api (in java) that
> > was domain specific so that the application didn't know it was talking
> > to hbase.
> >
>
> That makes sense.  I'd imagine though that you wouldn't mind having an
> existing IDL for inspiration.
>
> St.Ack
>
--
Best regards,

   - Andy

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