You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hbase.apache.org by Stack <st...@duboce.net> on 2010/03/09 01:03:46 UTC
Any objection if we up RPC version when we go from 0.20.3 to 0.20.4?
Are there objections to changing the RPC version number when we
release 0.20.4? Changing the RPC version number will mean that a
0.20.4 and 0.20.3 version hbase cluster will not be able to talk to
each other. Updating your cluster from 0.20.3 to 0.20.4 will require
a restart.
Currently our RPC is brittle in that we can't add or amend methods
without breaking the interface. We'd like to make the interface
tolerant of small changes. To do this, we need to change how RPC
identifies remote methods. See HBASE-2219 for details. The change in
method identification requires that we up the RPC version so
mismatched clients fail fast.
If above is amenable, we'll also pull into 0.20.4 other facility such
as the new multiput and fixes that make us less vulnerable to flapping
dns, etc.
Yours,
St.Ack
Re: Any objection if we up RPC version when we go from 0.20.3 to
0.20.4?
Posted by Stack <st...@duboce.net>.
I'm +1 too.
Will make sure that on release of 0.20.4, we're clear about the cold
reboot required (as per Kay Kay suggestion).
St.Ack
On Mon, Mar 8, 2010 at 7:41 PM, Ryan Rawson <ry...@gmail.com> wrote:
> As the author of the patch in question I unquestionably believe in it,
> so I give this a +1.
>
>
>
> On Mon, Mar 8, 2010 at 7:37 PM, Jonathan Gray <jl...@streamy.com> wrote:
>> +1
>>
>> JG
>>
>> -----Original Message-----
>> From: Dhruba Borthakur [mailto:dhruba@gmail.com]
>> Sent: Monday, March 08, 2010 5:59 PM
>> To: hbase-dev@hadoop.apache.org
>> Subject: Re: Any objection if we up RPC version when we go from 0.20.3 to
>> 0.20.4?
>>
>> +1.
>>
>> dhruba
>>
>> On Mon, Mar 8, 2010 at 5:16 PM, Kay Kay <ka...@gmail.com> wrote:
>>
>>> not at all, considering the important fixes, we get in return.
>>>
>>> Meanwhile - it might be worthwhile to have a 'bold' mention w.r.t backward
>>> incompatibility ( cold restart , as opposed to a rolling one) in the
>>> documentation home page for 0.20.4 about this. Thanks.
>>>
>>>
>>>
>>> On 3/8/10 4:03 PM, Stack wrote:
>>>
>>>> Are there objections to changing the RPC version number when we
>>>> release 0.20.4? Changing the RPC version number will mean that a
>>>> 0.20.4 and 0.20.3 version hbase cluster will not be able to talk to
>>>> each other. Updating your cluster from 0.20.3 to 0.20.4 will require
>>>> a restart.
>>>>
>>>> Currently our RPC is brittle in that we can't add or amend methods
>>>> without breaking the interface. We'd like to make the interface
>>>> tolerant of small changes. To do this, we need to change how RPC
>>>> identifies remote methods. See HBASE-2219 for details. The change in
>>>> method identification requires that we up the RPC version so
>>>> mismatched clients fail fast.
>>>>
>>>> If above is amenable, we'll also pull into 0.20.4 other facility such
>>>> as the new multiput and fixes that make us less vulnerable to flapping
>>>> dns, etc.
>>>>
>>>> Yours,
>>>> St.Ack
>>>>
>>>>
>>>
>>>
>>
>>
>> --
>> Connect to me at http://www.facebook.com/dhruba
>>
>>
>
Re: Any objection if we up RPC version when we go from 0.20.3 to
0.20.4?
Posted by Jean-Daniel Cryans <jd...@apache.org>.
We discussed jumping to 0.20.5 (either have 0.20.4 as a bug fix
release only or skipping it entirely) instead of continuing with
0.20.4 that breaks compatibility. 0.21 means it works on Hadoop 0.21,
we binded ourselves to that, so it's not really an option.
I'm +1 on a 0.20.4 that breaks backward compatibility if we advertise
it as such, like using a size 72 warning in bright blinking characters
(not really). Also it wouldn't be the first time we aren't able to
keep it compatible and until 1.0 I don't really mind it that much.
J-D
On Mon, Mar 8, 2010 at 10:33 PM, tsuna <ts...@gmail.com> wrote:
> It's going to confuse the hell out of people because going from 0.20.3
> to 0.20.4 seems to be only a minor update.
> Can't we name the new release somewhat differently to make it clear
> that there's a compatibility problem between the two releases? Is
> 0.21.x really not an option?
>
> --
> Benoit "tsuna" Sigoure
> Software Engineer @ www.StumbleUpon.com
>
Re: Any objection if we up RPC version when we go from 0.20.3 to
0.20.4?
Posted by Stack <st...@duboce.net>.
I think naming it hbase 0.21.x will confuse more since it implies
hadoop 0.21.x (up to this we've aligned our versions).
Its my sense that few rely on being able to upgrade using the rolling
restart feature -- i.e. updating a daemon at a time letting the
cluster run across the upgrade -- and that only these few need to be
warned that rolling restart is not possible going 0.20.3 to 0.20.4.
St.Ack
On Mon, Mar 8, 2010 at 10:33 PM, tsuna <ts...@gmail.com> wrote:
> It's going to confuse the hell out of people because going from 0.20.3
> to 0.20.4 seems to be only a minor update.
> Can't we name the new release somewhat differently to make it clear
> that there's a compatibility problem between the two releases? Is
> 0.21.x really not an option?
>
> --
> Benoit "tsuna" Sigoure
> Software Engineer @ www.StumbleUpon.com
>
Re: Any objection if we up RPC version when we go from 0.20.3 to
0.20.4?
Posted by tsuna <ts...@gmail.com>.
It's going to confuse the hell out of people because going from 0.20.3
to 0.20.4 seems to be only a minor update.
Can't we name the new release somewhat differently to make it clear
that there's a compatibility problem between the two releases? Is
0.21.x really not an option?
--
Benoit "tsuna" Sigoure
Software Engineer @ www.StumbleUpon.com
Re: Any objection if we up RPC version when we go from 0.20.3 to
0.20.4?
Posted by Ryan Rawson <ry...@gmail.com>.
As the author of the patch in question I unquestionably believe in it,
so I give this a +1.
On Mon, Mar 8, 2010 at 7:37 PM, Jonathan Gray <jl...@streamy.com> wrote:
> +1
>
> JG
>
> -----Original Message-----
> From: Dhruba Borthakur [mailto:dhruba@gmail.com]
> Sent: Monday, March 08, 2010 5:59 PM
> To: hbase-dev@hadoop.apache.org
> Subject: Re: Any objection if we up RPC version when we go from 0.20.3 to
> 0.20.4?
>
> +1.
>
> dhruba
>
> On Mon, Mar 8, 2010 at 5:16 PM, Kay Kay <ka...@gmail.com> wrote:
>
>> not at all, considering the important fixes, we get in return.
>>
>> Meanwhile - it might be worthwhile to have a 'bold' mention w.r.t backward
>> incompatibility ( cold restart , as opposed to a rolling one) in the
>> documentation home page for 0.20.4 about this. Thanks.
>>
>>
>>
>> On 3/8/10 4:03 PM, Stack wrote:
>>
>>> Are there objections to changing the RPC version number when we
>>> release 0.20.4? Changing the RPC version number will mean that a
>>> 0.20.4 and 0.20.3 version hbase cluster will not be able to talk to
>>> each other. Updating your cluster from 0.20.3 to 0.20.4 will require
>>> a restart.
>>>
>>> Currently our RPC is brittle in that we can't add or amend methods
>>> without breaking the interface. We'd like to make the interface
>>> tolerant of small changes. To do this, we need to change how RPC
>>> identifies remote methods. See HBASE-2219 for details. The change in
>>> method identification requires that we up the RPC version so
>>> mismatched clients fail fast.
>>>
>>> If above is amenable, we'll also pull into 0.20.4 other facility such
>>> as the new multiput and fixes that make us less vulnerable to flapping
>>> dns, etc.
>>>
>>> Yours,
>>> St.Ack
>>>
>>>
>>
>>
>
>
> --
> Connect to me at http://www.facebook.com/dhruba
>
>
RE: Any objection if we up RPC version when we go from 0.20.3 to 0.20.4?
Posted by Jonathan Gray <jl...@streamy.com>.
+1
JG
-----Original Message-----
From: Dhruba Borthakur [mailto:dhruba@gmail.com]
Sent: Monday, March 08, 2010 5:59 PM
To: hbase-dev@hadoop.apache.org
Subject: Re: Any objection if we up RPC version when we go from 0.20.3 to
0.20.4?
+1.
dhruba
On Mon, Mar 8, 2010 at 5:16 PM, Kay Kay <ka...@gmail.com> wrote:
> not at all, considering the important fixes, we get in return.
>
> Meanwhile - it might be worthwhile to have a 'bold' mention w.r.t backward
> incompatibility ( cold restart , as opposed to a rolling one) in the
> documentation home page for 0.20.4 about this. Thanks.
>
>
>
> On 3/8/10 4:03 PM, Stack wrote:
>
>> Are there objections to changing the RPC version number when we
>> release 0.20.4? Changing the RPC version number will mean that a
>> 0.20.4 and 0.20.3 version hbase cluster will not be able to talk to
>> each other. Updating your cluster from 0.20.3 to 0.20.4 will require
>> a restart.
>>
>> Currently our RPC is brittle in that we can't add or amend methods
>> without breaking the interface. We'd like to make the interface
>> tolerant of small changes. To do this, we need to change how RPC
>> identifies remote methods. See HBASE-2219 for details. The change in
>> method identification requires that we up the RPC version so
>> mismatched clients fail fast.
>>
>> If above is amenable, we'll also pull into 0.20.4 other facility such
>> as the new multiput and fixes that make us less vulnerable to flapping
>> dns, etc.
>>
>> Yours,
>> St.Ack
>>
>>
>
>
--
Connect to me at http://www.facebook.com/dhruba
Re: Any objection if we up RPC version when we go from 0.20.3 to
0.20.4?
Posted by Dhruba Borthakur <dh...@gmail.com>.
+1.
dhruba
On Mon, Mar 8, 2010 at 5:16 PM, Kay Kay <ka...@gmail.com> wrote:
> not at all, considering the important fixes, we get in return.
>
> Meanwhile - it might be worthwhile to have a 'bold' mention w.r.t backward
> incompatibility ( cold restart , as opposed to a rolling one) in the
> documentation home page for 0.20.4 about this. Thanks.
>
>
>
> On 3/8/10 4:03 PM, Stack wrote:
>
>> Are there objections to changing the RPC version number when we
>> release 0.20.4? Changing the RPC version number will mean that a
>> 0.20.4 and 0.20.3 version hbase cluster will not be able to talk to
>> each other. Updating your cluster from 0.20.3 to 0.20.4 will require
>> a restart.
>>
>> Currently our RPC is brittle in that we can't add or amend methods
>> without breaking the interface. We'd like to make the interface
>> tolerant of small changes. To do this, we need to change how RPC
>> identifies remote methods. See HBASE-2219 for details. The change in
>> method identification requires that we up the RPC version so
>> mismatched clients fail fast.
>>
>> If above is amenable, we'll also pull into 0.20.4 other facility such
>> as the new multiput and fixes that make us less vulnerable to flapping
>> dns, etc.
>>
>> Yours,
>> St.Ack
>>
>>
>
>
--
Connect to me at http://www.facebook.com/dhruba
Re: Any objection if we up RPC version when we go from 0.20.3 to
0.20.4?
Posted by Kay Kay <ka...@gmail.com>.
not at all, considering the important fixes, we get in return.
Meanwhile - it might be worthwhile to have a 'bold' mention w.r.t
backward incompatibility ( cold restart , as opposed to a rolling one)
in the documentation home page for 0.20.4 about this. Thanks.
On 3/8/10 4:03 PM, Stack wrote:
> Are there objections to changing the RPC version number when we
> release 0.20.4? Changing the RPC version number will mean that a
> 0.20.4 and 0.20.3 version hbase cluster will not be able to talk to
> each other. Updating your cluster from 0.20.3 to 0.20.4 will require
> a restart.
>
> Currently our RPC is brittle in that we can't add or amend methods
> without breaking the interface. We'd like to make the interface
> tolerant of small changes. To do this, we need to change how RPC
> identifies remote methods. See HBASE-2219 for details. The change in
> method identification requires that we up the RPC version so
> mismatched clients fail fast.
>
> If above is amenable, we'll also pull into 0.20.4 other facility such
> as the new multiput and fixes that make us less vulnerable to flapping
> dns, etc.
>
> Yours,
> St.Ack
>
Re: Any objection if we up RPC version when we go from 0.20.3 to 0.20.4?
Posted by Andrew Purtell <ap...@apache.org>.
+1
- Andy
----- Original Message ----
> From: Stack <st...@duboce.net>
> To: HBase Dev List <hb...@hadoop.apache.org>
> Sent: Mon, March 8, 2010 4:03:46 PM
> Subject: Any objection if we up RPC version when we go from 0.20.3 to 0.20.4?
>
> Are there objections to changing the RPC version number when we
> release 0.20.4? Changing the RPC version number will mean that a
> 0.20.4 and 0.20.3 version hbase cluster will not be able to talk to
> each other. Updating your cluster from 0.20.3 to 0.20.4 will require
> a restart.
>
> Currently our RPC is brittle in that we can't add or amend methods
> without breaking the interface. We'd like to make the interface
> tolerant of small changes. To do this, we need to change how RPC
> identifies remote methods. See HBASE-2219 for details. The change in
> method identification requires that we up the RPC version so
> mismatched clients fail fast.
>
> If above is amenable, we'll also pull into 0.20.4 other facility such
> as the new multiput and fixes that make us less vulnerable to flapping
> dns, etc.
>
> Yours,
> St.Ack