Next release

[Todd Lipcon <to...@cloudera.com>]

Hey all,

Hadoop World is just a little under a month away, so I wanted to start a
thread regarding planning for our next release. There's obviously no rule
that we must release anything for Hadoop World, but we had originally hoped
to ship 0.90 (or at least have it in a usable release candidate state) for
the conference. Lining releases up around conferences is a good way to build
buzz, etc.

To summarize the state of the world right now, we have basically two main
trees: (a) 0.89.20100830 in release candidates, and (b) trunk which includes
the master rewrite branch merged. Currently, 0.89.20100830 is pretty good
looking and Stumble is using it in production modulo a few changes. Trunk is
a bit messy, as it's failing its tests and apparently won't even start up a
cluster. It's certainly not in a releasable state right now, and it remains
to be seen how much more work we need to put into it to get it to be stable.

So, I think we have basically three choices:

1) Continue to work on stabilizing trunk, and aim to get it usable enough to
do a developer release for Hadoop World, still under the umbrella of the
0.89 series. Assumedly this release would be *less* stable than
0.89.20100830 given the scope of the changes and the amount of time we have
remaining to test and fix, so calling it 0.90 at that point is probably
unwise.

2) Add a few more patches to 0.89.20100830 to fix up some current issues
(assignment speed, read optimizations, et al), and rebrand it as 0.90. In
parallel we continue to work on trunk and start doing developer releases
called 0.91.* including all the master rewrite. Then when the master rewrite
is ready, we'll release that as 0.92.

3) We can completely ignore Hadoop World - it's a nice place to make a
release, but we can always miss the date by a month if we prefer.

I'm personally leaning towards option 2 above - there are a ton of great
fixes and features since 0.20, and getting them out to a broader user
community sooner rather than later seems prudent. We can then take our time
to fix up and stabilize trunk and get an 0.92 out with all the goodness when
it's ready.

Thoughts?
-Todd
-- 
Todd Lipcon
Software Engineer, Cloudera

Re: Next release

[Stack <st...@duboce.net>]

+1 on having a release in time for Hadoop World.

My thought is that we have until next week to decide among the
options. My hope is that 1.) will be viable by Wednesday of next week
or so (If this is not the case, I'll flip my vote to option 2.)).

St.Ack

On Wed, Sep 15, 2010 at 1:56 PM, Todd Lipcon <to...@cloudera.com> wrote:
> Hey all,
>
> Hadoop World is just a little under a month away, so I wanted to start a
> thread regarding planning for our next release. There's obviously no rule
> that we must release anything for Hadoop World, but we had originally hoped
> to ship 0.90 (or at least have it in a usable release candidate state) for
> the conference. Lining releases up around conferences is a good way to build
> buzz, etc.
>
> To summarize the state of the world right now, we have basically two main
> trees: (a) 0.89.20100830 in release candidates, and (b) trunk which includes
> the master rewrite branch merged. Currently, 0.89.20100830 is pretty good
> looking and Stumble is using it in production modulo a few changes. Trunk is
> a bit messy, as it's failing its tests and apparently won't even start up a
> cluster. It's certainly not in a releasable state right now, and it remains
> to be seen how much more work we need to put into it to get it to be stable.
>
> So, I think we have basically three choices:
>
> 1) Continue to work on stabilizing trunk, and aim to get it usable enough to
> do a developer release for Hadoop World, still under the umbrella of the
> 0.89 series. Assumedly this release would be *less* stable than
> 0.89.20100830 given the scope of the changes and the amount of time we have
> remaining to test and fix, so calling it 0.90 at that point is probably
> unwise.
>
> 2) Add a few more patches to 0.89.20100830 to fix up some current issues
> (assignment speed, read optimizations, et al), and rebrand it as 0.90. In
> parallel we continue to work on trunk and start doing developer releases
> called 0.91.* including all the master rewrite. Then when the master rewrite
> is ready, we'll release that as 0.92.
>
> 3) We can completely ignore Hadoop World - it's a nice place to make a
> release, but we can always miss the date by a month if we prefer.
>
> I'm personally leaning towards option 2 above - there are a ton of great
> fixes and features since 0.20, and getting them out to a broader user
> community sooner rather than later seems prudent. We can then take our time
> to fix up and stabilize trunk and get an 0.92 out with all the goodness when
> it's ready.
>
> Thoughts?
> -Todd
> --
> Todd Lipcon
> Software Engineer, Cloudera
>

Re: Next release

[Andrew Purtell <ap...@apache.org>]

We are in the process of getting coprocessor and security related patches out for trunk. Gary's RPC changes are already up on reviewboard, the first bits of coprocessor framework should be up this week, likewise the first cut of a coprocessor-based discretionary access control engine. 

We would like to have these be in a somewhat "real" state by Hadoop World, i.e. included in trunk, so I prefer option #1. 

Best regards,

    - Andy


--- On Wed, 9/15/10, Todd Lipcon <to...@cloudera.com> wrote:

> From: Todd Lipcon <to...@cloudera.com>
> Subject: Next release
> To: hbase-dev@hadoop.apache.org
> Date: Wednesday, September 15, 2010, 1:56 PM
> Hey all,
> 
> Hadoop World is just a little under a month away, so I
> wanted to start a
> thread regarding planning for our next release. There's
> obviously no rule
> that we must release anything for Hadoop World, but we had
> originally hoped
> to ship 0.90 (or at least have it in a usable release
> candidate state) for
> the conference. Lining releases up around conferences is a
> good way to build
> buzz, etc.
> 
> To summarize the state of the world right now, we have
> basically two main
> trees: (a) 0.89.20100830 in release candidates, and (b)
> trunk which includes
> the master rewrite branch merged. Currently, 0.89.20100830
> is pretty good
> looking and Stumble is using it in production modulo a few
> changes. Trunk is
> a bit messy, as it's failing its tests and apparently won't
> even start up a
> cluster. It's certainly not in a releasable state right
> now, and it remains
> to be seen how much more work we need to put into it to get
> it to be stable.
> 
> So, I think we have basically three choices:
> 
> 1) Continue to work on stabilizing trunk, and aim to get it
> usable enough to
> do a developer release for Hadoop World, still under the
> umbrella of the
> 0.89 series. Assumedly this release would be *less* stable
> than
> 0.89.20100830 given the scope of the changes and the amount
> of time we have
> remaining to test and fix, so calling it 0.90 at that point
> is probably
> unwise.
> 
> 2) Add a few more patches to 0.89.20100830 to fix up some
> current issues
> (assignment speed, read optimizations, et al), and rebrand
> it as 0.90. In
> parallel we continue to work on trunk and start doing
> developer releases
> called 0.91.* including all the master rewrite. Then when
> the master rewrite
> is ready, we'll release that as 0.92.
> 
> 3) We can completely ignore Hadoop World - it's a nice
> place to make a
> release, but we can always miss the date by a month if we
> prefer.
> 
> I'm personally leaning towards option 2 above - there are a
> ton of great
> fixes and features since 0.20, and getting them out to a
> broader user
> community sooner rather than later seems prudent. We can
> then take our time
> to fix up and stabilize trunk and get an 0.92 out with all
> the goodness when
> it's ready.
> 
> Thoughts?
> -Todd
> -- 
> Todd Lipcon
> Software Engineer, Cloudera
>