hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dave Latham <lat...@davelink.net>
Subject Re: Release 0.90.3 soon?
Date Wed, 04 May 2011 16:10:46 GMT
I was one of those bitten by the connection identity changes in 0.90 and
TableOutputFormat issues.  I was very glad to see HBASE-3777 addressed, and
many thanks to those who did the work to sort it out.  I'd love to have it
soon, but I have to agree that it's a pretty fundamental change and seems
high risk to put into a point release.

I knew there were some big changes moving from 0.20 to 0.90, so I did some
rigorous testing on our cluster (still haven't rolled out 0.90 yet) and
ended up making some non-trivial changes to our HBase usage and automated
tests to accomodate the the connection identity issues in 0.90.  I wouldn't
expect that level of change in a point release, and don't put point releases
through as much testing.

I think it's worth putting more effort into getting 0.92 out the door as
soon as is reasonable to get this change and others out.  Keep the point
releases as risk free as possible.  Users should have a high degree of
confidence that they are strictly improvements, and if their system works on
one point release, it should work on the next without modifications.


On Tue, May 3, 2011 at 9:07 PM, Stack <stack@duboce.net> wrote:

> IMO, HBASE-3777 is a critical fix -- it even addresses a regression
> introduced in 0.90.0 -- but its too risky putting it out now in a
> release from branch, at least just yet.  It was only committed a day
> or so ago (Thanks Karthick and Ted for the hard work getting it in).
> I think it needs a bit of bake-in.  We should be rolling a 0.92.0RC
> pretty soon.  It'll get some testing then.
> We can not risk a point release that is less stable than previous
> versions; if we err, the cost in terms of support and community trust
> is just too high.
> Meantime, any chance of a backport of hbase-3777 Ted?
> Good stuff,
> St.Ack
> On Tue, May 3, 2011 at 6:00 PM, Ted Yu <yuzhihong@gmail.com> wrote:
> > The possibility of HBASE-3777 creating bigger trouble than without is
> low,
> > in my opinion.
> > Maybe we should conduct a poll in user mailing list and count the votes.
> >
> > On Tue, May 3, 2011 at 5:34 PM, Jean-Daniel Cryans <jdcryans@apache.org
> >wrote:
> >
> >> > Actually these two actions are related.
> >> > I can imagine the disappointment among hbase users if HBASE-3777
> weren't
> >> > included in 0.90.3
> >>
> >> I can also imagine the disappointment if we release a 0.90.3 that
> >> contains more bugs than it fixes, it goes both ways. Moreover,
> >> HBASE-3777 wasn't targeted and still isn't targeted for 0.90.3, so I
> >> don't see how even if someone paid attention to the jira they would
> >> expect to see it in 0.90.3
> >>
> >> I'd like to state that I'm not trying to discredit the work that was
> >> done in that Jira, it was a perfect example of open source
> >> collaboration, but I'm rather trying to point out that it's a big
> >> change and that the bigger the change the better the chances are that
> >> there will be bugs lurking in it. You could easily list big patches
> >> that were committed to point releases in the past and I would agree
> >> with you that this is something we've done, but I can also recall a
> >> number of those changes that introduced more bugs and even made some
> >> releases unusable (like 0.20.4). Let's try to learn from our errors.
> >>
> >> Finally, even if it's not in 0.90.3, the fact that a backport be made
> >> available means that people can patch it in themselves or that other
> >> distros can decide to include it (like in the next CDH3 update). And
> >> finally we could do a 0.90.4 with it.
> >>
> >> J-D
> >>
> >

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