hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Todd Lipcon <t...@cloudera.com>
Subject Re: ANN: The third hbase 0.92.0 release candidate is available for download
Date Wed, 04 Jan 2012 03:45:07 GMT
On Tue, Jan 3, 2012 at 7:40 PM, Ted Yu <yuzhihong@gmail.com> wrote:
> I agree with Todd that we should cover the scenario where there're many
> regions.
>
>>> we at least have decent visibility and fix tools compared to what we had
> in 90.
> We have better tools, yes.
> Would they solve the known / potential problems ? I don't think we have
> enough cases to prove that.

Maybe not... but I don't think it's particularly worse off than 0.90.x
in this regard either.

I won't vote +1 on the release since I haven't had time to pound on it
personally. But I generally think that we need to just get this thing
out there, even if we label it as an alpha initially for 0.92.0, beta
for 0.92.1, or whatever. Remember when we wanted to get 92 out for
Hadoop World?

-Todd

>
> On Tue, Jan 3, 2012 at 7:36 PM, Todd Lipcon <todd@cloudera.com> wrote:
>
>> On Tue, Jan 3, 2012 at 7:16 PM, Jean-Daniel Cryans <jdcryans@apache.org>
>> wrote:
>>
>> > I would also suggest a third option that seems like a stretch but
>> > could be workable:
>> >
>> > - MAX_FILESIZE is 4x bigger so users are less likely to have a huge
>> > number of regions (plus all our education), so the TM is less likely
>> > to cause damage and could be very useful. What I mean is 5119 could be
>> > committed but not 5120 for 0.92.0
>> >
>>
>> That doesn't help the folks who will be upgrading from an existing
>> cluster and still have too-many regions. We've got some merge hacks
>> but nothing super-easy to use yet. So I think we have to expect that
>> some folks will have lots of regions for a little while yet.
>>
>> Given that people have been testing w/o the TimeoutMonitor, I think we
>> should leave these confs as they are in the current 92 rc and not futz
>> too much. If stuff gets stuck in transition we at least have decent
>> visibility and fix tools compared to what we had in 90.
>>
>> -Todd
>>
>> > On Tue, Jan 3, 2012 at 6:47 PM, Ted Yu <yuzhihong@gmail.com> wrote:
>> >> I cloned HBASE-5120 off of HBASE-5119 and marked it as a blocker.
>> >> I think it shows that we may need to revisit HBASE-4015.
>> >>
>> >> Regards
>> >>
>> >> On Tue, Jan 3, 2012 at 5:10 PM, Ted Yu <yuzhihong@gmail.com> wrote:
>> >>
>> >>> Shall we also address the scenario where timeout monitor and bulk
>> disabler
>> >>> race against the same region ?
>> >>> See
>> >>>
>> https://issues.apache.org/jira/browse/HBASE-5119?focusedCommentId=13179176&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13179176
>> >>>
>> >>> Cheers
>> >>>
>> >>>
>> >>> On Tue, Jan 3, 2012 at 5:04 PM, Stack <stack@duboce.net> wrote:
>> >>>
>> >>>> On Tue, Jan 3, 2012 at 3:12 PM, Jonathan Hsieh <jon@cloudera.com>
>> wrote:
>> >>>> > I am similarly concerned about the deadlocks in distributed
log
>> >>>> splitting
>> >>>> > that Jimmy and Prakash have been working on.
>> >>>> >
>> >>>> > If distributed long splitting is off by default, I'm +1.  If
it is
>> >>>> going to
>> >>>> > be default on, then I'd prefer getting those bugs bumped up
to
>> >>>> > blockers fixed and then spinning another rc.
>> >>>> >
>> >>>>
>> >>>> I'll put up a new RC after HBASE-5081 goes in (I'll update our hadoop
>> >>>> to the released 1.0.0).
>> >>>>
>> >>>> St.Ack
>> >>>>
>> >>>
>> >>>
>>
>>
>>
>> --
>> Todd Lipcon
>> Software Engineer, Cloudera
>>



-- 
Todd Lipcon
Software Engineer, Cloudera

Mime
View raw message