HBase, mail # dev - Re: 0.92.2?

Re: 0.92.2?

Devaraj Das 2012-05-03, 17:06

+1

On May 3, 2012, at 8:58 AM, Stack wrote:

> Should I put up a 0.92.2 candidate? There have been a bunch of fixes
> since 0.92.1: https://issues.apache.org/jira/secure/ReleaseNote.jspa?projectId=12310753&version=12319888
>
> St.Ack

Re: 0.92.2?

Todd Lipcon 2012-05-03, 17:09

Anyone looked into the apparent performance issue that Elliott reported
earlier this week? If we have a big perf regression from 0.90 we should
probably spend at least a few minutes looking into it.

-Todd

On Thu, May 3, 2012 at 8:58 AM, Stack <[EMAIL PROTECTED]> wrote:

> Should I put up a 0.92.2 candidate? There have been a bunch of fixes
> since 0.92.1:
> https://issues.apache.org/jira/secure/ReleaseNote.jspa?projectId=12310753&version=12319888
>
> St.Ack
>

--
Todd Lipcon
Software Engineer, Cloudera

Re: 0.92.2?

Elliott Clark 2012-05-03, 17:44

I'm going to try and test 0.94.0 RC3 a little today. After that I can try
and do some poking to see why 92 was so slow.

On Thu, May 3, 2012 at 10:09 AM, Todd Lipcon <[EMAIL PROTECTED]> wrote:

> Anyone looked into the apparent performance issue that Elliott reported
> earlier this week? If we have a big perf regression from 0.90 we should
> probably spend at least a few minutes looking into it.
>
> -Todd
>
> On Thu, May 3, 2012 at 8:58 AM, Stack <[EMAIL PROTECTED]> wrote:
>
> > Should I put up a 0.92.2 candidate? There have been a bunch of fixes
> > since 0.92.1:
> >
> https://issues.apache.org/jira/secure/ReleaseNote.jspa?projectId=12310753&version=12319888
> >
> > St.Ack
> >
>
>
>
> --
> Todd Lipcon
> Software Engineer, Cloudera
>

Re: 0.92.2?

Enis Söztutar 2012-05-03, 23:22

Shall we open a jira for investigating the performance issue, and mark it
as a blocker for 0.92.2?

Enis

On Thu, May 3, 2012 at 10:44 AM, Elliott Clark <[EMAIL PROTECTED]>wrote:

> I'm going to try and test 0.94.0 RC3 a little today. After that I can try
> and do some poking to see why 92 was so slow.
>
> On Thu, May 3, 2012 at 10:09 AM, Todd Lipcon <[EMAIL PROTECTED]> wrote:
>
> > Anyone looked into the apparent performance issue that Elliott reported
> > earlier this week? If we have a big perf regression from 0.90 we should
> > probably spend at least a few minutes looking into it.
> >
> > -Todd
> >
> > On Thu, May 3, 2012 at 8:58 AM, Stack <[EMAIL PROTECTED]> wrote:
> >
> > > Should I put up a 0.92.2 candidate? There have been a bunch of fixes
> > > since 0.92.1:
> > >
> >
> https://issues.apache.org/jira/secure/ReleaseNote.jspa?projectId=12310753&version=12319888
> > >
> > > St.Ack
> > >
> >
> >
> >
> > --
> > Todd Lipcon
> > Software Engineer, Cloudera
> >
>

Re: 0.92.2?

Ted Yu 2012-05-04, 20:27

I think HBASE-5942 should be fixed before rolling out 0.92.2 RC.

Cheers

On Thu, May 3, 2012 at 4:22 PM, Enis Söztutar <[EMAIL PROTECTED]> wrote:

> Shall we open a jira for investigating the performance issue, and mark it
> as a blocker for 0.92.2?
>
> Enis
>
> On Thu, May 3, 2012 at 10:44 AM, Elliott Clark <[EMAIL PROTECTED]
> >wrote:
>
> > I'm going to try and test 0.94.0 RC3 a little today. After that I can
> try
> > and do some poking to see why 92 was so slow.
> >
> > On Thu, May 3, 2012 at 10:09 AM, Todd Lipcon <[EMAIL PROTECTED]> wrote:
> >
> > > Anyone looked into the apparent performance issue that Elliott reported
> > > earlier this week? If we have a big perf regression from 0.90 we should
> > > probably spend at least a few minutes looking into it.
> > >
> > > -Todd
> > >
> > > On Thu, May 3, 2012 at 8:58 AM, Stack <[EMAIL PROTECTED]> wrote:
> > >
> > > > Should I put up a 0.92.2 candidate? There have been a bunch of fixes
> > > > since 0.92.1:
> > > >
> > >
> >
> https://issues.apache.org/jira/secure/ReleaseNote.jspa?projectId=12310753&version=12319888
> > > >
> > > > St.Ack
> > > >
> > >
> > >
> > >
> > > --
> > > Todd Lipcon
> > > Software Engineer, Cloudera
> > >
> >
>

Re: 0.92.2?

Luke Lu 2012-05-10, 17:39

Any update on the slowness of 0.92 especially for YCSB workload A
(50/50 read/write, which shows about 40% regression. A 15/85
read/write work load shows 60% regression)? 0.92 seem fine for pure
writes/reads.

On Thu, May 3, 2012 at 10:44 AM, Elliott Clark <[EMAIL PROTECTED]> wrote:
> I'm going to try and test 0.94.0 RC3 a little today. After that I can try
> and do some poking to see why 92 was so slow.
>
> On Thu, May 3, 2012 at 10:09 AM, Todd Lipcon <[EMAIL PROTECTED]> wrote:
>
>> Anyone looked into the apparent performance issue that Elliott reported
>> earlier this week? If we have a big perf regression from 0.90 we should
>> probably spend at least a few minutes looking into it.
>>
>> -Todd
>>
>> On Thu, May 3, 2012 at 8:58 AM, Stack <[EMAIL PROTECTED]> wrote:
>>
>> > Should I put up a 0.92.2 candidate? There have been a bunch of fixes
>> > since 0.92.1:
>> >
>> https://issues.apache.org/jira/secure/ReleaseNote.jspa?projectId=12310753&version=12319888
>> >
>> > St.Ack
>> >
>>
>>
>>
>> --
>> Todd Lipcon
>> Software Engineer, Cloudera
>>

Re: 0.92.2?

Stack 2012-05-12, 21:59

On Thu, May 10, 2012 at 10:39 AM, Luke Lu <[EMAIL PROTECTED]> wrote:
> Any update on the slowness of 0.92 especially for YCSB workload A
> (50/50 read/write, which shows about 40% regression. A 15/85
> read/write work load shows 60% regression)? 0.92 seem fine for pure
> writes/reads.
>

Hey Luke:

What did you test on? What was your cluster like? Unadorned YCSB?
Can I have your commands? I'd like to repro.

Thanks,
St.Ack

Re: 0.92.2?

Jean-Daniel Cryans 2012-05-14, 21:58

FWIW we're running in production with a snapshot of 0.92 from May 8th
(so we're missing HBASE-5973 and 5922) and we don't see any issues.

I think we should release a first RC.

J-D

On Sat, May 12, 2012 at 2:59 PM, Stack <[EMAIL PROTECTED]> wrote:
> On Thu, May 10, 2012 at 10:39 AM, Luke Lu <[EMAIL PROTECTED]> wrote:
>> Any update on the slowness of 0.92 especially for YCSB workload A
>> (50/50 read/write, which shows about 40% regression. A 15/85
>> read/write work load shows 60% regression)? 0.92 seem fine for pure
>> writes/reads.
>>
>
> Hey Luke:
>
> What did you test on? What was your cluster like? Unadorned YCSB?
> Can I have your commands? I'd like to repro.
>
> Thanks,
> St.Ack

Re: 0.92.2?

Luke Lu 2012-05-14, 23:55

Looks like the workload A issue was due to an incorrect recordcount
specified for 0.92 runs, as I asked for more details. AFAICT, 0.92 is
significantly faster than 0.90 on a 3 (hbase/hdfs) +1 (ycsb) node
cluster. Sorry for the noise and thanks for the great work :)

On Sat, May 12, 2012 at 2:59 PM, Stack <[EMAIL PROTECTED]> wrote:
> On Thu, May 10, 2012 at 10:39 AM, Luke Lu <[EMAIL PROTECTED]> wrote:
>> Any update on the slowness of 0.92 especially for YCSB workload A
>> (50/50 read/write, which shows about 40% regression. A 15/85
>> read/write work load shows 60% regression)? 0.92 seem fine for pure
>> writes/reads.
>>
>
> Hey Luke:
>
> What did you test on? What was your cluster like? Unadorned YCSB?
> Can I have your commands? I'd like to repro.
>
> Thanks,
> St.Ack

Re: 0.92.2?

Stack 2012-05-15, 02:46

On Mon, May 14, 2012 at 4:55 PM, Luke Lu <[EMAIL PROTECTED]> wrote:
> Looks like the workload A issue was due to an incorrect recordcount
> specified for 0.92 runs, as I asked for more details. AFAICT, 0.92 is
> significantly faster than 0.90 on a 3 (hbase/hdfs) +1 (ycsb) node
> cluster. Sorry for the noise and thanks for the great work :)
>

np Luke.

Let me role an RC for 0.92.2....

St.Ack