Hi all.

I have requirements to use hbase in several datacenters.
But HDFS has as SPOF, so we can't use it.
I plan to use ceph as the file system for hbase.

In general, interested in issues of:
a) use the hadoop + сeph in the production environment
b) using hbase + ceph in the production environment

In particular:
Ceph Wiki states that the integration of hadoop/ceph has the problem in the
form of modification dates of the files.
Does hbase uses modification times for some critical tasks?

--
Andrey.

Hi Andrey,

I don't think anyone has ever run Ceph in a production use case, much
less run it in a production use case with HBase.

The SPOF problem is currently being addressed and will be available
for production use by the middle of next calendar year. It will be
available for preview/beta usage in Q1 (it's already working in
development setups - feel free to follow HDFS-1623 for details)

I think you're way better off waiting for the HA NameNode support in
HDFS than going with an unproven FS like Ceph. Ceph certainly has
promise and is a nice design, but it hasn't had the years of pounding
that HDFS has.

-Todd

On Mon, Dec 19, 2011 at 10:23 PM, Andrey Stepachev <[EMAIL PROTECTED]> wrote:
> Hi all.
>
> I have requirements to use hbase in several datacenters.
> But HDFS has as SPOF, so we can't use it.
> I plan to use ceph as the file system for hbase.
>
> In general, interested in issues of:
> a) use the hadoop + сeph in the production environment
> b) using hbase + ceph in the production environment
>
> In particular:
> Ceph Wiki states that the integration of hadoop/ceph has the problem in the
> form of modification dates of the files.
> Does hbase uses modification times for some critical tasks?
>
> --
> Andrey.

--
Todd Lipcon
Software Engineer, Cloudera

Thanks for reply Todd.

Great news, I'll, definitely, prefer hdfs if HA already
available for development.

I see, that most issues addressed to HA branch.
What version this branch is applied to? 0.22 or
i should build hadoop from sources?

20 декабря 2011 г. 11:46 п��льзователь Todd Lipcon <[EMAIL PROTECTED]>написал:

> Hi Andrey,
>
> I don't think anyone has ever run Ceph in a production use case, much
> less run it in a production use case with HBase.
>
> The SPOF problem is currently being addressed and will be available
> for production use by the middle of next calendar year. It will be
> available for preview/beta usage in Q1 (it's already working in
> development setups - feel free to follow HDFS-1623 for details)
>
> I think you're way better off waiting for the HA NameNode support in
> HDFS than going with an unproven FS like Ceph. Ceph certainly has
> promise and is a nice design, but it hasn't had the years of pounding
> that HDFS has.
>
> -Todd
>
> On Mon, Dec 19, 2011 at 10:23 PM, Andrey Stepachev <[EMAIL PROTECTED]>
> wrote:
> > Hi all.
> >
> > I have requirements to use hbase in several datacenters.
> > But HDFS has as SPOF, so we can't use it.
> > I plan to use ceph as the file system for hbase.
> >
> > In general, interested in issues of:
> > a) use the hadoop + сeph in the production environment
> > b) using hbase + ceph in the production environment
> >
> > In particular:
> > Ceph Wiki states that the integration of hadoop/ceph has the problem in
> the
> > form of modification dates of the files.
> > Does hbase uses modification times for some critical tasks?
> >
> > --
> > Andrey.
>
>
>
> --
> Todd Lipcon
> Software Engineer, Cloudera
>

--
Андрей.

On Tue, Dec 20, 2011 at 12:56 AM, Andrey Stepachev <[EMAIL PROTECTED]> wrote:
> I see, that most issues addressed to HA branch.
> What version this branch is applied to? 0.22 or
> i should build hadoop from sources?

It's currently only available in the HDFS-1623 branch, which is in
active development and definitely shouldn't be used unless you plan to
help us finish debugging it :) There are still some known bugs and
deficiencies to address before we merge.

The plan is to merge it into 0.23 but not 0.22 - it will hopefully be
merged in January. We'll also make it available in CDH4 beta releases
early next year.

-Todd

>
> 20 декабря 2011 г. 11:46 пользователь Todd Lipcon <[EMAIL PROTECTED]>написал:
>
>> Hi Andrey,
>>
>> I don't think anyone has ever run Ceph in a production use case, much
>> less run it in a production use case with HBase.
>>
>> The SPOF problem is currently being addressed and will be available
>> for production use by the middle of next calendar year. It will be
>> available for preview/beta usage in Q1 (it's already working in
>> development setups - feel free to follow HDFS-1623 for details)
>>
>> I think you're way better off waiting for the HA NameNode support in
>> HDFS than going with an unproven FS like Ceph. Ceph certainly has
>> promise and is a nice design, but it hasn't had the years of pounding
>> that HDFS has.
>>
>> -Todd
>>
>> On Mon, Dec 19, 2011 at 10:23 PM, Andrey Stepachev <[EMAIL PROTECTED]>
>> wrote:
>> > Hi all.
>> >
>> > I have requirements to use hbase in several datacenters.
>> > But HDFS has as SPOF, so we can't use it.
>> > I plan to use ceph as the file system for hbase.
>> >
>> > In general, interested in issues of:
>> > a) use the hadoop + сeph in the production environment
>> > b) using hbase + ceph in the production environment
>> >
>> > In particular:
>> > Ceph Wiki states that the integration of hadoop/ceph has the problem in
>> the
>> > form of modification dates of the files.
>> > Does hbase uses modification times for some critical tasks?
>> >
>> > --
>> > Andrey.
>>
>>
>>
>> --
>> Todd Lipcon
>> Software Engineer, Cloudera
>>
>
>
>
> --
> Андрей.

--
Todd Lipcon
Software Engineer, Cloudera

23 декабря 2011 г. 22:48 п��льзователь Todd Lipcon <[EMAIL PROTECTED]>написал:

> On Tue, Dec 20, 2011 at 12:56 AM, Andrey Stepachev <[EMAIL PROTECTED]>
> wrote:
> > I see, that most issues addressed to HA branch.
> > What version this branch is applied to? 0.22 or
> > i should build hadoop from sources?
>
> It's currently only available in the HDFS-1623 branch, which is in
> active development and definitely shouldn't be used unless you plan to
> help us finish debugging it :) There are still some known bugs and
> deficiencies to address before we merge.
>

Looks like this is one of possible scenarios :). We need
some solution for HA, but no one (except MapR) still exists.
So we must use MapR (costs, closed) or debug some
not-ready-yet solution.
>
> The plan is to merge it into 0.23 but not 0.22 - it will hopefully be
> merged in January. We'll also make it available in CDH4 beta releases
> early next year.
>

Sounds good. But still can't find such branch in cloudera (or any other)
github
repositories.
Is it publicly available right now, or it in some private repository?
>
> -Todd
>
> >
> > 20 декабря 2011 г. 11:46 пользователь Todd Lipcon <[EMAIL PROTECTED]
> >написал:
> >
> >> Hi Andrey,
> >>
> >> I don't think anyone has ever run Ceph in a production use case, much
> >> less run it in a production use case with HBase.
> >>
> >> The SPOF problem is currently being addressed and will be available
> >> for production use by the middle of next calendar year. It will be
> >> available for preview/beta usage in Q1 (it's already working in
> >> development setups - feel free to follow HDFS-1623 for details)
> >>
> >> I think you're way better off waiting for the HA NameNode support in
> >> HDFS than going with an unproven FS like Ceph. Ceph certainly has
> >> promise and is a nice design, but it hasn't had the years of pounding
> >> that HDFS has.
> >>
> >> -Todd
> >>
> >> On Mon, Dec 19, 2011 at 10:23 PM, Andrey Stepachev <[EMAIL PROTECTED]>
> >> wrote:
> >> > Hi all.
> >> >
> >> > I have requirements to use hbase in several datacenters.
> >> > But HDFS has as SPOF, so we can't use it.
> >> > I plan to use ceph as the file system for hbase.
> >> >
> >> > In general, interested in issues of:
> >> > a) use the hadoop + сeph in the production environment
> >> > b) using hbase + ceph in the production environment
> >> >
> >> > In particular:
> >> > Ceph Wiki states that the integration of hadoop/ceph has the problem
> in
> >> the
> >> > form of modification dates of the files.
> >> > Does hbase uses modification times for some critical tasks?
> >> >
> >> > --
> >> > Andrey.
> >>
> >>
> >>
> >> --
> >> Todd Lipcon
> >> Software Engineer, Cloudera
> >>
> >
> >
> >
> > --
> > Андрей.
>
>
>
> --
> Todd Lipcon
> Software Engineer, Cloudera
>

--
Андрей.

On Dec 24, 2011 10:40 PM, "Andrey Stepachev" <[EMAIL PROTECTED]> wrote:
>
> 23 декабря 2011 г. 22:48 пользователь Todd Lipcon <[EMAIL PROTECTED]
>написал:
>
> > On Tue, Dec 20, 2011 at 12:56 AM, Andrey Stepachev <[EMAIL PROTECTED]>
> > wrote:
> > > I see, that most issues addressed to HA branch.
> > > What version this branch is applied to? 0.22 or
> > > i should build hadoop from sources?
> >
> > It's currently only available in the HDFS-1623 branch, which is in
> > active development and definitely shouldn't be used unless you plan to
> > help us finish debugging it :) There are still some known bugs and
> > deficiencies to address before we merge.
> >
>
> Looks like this is one of possible scenarios :). We need
> some solution for HA, but no one (except MapR) still exists.
> So we must use MapR (costs, closed) or debug some
> not-ready-yet solution.
>
>
You have also the option of DB, with avatar node,  the code is available on
the net
regarding mapr you have their m3 version which is free as far as I
remember.
Mikael.S
> >
> > The plan is to merge it into 0.23 but not 0.22 - it will hopefully be
> > merged in January. We'll also make it available in CDH4 beta releases
> > early next year.
> >
>
> Sounds good. But still can't find such branch in cloudera (or any other)
> github
> repositories.
> Is it publicly available right now, or it in some private repository?
>
>
> >
> > -Todd
> >
> > >
> > > 20 декабря 2011 г. 11:46 ��ользователь Todd Lipcon <[EMAIL PROTECTED]
> > >написал:
> > >
> > >> Hi Andrey,
> > >>
> > >> I don't think anyone has ever run Ceph in a production use case, much
> > >> less run it in a production use case with HBase.
> > >>
> > >> The SPOF problem is currently being addressed and will be available
> > >> for production use by the middle of next calendar year. It will be
> > >> available for preview/beta usage in Q1 (it's already working in
> > >> development setups - feel free to follow HDFS-1623 for details)
> > >>
> > >> I think you're way better off waiting for the HA NameNode support in
> > >> HDFS than going with an unproven FS like Ceph. Ceph certainly has
> > >> promise and is a nice design, but it hasn't had the years of pounding
> > >> that HDFS has.
> > >>
> > >> -Todd
> > >>
> > >> On Mon, Dec 19, 2011 at 10:23 PM, Andrey Stepachev <[EMAIL PROTECTED]>
> > >> wrote:
> > >> > Hi all.
> > >> >
> > >> > I have requirements to use hbase in several datacenters.
> > >> > But HDFS has as SPOF, so we can't use it.
> > >> > I plan to use ceph as the file system for hbase.
> > >> >
> > >> > In general, interested in issues of:
> > >> > a) use the hadoop + сeph in the production environment
> > >> > b) using hbase + ceph in the production environment
> > >> >
> > >> > In particular:
> > >> > Ceph Wiki states that the integration of hadoop/ceph has the
problem
> > in
> > >> the
> > >> > form of modification dates of the files.
> > >> > Does hbase uses modification times for some critical tasks?
> > >> >
> > >> > --
> > >> > Andrey.
> > >>
> > >>
> > >>
> > >> --
> > >> Todd Lipcon
> > >> Software Engineer, Cloudera
> > >>
> > >
> > >
> > >
> > > --
> > > Андрей.
> >
> >
> >
> > --
> > Todd Lipcon
> > Software Engineer, Cloudera
> >
>
>
>
> --
> Андрей.

28.12.2011 19:52 пользоват��ль "Mikael Sitruk" <[EMAIL PROTECTED]>
написал:
>
> On Dec 24, 2011 10:40 PM, "Andrey Stepachev" <[EMAIL PROTECTED]> wrote:
> >
> > 23 декабря 2011 г. 22:48 пользователь Todd Lipcon <[EMAIL PROTECTED]
> >написал:
> >
> > > On Tue, Dec 20, 2011 at 12:56 AM, Andrey Stepachev <[EMAIL PROTECTED]>
> > > wrote:
> > > > I see, that most issues addressed to HA branch.
> > > > What version this branch is applied to? 0.22 or
> > > > i should build hadoop from sources?
> > >
> > > It's currently only available in the HDFS-1623 branch, which is in
> > > active development and definitely shouldn't be used unless you plan to
> > > help us finish debugging it :) There are still some known bugs and
> > > deficiencies to address before we merge.
> > >
> >
> > Looks like this is one of possible scenarios :). We need
> > some solution for HA, but no one (except MapR) still exists.
> > So we must use MapR (costs, closed) or debug some
> > not-ready-yet solution.
> >
> >
> You have also the option of DB, with avatar node,  the code is available
on
> the net
Avatar is not an automatic solution. And it need secure nfs (in multiple
datacenters).

> regarding mapr you have their m3 version which is free as far as I
> remember.
M3 doesn't has ha.

> Mikael.S
> > >
> > > The plan is to merge it into 0.23 but not 0.22 - it will hopefully be
> > > merged in January. We'll also make it available in CDH4 beta releases
> > > early next year.
> > >
> >
> > Sounds good. But still can't find such branch in cloudera (or any other)
> > github
> > repositories.
> > Is it publicly available right now, or it in some private repository?
> >
> >
> > >
> > > -Todd
> > >
> > > >
> > > > 20 декабря 2011 г. 11:46 пользователь Todd Lipcon <[EMAIL PROTECTED]
> > > >написал:
> > > >
> > > >> Hi Andrey,
> > > >>
> > > >> I don't think anyone has ever run Ceph in a production use case,
much
> > > >> less run it in a production use case with HBase.
> > > >>
> > > >> The SPOF problem is currently being addressed and will be available
> > > >> for production use by the middle of next calendar year. It will be
> > > >> available for preview/beta usage in Q1 (it's already working in
> > > >> development setups - feel free to follow HDFS-1623 for details)
> > > >>
> > > >> I think you're way better off waiting for the HA NameNode support
in
> > > >> HDFS than going with an unproven FS like Ceph. Ceph certainly has
> > > >> promise and is a nice design, but it hasn't had the years of
pounding
> > > >> that HDFS has.
> > > >>
> > > >> -Todd
> > > >>
> > > >> On Mon, Dec 19, 2011 at 10:23 PM, Andrey Stepachev <
[EMAIL PROTECTED]>
> > > >> wrote:
> > > >> > Hi all.
> > > >> >
> > > >> > I have requirements to use hbase in several datacenters.
> > > >> > But HDFS has as SPOF, so we can't use it.
> > > >> > I plan to use ceph as the file system for hbase.
> > > >> >
> > > >> > In general, interested in issues of:
> > > >> > a) use the hadoop + сeph in the production environment
> > > >> > b) using hbase + ceph in the production environment
> > > >> >
> > > >> > In particular:
> > > >> > Ceph Wiki states that the integration of hadoop/ceph has the
> problem
> > > in
> > > >> the
> > > >> > form of modification dates of the files.
> > > >> > Does hbase uses modification times for some critical tasks?
> > > >> >
> > > >> > --
> > > >> > Andrey.
> > > >>
> > > >>
> > > >>
> > > >> --
> > > >> Todd Lipcon
> > > >> Software Engineer, Cloudera
> > > >>
> > > >
> > > >
> > > >
> > > > --
> > > > Андрей.
> > >
> > >
> > >
> > > --
> > > Todd Lipcon
> > > Software Engineer, Cloudera
> > >
> >
> >
> >
> > --
> > Андрей.