hadoop-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Harsh J <ha...@cloudera.com>
Subject Re: Question about HA and Federation
Date Fri, 21 Dec 2012 08:34:00 GMT
Appears alright to me!

On Fri, Dec 21, 2012 at 1:15 PM, ESGLinux <esggrupos@gmail.com> wrote:
> Hi,
>
> Finally I´m going to try this:
>
> 1 Machine: Active Name Node for NS1
> 1 Machine: Passive Name Node for NS1
> 1 Machine: NameNode for NS2 + NameNode for NS3
> 1 Machine: Secondary NameNode for NS2 + Secondary NameNode for NS3
>
> Is this correct?
>
> thanks,
>
> ESGLinux
>
> 2012/12/20 Harsh J <harsh@cloudera.com>
>>
>> Btw, you can co-locate NameNodes (unique namespace ones) onto the same
>> machine if you need to - the configs easily allow this via rpc/http
>> port specifiers.
>>
>> On Thu, Dec 20, 2012 at 11:33 PM, ESGLinux <esggrupos@gmail.com> wrote:
>> > Thank you very much,
>> >
>> > your answer have clarified me these concepts very much,
>> >
>> > I didn't understand how could I mix HA and Federation and how many nodes
>> > I
>> > need....
>> >
>> > Kind Regards,
>> >
>> > ESGLinux,
>> >
>> > 2012/12/20 Harsh J <harsh@cloudera.com>
>> >>
>> >> Yes I think its safe to say that - sorry that I missed out SNNs in my
>> >> first response (I counted only the regular serving namenodes) :)
>> >>
>> >> On Thu, Dec 20, 2012 at 11:25 PM, ESGLinux <esggrupos@gmail.com> wrote:
>> >> > Hi again,
>> >> >
>> >> > So finally the number of nodes are these:
>> >> >
>> >> > 1 Active NameNode + 1 Passive NameNode (it does the work of the old
>> >> > Secondary NameNode) for NS1 NameSpace (these are 2 diferent machines)
>> >> > 1 NameNode for NS2 + 1 Secondary NameNode
>> >> > 1 NameNode for NS3 + 1 Secondary NameNode
>> >> >
>> >> > We can say that we need 2 nodes per NameSpace, is that true?
>> >> >
>> >> > Thanks,
>> >> >
>> >> > ESGLinux
>> >> >
>> >> >
>> >> >
>> >> > 2012/12/20 Harsh J <harsh@cloudera.com>
>> >> >>
>> >> >> Hi,
>> >> >>
>> >> >> To put it simply: If you use a NameNode, you need a
>> >> >> SecondaryNameNode.
>> >> >> In HA-mode, a StandbyNameNode acts as a SecondaryNameNode (so you
>> >> >> don't need to run an extra).
>> >> >>
>> >> >> Either way, you definitely need the checkpoint operation happening
>> >> >> and
>> >> >> being monitored for.
>> >> >>
>> >> >> On Thu, Dec 20, 2012 at 11:09 PM, ESGLinux <esggrupos@gmail.com>
>> >> >> wrote:
>> >> >> > Hi Harsh,
>> >> >> >
>> >> >> > First thank you very much for your answer,
>> >> >> >
>> >> >> > following your example:
>> >> >> >
>> >> >> > You have:
>> >> >> >
>> >> >> > 1 Active NameNode + 1 Passive NameNode (it does the work of
the
>> >> >> > old
>> >> >> > Secondary NameNode) for NS1 NameSpace (these are 2 diferent
>> >> >> > machines)
>> >> >> > 1 NameNode for NS2
>> >> >> > 1 NameNode for NS3
>> >> >> >
>> >> >> > but what about the Secondary Name Nodes for NS2 and NS3? or
I
>> >> >> > don´t
>> >> >> > need
>> >> >> > it?
>> >> >> > perhaps I´m mixing concepts....
>> >> >> >
>> >> >> > Thanks again,
>> >> >> >
>> >> >> > Greetings,
>> >> >> >
>> >> >> > ESGLinux
>> >> >> >
>> >> >> >
>> >> >> >
>> >> >> >
>> >> >> > 2012/12/20 Harsh J <harsh@cloudera.com>
>> >> >> >>
>> >> >> >> Hi ESGLinux,
>> >> >> >>
>> >> >> >> Federation and HA are two distinct features that share
some
>> >> >> >> common
>> >> >> >> properties but nothing more. You can turn on HA for any
selected
>> >> >> >> Namespace but it is not necessarily needed to be that
all
>> >> >> >> Namespaces
>> >> >> >> have HA.
>> >> >> >>
>> >> >> >> Perhaps an example will clear it up for you.
>> >> >> >>
>> >> >> >> I have a local instance that is configured to run several
>> >> >> >> namespaces:
>> >> >> >> ns1, ns2, and ns3 (Federated Namespaces).
>> >> >> >> The namespace ns1 hosts my HBase tables and is critical
to me, so
>> >> >> >> I
>> >> >> >> have also turned on HA for this namespace alone.
>> >> >> >> The other two namespaces ns2 and ns3 are used only for
regular
>> >> >> >> query
>> >> >> >> jobs so its not yet very important for me to have HA on
it. So I
>> >> >> >> run
>> >> >> >> them without HA.
>> >> >> >>
>> >> >> >> Thus I have 4 NameNode processes in my cluster in all,
given my
>> >> >> >> design
>> >> >> >> above: (2 NNs under ns1, in HA mode) + (1 NN of ns2) +
(1 NN of
>> >> >> >> ns3).
>> >> >> >>
>> >> >> >> On Thu, Dec 20, 2012 at 10:00 PM, ESGLinux <esggrupos@gmail.com>
>> >> >> >> wrote:
>> >> >> >> > Hi All,
>> >> >> >> >
>> >> >> >> > I´m going to test a hadoop cluster and I have a
doubt about HA
>> >> >> >> > and
>> >> >> >> > Federation.
>> >> >> >> >
>> >> >> >> > With federation I Have a NameNode per namespace and
with HA I
>> >> >> >> > have
>> >> >> >> > an
>> >> >> >> > Active
>> >> >> >> > NameNode and a standby NameNode.
>> >> >> >> >
>> >> >> >> > so, as I have sevaral namespaces, do I need an Active
NameNode
>> >> >> >> > and
>> >> >> >> > a
>> >> >> >> > standby
>> >> >> >> > nameNode per namespace?
>> >> >> >> >
>> >> >> >> > I have read this documentation but It´s not clear
for me :-(
>> >> >> >> >
>> >> >> >> >
>> >> >> >> >
>> >> >> >> >
>> >> >> >> >
>> >> >> >> > https://ccp.cloudera.com/display/CDH4DOC/Introduction+to+Hadoop+High+Availability
>> >> >> >> >
>> >> >> >> >
>> >> >> >> >
>> >> >> >> >
>> >> >> >> > http://hadoop.apache.org/docs/r0.23.0/hadoop-yarn/hadoop-yarn-site/Federation.html
>> >> >> >> >
>> >> >> >> > Thanks in advance
>> >> >> >> >
>> >> >> >> > ESGLinux
>> >> >> >> >
>> >> >> >>
>> >> >> >>
>> >> >> >>
>> >> >> >> --
>> >> >> >> Harsh J
>> >> >> >
>> >> >> >
>> >> >>
>> >> >>
>> >> >>
>> >> >> --
>> >> >> Harsh J
>> >> >
>> >> >
>> >>
>> >>
>> >>
>> >> --
>> >> Harsh J
>> >
>> >
>>
>>
>>
>> --
>> Harsh J
>
>



-- 
Harsh J

Mime
View raw message