Mailing-List: contact general-help@lucene.apache.org; run by ezmlm
Precedence: bulk
Reply-To: general@lucene.apache.org
From: =?utf-8?Q?Jan_H=C3=B8ydahl?= <jan.asf@cominvent.com>
Content-Type: multipart/alternative;
 boundary="Apple-Mail=_9B2A15C9-79AC-4AD3-BA9B-61A6FAF26828"
Mime-Version: 1.0 (Mac OS X Mail 10.1 \(3251\))
Subject: Re: Question: Does Zookeeper/SolrCloud handle the failover?
Date: Fri, 6 Jan 2017 15:31:33 +0100
References: <d93bda3b90594a308ff4c5d7b4e308dd@FarmMail.acsysoffice.com>
 <CAF8TkC62jHjgL3d=5_-sR3QTs2UddrGoBNKiwBwY7FcT5EcKiA@mail.gmail.com>
 <65090959df5b4f4e880bbfb983e1ddee@FarmMail.acsysoffice.com>
To: general@lucene.apache.org
In-Reply-To: <65090959df5b4f4e880bbfb983e1ddee@FarmMail.acsysoffice.com>
Message-Id: <ECE44EAE-6A7C-46D7-95EC-DD01198399DF@cominvent.com>
archived-at: Fri, 06 Jan 2017 14:31:44 -0000

--Apple-Mail=_9B2A15C9-79AC-4AD3-BA9B-61A6FAF26828
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain;
	charset=utf-8

Hi,

That should work well, as long as the list of Solr nodes is fairly =
constant. Using ZK your client will always have the correct list of =
nodes and will always avoid unnecessary hops. With a hardcoded list of =
node names you=E2=80=99ll need to update your LB when you make changes =
to your cluster, and there will be more hops, but that is perfectly =
acceptable and neglectible for many users.

Another solution is to talk to ZK from your .NET application code =
directly, just to pull out the cluster state for each collection, and =
thus be able to pick nodes yourself in an intelligent way...

--
Jan H=C3=B8ydahl, search solution architect
Cominvent AS - www.cominvent.com

> 6. jan. 2017 kl. 15.20 skrev Mike Linnetz =
<Mike.Linnetz@theprimacy.com>:
>=20
> Thanks for the reply!
> What you describe makes sense.
>=20
> I have another question as I'd like to present my team all of the =
options.
>=20
> As an alternative to using SolrNetCloud to load balance, couldn't I =
just configure my network load balancer to distribute requests among my =
cloud instances?  As I understand it, Zookeeper and SolrCloud should =
take care of propagating any updates across all nodes in a collection if =
a request is routed to any node in the collection.
> Is there anything I'm missing such that that solution would not work?
>=20
> Thanks,
> Mike
>=20
>=20
> -----Original Message-----
> From: Mikhail Khludnev [mailto:mkhl@apache.org]=20
> Sent: Thursday, January 5, 2017 4:26 PM
> To: general@lucene.apache.org
> Subject: Re: Question: Does Zookeeper/SolrCloud handle the failover?
>=20
> Hello, Mike!
>=20
> As far as I've got, you need to connect to SolrCloud (Zookeeper) via
> Solr.Net library.
> Here is the reference to the fork where it's done.
> https://github.com/mausch/SolrNet/issues/174#issuecomment-143697474
> Let me know if you need more info.
>=20
> On Thu, Jan 5, 2017 at 8:50 PM, Mike Linnetz =
<Mike.Linnetz@theprimacy.com>
> wrote:
>=20
>> Hello,
>>=20
>> I am new to SolrCloud and Zookeeper, and there's a piece of the =
puzzle on
>> which I'm unclear.  I don't understand whether/how SolrCloud and =
Zookeeper
>> handles the high availability aspect.  That is, once I have the =
Zookeeper
>> ensemble set up, now how do I reference the collection in a
>> "high-availability" fashion so that if one Solr instance is down, it =
will
>> automatically route to another instance?
>>=20
>> Am I thinking that SolrCloud does more than it actually does?  That =
is,
>> could I instead handle all Load Balancing outside of Zookeeper, and =
just
>> use our company's network load balancer to route/distribute traffic =
for a
>> single consolidated URL:port to the different actual Solr =
Instances/Ports?
>>=20
>> Here's more details in case it helps:
>> I have an ASP.NET application which uses SolrNet to query and write =
to a
>> standalone Solr instance.  In my program , I specify the =
hostname:port of
>> my Solr instance:
>>                Startup.Init<SolrDoc>("hostname:port");
>> This setup was used for development and testing.
>>=20
>> Now, for production, I need to implement Solr in a H.A. fashion so =
that I
>> don't have a single point of failure.    So, following Apache's
>> documentation, I understand that the solution would be to set up an
>> external Zookeeper Ensemble.
>> So, let's say I set up a new Zookeeper Ensemble (three instances). =
Let's
>> also say I have three Solr instances, #shards =3D 2, #replicas per =
shard =3D 2.
>>=20
>> Once I have the Zookeeper ensemble set up, now how do I reference the
>> collection?
>> Previously, my code references a single standalone Solr instance, "
>> http://solrserver1:9983". What do I point it to now?
>>=20
>> If I point it to any of the three Solr instances, and if that =
instance I'm
>> referencing goes down, it means the request would fail, right?
>>=20
>> I read similar questions online, for example =
https://groups.google.com/
>> forum/#!msg/solrnet/-PeaGrLAMtw/pAfxuoYLVnIJ, and the answer seems to =
be
>> that SolrNet doesn't support this type of HA.  So, do I understand
>> correctly that the load balancing aspect needs to be handled on the
>> client-side (or via a network load balancer), not the =
zookeeper/solrcloud
>> server side?
>>=20
>> Thanks,
>> Mike
>>=20
>>=20
>>=20
>>=20
>=20
>=20
> --=20
> Sincerely yours
> Mikhail Khludnev


--Apple-Mail=_9B2A15C9-79AC-4AD3-BA9B-61A6FAF26828--