"I just want to utilize my servers maximum for select query . Also i have
approximately thousand of concurrent update/delete request to cluster and
ten thousand of concurrent select request ."

And remember, if you have, say, 100 shards and a thousand updates per
second, that's only 10 updates/sec/shard, and a similar consideration for
queries and the number of replicase per shard.

If that's per second, then you may need to move to TLOG or PULL replicas.
If it's per hour then _probably_ not.

This is really "the sizing question" and the only way to answer that question
that I know of is to prototype. Here's a long discourse on the subject.

https://lucidworks.com/2012/07/23/sizing-hardware-in-the-abstract-why-we-dont-have-a-definitive-answer/

Best,
Erick

On Thu, Apr 26, 2018 at 11:52 AM, kumar gaurav <kg2...@gmail.com> wrote:
> Hi Erick
>
> Pleased to see your reply .
>
> I just want to utilize my servers maximum for select query . Also i have
> approximately thousand of concurrent update/delete request to cluster and
> ten thousand of concurrent select request .
>
> So Is it the right way to let segment merging in all replicas in case of
> only NRT chosen ?
>
> i am assuming, PULL replicas will server select query better than other
> type of replica because it is free from indexing and TLOG balancing .
>
> Please clarify best possible solution for my case ?
>
> Waiting for your answer ... :)
>
>
>
>
> On Thu, Apr 26, 2018 at 9:08 PM, Erick Erickson <erickerick...@gmail.com>
> wrote:
>
>> First I'd ask if you want anything except NRT types. Do you have any
>> proof that this won't work?
>>
>> Here's the problem. Both TLOG and PULL replicas will lag behind the
>> master. The polling is pretty
>> quick IIRC, but there'll still be a lag. Plus that'll copy your
>> indexes around as time passes, perhaps
>> quite a bit.
>>
>> And you're saying that your indexing load is much less than your query
>> load, so I'd test NRT
>> before assuming you need different replica types.
>>
>> Solr 7.3 is pretty stable, although 7.3.1 will be released soon, I'd
>> wait for that.
>>
>> Best,
>> Erick
>>
>> On Wed, Apr 25, 2018 at 11:52 PM, kumar gaurav <kg2...@gmail.com> wrote:
>> > Hi
>> >
>> > I am setting up new cluster in solr cloud 7.3 . i am just confused to
>> make
>> > a combination of replica types . Real time indexing update/delete request
>> > will coming in cluster continuously but very less than select query .
>> > Cluster will need to handle high concurrent select queries than update.
>> >
>> > I am thinking to create combination of two TLOG replica + one PULL
>> replica
>> >  per shard because one NRT + one TLOG + one PULL is not recommended as
>> per
>> > documentation
>> >
>> > source -
>> > https://lucene.apache.org/solr/guide/7_3/shards-and-
>> indexing-data-in-solrcloud.html#recovery-with-pull-replicas
>> >
>> > Anyone please suggest if i am wrong ?
>> >
>> > Also i want to know if solr 7.3 version is stable or not for cloud mode ?
>> >
>> >
>> > Thanks in advance :)
>>

Reply via email to