Internal details on how ADDREPLICA executes?

[Nawab Zada Asad Iqbal <kh...@gmail.com>]

Hi,

I am considering using SolrCloud and enable autoADDREPLICA.
I am curious on how long does it take for SolrCloud to setup the replica.
Before the new replica can start to serve queries, it needs to copy all the
documents in the current leader and also index whatever new traffic is
arriving. How is this state tracked?  I assume SolrCloud knows about this
transient state between a new replica creation and being ready to server
queries and handles it appropriately.

Do I need to consider any performance issues after executing addReplica
(i.e., will it affect the current leader's index or querying response time
when older segment files are being copied to the new replica).


Thanks
Nawab

Re: Internal details on how ADDREPLICA executes?

[Nawab Zada Asad Iqbal <kh...@gmail.com>]

Thanks Erick!

On Thu, Aug 2, 2018 at 1:26 PM, Erick Erickson <er...@gmail.com>
wrote:

> Oh my, I see confusion on the horizon. _Which_ autoaddreplica?
>
> Pre 7x and autoscaling, autoAddReplica is all about HDFS and
> spinning up a new Solr instance that points to an existing index
> so there's no impact on the leader (other than normal peer sync
> if you're actively indexing).
>
> Autoscaling, on the other hand (7x) spins up a new replica and
> it goes through sync before it's ready to server queries. This is
> really just an ADDREPLICA and you can test the impacts on
> your particular installation by issuing that collections API
> command.
>
> Best,
> Erick
>
> On Thu, Aug 2, 2018 at 12:20 PM, Nawab Zada Asad Iqbal <kh...@gmail.com>
> wrote:
> > Hi,
> >
> > I am considering using SolrCloud and enable autoADDREPLICA.
> > I am curious on how long does it take for SolrCloud to setup the replica.
> > Before the new replica can start to serve queries, it needs to copy all
> the
> > documents in the current leader and also index whatever new traffic is
> > arriving. How is this state tracked?  I assume SolrCloud knows about this
> > transient state between a new replica creation and being ready to server
> > queries and handles it appropriately.
> >
> > Do I need to consider any performance issues after executing addReplica
> > (i.e., will it affect the current leader's index or querying response
> time
> > when older segment files are being copied to the new replica).
> >
> >
> > Thanks
> > Nawab
>

Re: Internal details on how ADDREPLICA executes?

[Erick Erickson <er...@gmail.com>]

Oh my, I see confusion on the horizon. _Which_ autoaddreplica?

Pre 7x and autoscaling, autoAddReplica is all about HDFS and
spinning up a new Solr instance that points to an existing index
so there's no impact on the leader (other than normal peer sync
if you're actively indexing).

Autoscaling, on the other hand (7x) spins up a new replica and
it goes through sync before it's ready to server queries. This is
really just an ADDREPLICA and you can test the impacts on
your particular installation by issuing that collections API
command.

Best,
Erick

On Thu, Aug 2, 2018 at 12:20 PM, Nawab Zada Asad Iqbal <kh...@gmail.com> wrote:
> Hi,
>
> I am considering using SolrCloud and enable autoADDREPLICA.
> I am curious on how long does it take for SolrCloud to setup the replica.
> Before the new replica can start to serve queries, it needs to copy all the
> documents in the current leader and also index whatever new traffic is
> arriving. How is this state tracked?  I assume SolrCloud knows about this
> transient state between a new replica creation and being ready to server
> queries and handles it appropriately.
>
> Do I need to consider any performance issues after executing addReplica
> (i.e., will it affect the current leader's index or querying response time
> when older segment files are being copied to the new replica).
>
>
> Thanks
> Nawab