Replicas, balancing and cache


(Rodrigo Farto Alonso) #1

Hi.

With our current configuration we are not capable of taking full advantage
of the cache.

We have 4 replicas, 15 nodes and node filter cache configuration. We need
to execute a filter 5 consecutive times in order for it to be cached (we
have run several test combining different filters and we are pretty sure
that's how it is working)

So I guess there is some kind of round-robin balancing, the easy way out
will be to set a index filter cache but as it is a not recommended featureI wonder if there is a way to change the balancing configuration (to make
sure all the petitions in the same session are serviced by the same
node-replica)

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.


(Mark Walkom) #2

We've not encountered this as yet so I'm not 100% on this next bit, but
would
http://exploringelasticsearch.com/book/advanced-techniques/routing.html be
helpful?

Regards,
Mark Walkom

Infrastructure Engineer
Campaign Monitor
email: markw@campaignmonitor.com
web: www.campaignmonitor.com

On 31 October 2013 22:01, Rodrigo Farto Alonso <rodrigofartoalonso@gmail.com

wrote:

Hi.****

With our current configuration we are not capable of taking full
advantage of the cache.

We have 4 replicas, 15 nodes and node filter cache configuration. We need
to execute a filter 5 consecutive times in order for it to be cached (we
have run several test combining different filters and we are pretty sure
that's how it is working)

So I guess there is some kind of round-robin balancing, the easy way out
will be to set a index filter cache but as it is a not recommended
feature
I wonder if there is a way to change the balancing configuration
(to make sure all the petitions in the same session are serviced by the
same node-replica)****

--
You received this message because you are subscribed to the Google Groups
"elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.


(Rodrigo Farto Alonso) #3

We are already using routing our query looks like:

curl -s -XPOST
'http://lpels001:9210/ebes_index/movimientos/_search?routing=XXXXXXXXX'
--data-binary {

"query": {

"filtered": {

  "query": {

    "bool": {

      "should": [

        {

          "prefix": {

            "descripcionConcepto": "SOME TEXT"

          }

        }

      ]

    }

  },

  "filter": {

    "bool": {

      "must": {

        "terms": {

            "codigoInternoDelContratoIuc": [

             XXXXXXXXX #As in the routing parameter

            ]

        }

      }

    }

  }

}

},

"facets": {

"facetTextoPredictivo": {

  "terms": {

    "fields": [

      "descripcionConcepto.raw",

      "descripcionMovimiento.raw",

      "comentarios.raw",

      "notasDeUsuario.notas.raw"

    ],

    "size": 100

  }

}

},

"size": 0

}

That particular query takes around 50 seconds when not routed, 8s when
routed but not cached and 20 ms when cached (we have over 3 TB of data).

So goal is for the query to take 8s just the first time, and not 8s for the
first execution against each replica.

On Thursday, October 31, 2013 12:09:56 PM UTC+1, Mark Walkom wrote:

We've not encountered this as yet so I'm not 100% on this next bit, but
would
http://exploringelasticsearch.com/book/advanced-techniques/routing.htmlbe helpful?

Regards,
Mark Walkom

Infrastructure Engineer
Campaign Monitor
email: ma...@campaignmonitor.com <javascript:>
web: www.campaignmonitor.com

On 31 October 2013 22:01, Rodrigo Farto Alonso <rodrigofa...@gmail.com<javascript:>

wrote:

Hi.****

With our current configuration we are not capable of taking full
advantage of the cache.

We have 4 replicas, 15 nodes and node filter cache configuration. We need
to execute a filter 5 consecutive times in order for it to be cached (we
have run several test combining different filters and we are pretty sure
that's how it is working)

So I guess there is some kind of round-robin balancing, the easy way out
will be to set a index filter cache but as it is a not recommended
feature
I wonder if there is a way to change the balancing
configuration (to make sure all the petitions in the same session are
serviced by the same node-replica)****

--
You received this message because you are subscribed to the Google Groups
"elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to elasticsearc...@googlegroups.com <javascript:>.
For more options, visit https://groups.google.com/groups/opt_out.

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.


(system) #4