Huge performance hit when using fuzzy and highlights


(José de Zárate) #1

index with about 150k documents.

when using fuzzy search and highlight on three fields, there is a huge
performance impact. what is 1 second without highligt turnos into a 7
second query when enabling highlights on three fieds.

thought it was related
to http://elasticsearch-users.115913.n3.nabble.com/highlighting-just-fields-td3916391.html
, but tried with 0.19.3 and have had same results.

willing to post more detailed info.

txs!!!


(José de Zárate) #2

Is this behavior to be expected (big performance loss when combining fuzzy
and highlighting) ? Or is it bug-related or just the way it is? some
tweaking available?

On Tuesday, May 22, 2012 8:22:16 PM UTC-4, JoeZ99 wrote:

index with about 150k documents.

when using fuzzy search and highlight on three fields, there is a huge
performance impact. what is 1 second without highligt turnos into a 7
second query when enabling highlights on three fieds.

thought it was related to
http://elasticsearch-users.115913.n3.nabble.com/highlighting-just-fields-td3916391.html, but tried with 0.19.3 and have had same results.

willing to post more detailed info.

txs!!!


(Shay Banon) #3

First, fuzzy is not amazing when it comes to performance wise, second, can
you try and use the fast vector highlighter by enabling term vectors in teh
mappings for the fields you index on?

On Wed, May 23, 2012 at 5:17 PM, JoeZ99 jzarate@gmail.com wrote:

Is this behavior to be expected (big performance loss when combining fuzzy
and highlighting) ? Or is it bug-related or just the way it is? some
tweaking available?

On Tuesday, May 22, 2012 8:22:16 PM UTC-4, JoeZ99 wrote:

index with about 150k documents.

when using fuzzy search and highlight on three fields, there is a huge
performance impact. what is 1 second without highligt turnos into a 7
second query when enabling highlights on three fieds.

thought it was related to http://elasticsearch-users.**
115913.n3.nabble.com/**highlighting-just-fields-**td3916391.htmlhttp://elasticsearch-users.115913.n3.nabble.com/highlighting-just-fields-td3916391.html, but tried with 0.19.3 and have had same results.

willing to post more detailed info.

txs!!!


(José de Zárate) #4

I'm afraid so. this is a simplified mapping I'm using:

"types": {
...
"typeA": {
"mapping": {
"fieldA": {
"type":"string",
"term_vector" : "with_positions_offsets"
},
...
}
}

and when searching in "typeA":
{
"query": {
"fuzzy": {
"_all": "whatever"
}
},
"highlight": {
"fields": {
"fieldA":{}
}
}
}

So I don't have much hope left. The only thing I haven't tried yet is to
set "store=true" for the "fieldA" field (I've read highighter uses it if
available). I don't know if there is such a difference between using that
and _store

On Friday, May 25, 2012 4:36:44 PM UTC-4, kimchy wrote:

First, fuzzy is not amazing when it comes to performance wise, second, can
you try and use the fast vector highlighter by enabling term vectors in teh
mappings for the fields you index on?

On Wed, May 23, 2012 at 5:17 PM, JoeZ99 wrote:

Is this behavior to be expected (big performance loss when combining
fuzzy and highlighting) ? Or is it bug-related or just the way it is? some
tweaking available?

On Tuesday, May 22, 2012 8:22:16 PM UTC-4, JoeZ99 wrote:

index with about 150k documents.

when using fuzzy search and highlight on three fields, there is a huge
performance impact. what is 1 second without highligt turnos into a 7
second query when enabling highlights on three fieds.

thought it was related to http://elasticsearch-users.**
115913.n3.nabble.com/**highlighting-just-fields-**td3916391.htmlhttp://elasticsearch-users.115913.n3.nabble.com/highlighting-just-fields-td3916391.html, but tried with 0.19.3 and have had same results.

willing to post more detailed info.

txs!!!


(José de Zárate) #5

An update. Reindexed the 130k documents with store set to true for the
field to be highlighted, and gained a 20%, from 7 seconds to aprox. 5.

any ideas on when lucene 4.0 is going to be out?

On Tuesday, May 22, 2012 8:22:16 PM UTC-4, JoeZ99 wrote:

index with about 150k documents.

when using fuzzy search and highlight on three fields, there is a huge
performance impact. what is 1 second without highligt turnos into a 7
second query when enabling highlights on three fieds.

thought it was related to
http://elasticsearch-users.115913.n3.nabble.com/highlighting-just-fields-td3916391.html, but tried with 0.19.3 and have had same results.

willing to post more detailed info.

txs!!!


(system) #6