Elastic 2.x function_score week performance with script

Elastic Stack Elasticsearch
1

Hi

We have just migrated from elastic version 1.7 to version 2.2, and we have encountered some extremely weird behavior.

We have a score function with a script:

{ "script_score": { "params": { "fields": [ "country_code_volumes/840", "country_code_volumes/826", "country_code_volumes/036", "country_code_volumes/124" ], "div_f": "volume", "multiply": 1, "add": 0 }, script": "doc_v=doc[div_f].value; if (doc_v == 0) {return score;}; score=0; for(field in fields) { score += doc[field].value;}; score /= (doc_v*1d); ((score + add) * multiply)" } }

we notice that taking this function with a query takes x10 times slower then if for example there is only one element in the fields array.

we tried to narrow it down more.. and we have come to a conclusion that the line inside the for loop:
score += doc[field].value is the reason for the slow performance..
for example if I change that line to: score = doc[field].value the query return 10 times faster then with the + operator..

Anyone have any idea why is that?

2

Not a groovy expert, but I suspect this has something to do with type promotion. Maybe groovy promotes you numbers to big decimals in order to perform the addition.

3

Not sure that is it because if we take the same script, and instead of preforming the for loop we do something like this:
score += doc[fields[0]].value; ..... ... score += doc[fields[last]].value;
The performance is still good.