No reindexing of NumericFieldValues?


(Valentin Pletzer) #1

According to this article http://shaierera.blogspot.com/2013/09/boosting-documents-in-lucene.html Lucene allows for updating NumericFieldValues without reindexing the whole document. The help doc on Partial Updates in Elasticsearch implies otherwise. Can anyone explain how it is solved in Elasticsearch?


(Adrien Grand) #2

Elasticsearch doesn't expose this feature for now. Note that the understanding that even if the document does not need to be reindexed, this feature in Lucene requires doc values for all documents in the same segment to be rewritten. So while this has benefits in certain cases, it definitely has costs too, they are just different.


(Adrien Grand) #3

Something else that I forgot to mention: this works for doc values but not for indexed fields. So you could not search on them, only sort or aggregate.


(Valentin Pletzer) #4

Thanks.

I am currently looking for the best way to integrate popularity data into elasticsearch. We would like to use pageview data which is changing constantly (eg daily).

Another idea of mine is to use a parent-child-relation in order to move the changing data into another "document".

What do you think? Is there any literature/best practise documentation on that topic?


(Adrien Grand) #5

This is a hard issue. Parent/child could work but it typically makes things much slower too. Also note that parent/child will only work better than regular updates if documents are very large, otherwise it would be better to just use the _update API.


(system) #6