Is it possible to retrieve only doc id and score from Elasticsearch without performing the fetch phase of search? - elasticsearch

Question

Understanding "Query Then Fetch" shows that an Elasticsearch query is a two step process of query (find/score/sort matching documents from all servers) and fetching (go back to the servers and collect the content of the matching documents).
Is there a way to retrieve only a list of sorted doc_id and score but avoid the fetch? I know that fetch can be avoided by setting size to 0... but I still need the matching docs and their scores and that would return none.
I figure I might be able to turn off _source, but I'm not sure that would work if, for example, the query portion of the search only knows the internal doc_id and needs to go and retrieve the public doc_id.

Answer 1

GET /_search
{
"_source": false,
"query" : {
"term" : { "user" : "kimchy" }
}
}
off course, you have to use your own ids, not auto-generated ones

Answer 2

The scores are separated from the docs' sources so I don't see why a fetch would be necessary to retrieve them.
You can surely turn off _source and then also sort by _id, like so:
GET your_index/_search
{
"_source": false,
"size": 200,
"sort": [
{
"_id": {
"order": "asc"
}
},
{
"_score": {
"order": "desc"
}
}
]
}
Interestingly enough, sorting the response by doc's _source field seems to be ~3x faster than sorting by the inner _id (Contrary to what I expected). I've tested this w/ quite a small index -- ~1.5M docs. I wonder what you get when you run
GET your_index/_search?request_cache=false
{
"_source": false,
"size": 200,
"sort": [
{
"_id": {
"order": "asc"
}
}
]
}
and then replace _id with another doc's _source sortable field.

Answer 3

Indeed by setting size to 0, we will be skipping the fetch phase. In all other cases, if have even a single hit, the fetch phase will be executed and there is no way to skip it.
As you correctly noted, the query phase doesn't know the real _ids of matched documents, only their internal doc ids on respected shard. As a part of the fetch phase we will retrieve those _ids, which are stored as a stored field in Lucene. _source is a separate stored field from _id, which is also loaded during the fetch phase. But to speed the fetch phase, you can disable loading _source if you don't need it. Being a separate field from _id, disabling _source doesn't affect the correct loading of _ids.