Hi
A negative query is a subtraction from a set of matched documents. With
(*:* NOT user_id:0) you are subtracting from the set of all documents in
the index first, then intersecting with the documents that match the other
clauses. With (NOT user_id:0) you are subtracting directly from the smaller
set of documents that match the other query clauses.
It might be interesting to compare your benchmarks with a filter query for
-user_id:0 instead, especially if that's a clause that's used repeatedly in
different queries.
Thomas
Op vr 6 jan. 2023 om 09:00 schreef Changcheng Shao <changchen...@gmail.com>:
> Hi, Solr team
> I am using Solr 8.11, and I want to ask some questions about NOT query.
>
> My original query is:
> network_id:379619 AND (object_type:("ssp_deal" OR "ssp_buyer_group" OR
> "ssp_buyer")) AND (network_id:("379619")) AND (*:* NOT user_id: 0)
> And I opened the debugQuery, the log is:
> "rawquerystring":"network_id:379619 AND (object_type:(\"ssp_deal\" OR
> \"ssp_buyer_group\" OR \"ssp_buyer\")) AND (network_id:(\"379619\")) AND
> (*:* NOT user_id: 0)",
> "querystring":"network_id:379619 AND (object_type:(\"ssp_deal\" OR
> \"ssp_buyer_group\" OR \"ssp_buyer\")) AND (network_id:(\"379619\")) AND
> (*:* NOT user_id: 0)",
> "parsedquery":"+network_id:379619 +(object_type:ssp_deal
> object_type:ssp_buyer_group object_type:ssp_buyer) +network_id:379619
> +(MatchAllDocsQuery(*:*) -user_id:0)",
>
>
> Then I changed the query to:
> network_id:379619 AND (object_type:("ssp_deal" OR "ssp_buyer_group" OR
> "ssp_buyer")) AND (network_id:("379619")) AND NOT user_id: 0
> And the debug log is:
> "rawquerystring":"network_id:379619 AND (object_type:(\"ssp_deal\" OR
> \"ssp_buyer_group\" OR \"ssp_buyer\")) AND (network_id:(\"379619\")) AND
> NOT user_id: 0",
> "querystring":"network_id:379619 AND (object_type:(\"ssp_deal\" OR
> \"ssp_buyer_group\" OR \"ssp_buyer\")) AND (network_id:(\"379619\")) AND
> NOT user_id: 0",
> "parsedquery":"+network_id:379619 +(object_type:ssp_deal
> object_type:ssp_buyer_group object_type:ssp_buyer) +network_id:379619
> -user_id:0",
>
> And the difference of queries are (*:* NOT user_id: 0) and NOT user_id: 0
> the parsedquery are : +(MatchAllDocsQuery(*:*) -user_id:0) and
> -user_id:0.
> And the results of both queries are the same.
>
> Then I test, the query which used (NOT user_id: 0) took less time than used
> (*:* NOT user_id: 0)
> So I want to ask, I hope you can analyze theoretically which query takes
> less time to select?
>
> Hope to get your response, thanks!
>
