1. Query DSL vs Query String

there are 2 ways of querying
1. using query DSL
2. using query string

1-1. Using query DSL

2 types
1. leaf query : search for values within particular fields
2. compound query : contain multiple leaf queries or compound queries

• most flexible and common way of writing search query

GET /product/_search
{
    "query" : {
        "match" : {
            "description" : {
                "value" : "red wine"
            }
        }
    }
}

• query object in _search API contains query definition

1-2. Using query string

GET /product/_search?q=name:pasta

name = field name
pasta = value

2. How search query works

• search query reaches node with shard b, which becomes coordinating node
• coordinating node then queries nodes with different shards

2-1. Relevance Scoring

full text search results are sorted by their relevance score

• relevance scoring is what differenciates Elasticsearch from RDB
• until recently TF/IDF was used, but now BM 25 is used
  • better handling of stop words using nonlinear term frequency saturation
  • improving field-length norm factor by treating each field separately instead of just treating a field in the same way across all docs

⚠️ IDF might be different for relevance scoring as by default, scoring is per shard based

3. Search Contexts

• query clause can be executed on 2 contexts

1. Query Context : how well do docs match the query (match + relevance scoring)
2. Filter Context : do docs match? (only match)

4. Full text vs term search

• term level : search for exact value, input is not processed by analyzer

• full text : search query is analyzed before matching. Analyzer specified by the field is used, if none, standard analyzer is used