elasticsearch-搜索-parent/child

mapping

olap_patient

{
  "_routing": {
    "required": true
  },
  "_source": {
    "enabled": true
  },
  "properties": {
    "visits": {
      "ignore_above": 256,
      "type": "keyword"
    },
    "patient_id": {
      "ignore_above": 256,
      "type": "keyword"
    }
  }
}

olap_visit

{
  "_parent": {
    "type": "olap_patient"
  },
  "_routing": {
    "required": true
  },
  "_source": {
    "enabled": true
  },
  "properties": {
    "patient_id": {
      "doc_values": true,
      "ignore_above": 256,
      "type": "keyword"
    },
    "visit_id": {
      "doc_values": true,
      "ignore_above": 256,
      "type": "keyword"
    }
  }
}

index data

curl -XPUT "localhost:9200/test/olap_patient/2?routing=2" -d '{"visits":["2+001", "2+002", "2+003"], "patient_id":"2"}'
curl -XPUT "localhost:9200/test/olap_visit/2+001?routing=2?parent=2" -d '{"visit_id":"2+001", "patient_id":"2"}'
curl -XPUT "localhost:9200/test/olap_visit/2+002?routing=2?parent=2" -d '{"visit_id":"2+002", "patient_id":"2"}'
curl -XPUT "localhost:9200/test/olap_visit/2+003?routing=2?parent=2" -d '{"visit_id":"2+003", "patient_id":"2"}'

索引过程

olap_patient#1

[stored,indexed,omitNorms,indexOptions=DOCS<_routing:1>,
stored<_source:[7b 22 76 69 73 69 74 73 22 3a 5b 22 31 2b 30 30 31 22 2c 20 22 31 2b 30 30 32 22 2c 20 22 31 2b 30 30 33 22 5d 2c 20 22 70 61 74 69 65 6e 74 5f 69 64 22 3a 22 31 22 7d]>, indexed,omitNorms,indexOptions=DOCS<_type:olap_patient>,
docValuesType=SORTED_SET<_type:[6f 6c 61 70 5f 70 61 74 69 65 6e 74]>,
stored,indexed,omitNorms,indexOptions=DOCS<_uid:olap_patient#1>,
docValuesType=NUMERIC<_version:-1>,
indexed,omitNorms,indexOptions=DOCS,
docValuesType=SORTED_SET,
indexed,omitNorms,indexOptions=DOCS,
docValuesType=SORTED_SET,
indexed,omitNorms,indexOptions=DOCS,
docValuesType=SORTED_SET,
indexed,omitNorms,indexOptions=DOCS,
docValuesType=SORTED_SET,
indexed,omitNorms,indexOptions=DOCS<_field_names:_routing>,
indexed,omitNorms,indexOptions=DOCS<_field_names:_source>,
indexed,omitNorms,indexOptions=DOCS<_field_names:_type>,
indexed,omitNorms,indexOptions=DOCS<_field_names:_type>,
indexed,omitNorms,indexOptions=DOCS<_field_names:_uid>,
indexed,omitNorms,indexOptions=DOCS<_field_names:_version>,
indexed,omitNorms,indexOptions=DOCS<_field_names:visits>,
indexed,omitNorms,indexOptions=DOCS<_field_names:visits>,
indexed,omitNorms,indexOptions=DOCS<_field_names:visits>,
indexed,omitNorms,indexOptions=DOCS<_field_names:visits>,
indexed,omitNorms,indexOptions=DOCS<_field_names:visits>,
indexed,omitNorms,indexOptions=DOCS<_field_names:visits>,
indexed,omitNorms,indexOptions=DOCS<_field_names:patient_id>,
indexed,omitNorms,indexOptions=DOCS<_field_names:patient_id>,
docValuesType=SORTED<_parent#olap_patient:[31]>]

olap_visit#2+001

[stored,indexed,omitNorms,indexOptions=DOCS<_routing:2>,
stored<_source:[7b 22 76 69 73 69 74 5f 69 64 22 3a 22 32 2b 30 30 31 22 2c 20 22 70 61 74 69 65 6e 74 5f 69 64 22 3a 22 32 22 7d]>,
indexed,omitNorms,indexOptions=DOCS<_type:olap_visit>,
docValuesType=SORTED_SET<_type:[6f 6c 61 70 5f 76 69 73 69 74]>,
stored,indexed,omitNorms,indexOptions=DOCS<_uid:olap_visit#2 001>,
docValuesType=NUMERIC<_version:-1>,
indexed,omitNorms,indexOptions=DOCS,
docValuesType=SORTED_SET,
indexed,omitNorms,indexOptions=DOCS,
docValuesType=SORTED_SET,
indexed,omitNorms,indexOptions=DOCS<_field_names:_routing>,
indexed,omitNorms,indexOptions=DOCS<_field_names:_source>,
indexed,omitNorms,indexOptions=DOCS<_field_names:_type>,
indexed,omitNorms,indexOptions=DOCS<_field_names:_type>,
indexed,omitNorms,indexOptions=DOCS<_field_names:_uid>,
indexed,omitNorms,indexOptions=DOCS<_field_names:_version>,
indexed,omitNorms,indexOptions=DOCS<_field_names:visit_id>,
indexed,omitNorms,indexOptions=DOCS<_field_names:visit_id>,
indexed,omitNorms,indexOptions=DOCS<_field_names:patient_id>,
indexed,omitNorms,indexOptions=DOCS<_field_names:patient_id>,
docValuesType=SORTED<_parent#olap_patient:[32]>]

注意索引字段_parent#olap_patient:[32]

search

/test2/olap_visit/_search?pretty -d '{"query":{"has_parent":{"type":"olap_patient", "query":{"terms":{"visits":["3+002"]}}}}}'

通过TermQuery（visits:["3+002"]）+ TermQuery（_type:olap_patient）得到olap_patient结果集；对每一个olap_patient，通过DocValue（_parent#olap_patient）拿到olap_patient的_parent#olap_patient字段存在doc_values里的序号；对所有序号构建一个BitSets，这个过程由org.apache.lucene.search.join.GlobalOrdinalsCollector.OrdinalMapCollector完成；注意这里的BitSets和通常的倒排链归并不一样，这里仅作为random access filter使用
通过TermQuery（_type:olap_visit）得到olap_visit结果集；对每一个olap_visit，通过DocValue（_parent#olap_patient）拿到olap_visit的_parent#olap_patient字段存在doc_values里的序号；判断该序号是否在第一个阶段生成的BitSets里；如果存在，则表示该olap_visit的parent满足has_parent查询条件，该olap_visit即命中

总的来说，ES构建了一个DocValues字段，不管是父文档还是子文档，都将父文档的id存在了该DocValues字段里，这样做可以在查询得到parent结果集后将parent的doc_values取出来，作为child查询结果集的一个过滤条件

为什么要基于doc_values去实现呢？因为doc_values内部维护了两个对象，一个是ordinals，一个是values；ordinals可以查找doc_id对应的序数，values可以查找序数对应的字段值；也就是说，doc_values可以将文档集对应的值集用bitset来表示，方便过滤及聚合

tips

增加一个doc_type时不能指定已存在的非parent的doc_type为_parent，因为ES会为父子文档都创建joinField的DocValues索引。异常信息：{"error":{"root_cause":[{"type":"illegal_argument_exception","reason":"can't add a _parent field that points to an already existing type, that isn't already a parent"}],"type":"illegal_argument_exception","reason":"can't add a _parent field that points to an already existing type, that isn't already a parent"},"status":400}