如何实现ArangoSearch-View仅检索指定链接集合的目标字段

现有如下ArangoSearch View定义，其关联的col_person与col_hotel两个集合均包含名为f_name的字段：

{
  "cleanupIntervalStep": 2,
  "writebufferSizeMax": 33554432,
  "commitIntervalMsec": 1000,
  "id": "291316",
  "consolidationPolicy": {
    "type": "tier",
    "segmentsBytesFloor": 2097152,
    "segmentsBytesMax": 5368709120,
    "segmentsMax": 10,
    "segmentsMin": 1,
    "minScore": 0
  },
  "consolidationIntervalMsec": 1000,
  "globallyUniqueId": "h506F923AD141/291316",
  "links": {
    "col_person": {
      "analyzers": [],
      "fields": {
        "f_name": {
          "analyzers": [
            "arabic_text_analyzer",
            "arabic_collation_analyzer",
            "identity",
            "text_en"
          ]
        }
      },
      "includeAllFields": false,
      "storeValues": "none",
      "trackListPositions": false
    },
    "col_hotel": {
      "analyzers": [],
      "fields": {
        "f_name": {
          "analyzers": [
            "arabic_text_analyzer",
            "arabic_collation_analyzer",
            "text_en",
            "identity"
          ]
        }
      },
      "includeAllFields": false,
      "storeValues": "none",
      "trackListPositions": false
    }
  },
  "writebufferIdle": 64,
  "primarySort": [],
  "primarySortCompression": "lz4",
  "writebufferActive": 0,
  "storedValues": [],
  "type": "arangosearch"
}

执行如下AQL查询时，ArangoDB会同时在两个集合的f_name字段中执行检索，需要调整为仅检索col_person.f_name字段，不匹配col_hotel集合的同名字段：

LET QR1 = (
    FOR doc_asview_global1 IN asview_global1
    SEARCH
    ANALYZER(
        Like(doc_asview_global1.f_name, "%Regional%"),"identity") OR
    ANALYZER(
        LEVENSHTEIN_MATCH(doc_asview_global1.f_name, "Regional",1,true),"arabic_text_analyzer") OR
    ANALYZER(
        PHRASE(doc_asview_global1.f_name, "Regional"), "arabic_text_analyzer")
    LIMIT 10000
    SORT BM25(doc_asview_global1) DESC
    RETURN doc_asview_global1
)

RETURN QR1

核心逻辑是在SEARCH子句中显式限定检索的文档所属集合，使用ArangoDB内置的IS_SAME_COLLECTION()函数做集合匹配，该过滤条件会在ArangoSearch检索阶段直接生效，不会产生额外的后置过滤开销，性能最优。

修改后的AQL如下，仅需要在原有SEARCH条件最前方增加集合判断，和原有检索条件用AND连接即可：

LET QR1 = (
    FOR doc_asview_global1 IN asview_global1
    SEARCH
    // 限定仅检索col_person集合的文档
    IS_SAME_COLLECTION("col_person", doc_asview_global1)
    AND (
        ANALYZER(
            Like(doc_asview_global1.f_name, "%Regional%"),"identity") OR
        ANALYZER(
            LEVENSHTEIN_MATCH(doc_asview_global1.f_name, "Regional",1,true),"arabic_text_analyzer") OR
        ANALYZER(
            PHRASE(doc_asview_global1.f_name, "Regional"), "arabic_text_analyzer")
    )
    LIMIT 10000
    SORT BM25(doc_asview_global1) DESC
    RETURN doc_asview_global1
)

RETURN QR1

注意点 #

• 必须把多个OR连接的字段检索条件用括号包裹，再和集合过滤条件做AND运算，避免逻辑运算优先级出错导致过滤失效。
• 该方案不需要修改现有ArangoSearch View的配置，不需要重建索引，改完查询即可生效。
• 不要把集合过滤条件放到SEARCH子句外面用FILTER实现，那样会先检索两个集合的所有匹配文档再做过滤，会产生不必要的性能损耗。

内容的提问来源于stack exchange，提问作者peter bence