CTS范围查询与SPARQL查询性能差异原因咨询

观察到的性能差异 #

• CTS字段查询：耗时0.8ms，依赖字段索引支持，查询代码：

cts:field-values("productid", (), (), cts:and-query(
              (
                cts:field-value-query("countryCode", "us", ("unstemmed","case-insensitive", "whitespace-insensitive", "punctuation-insensitive", "diacritic-insensitive")),
                cts:field-value-query("status", "published", ("unstemmed","case-insensitive", "whitespace-insensitive", "punctuation-insensitive", "diacritic-insensitive"))
              )
          ))

• SPARQL查询：耗时18ms，依赖TDE生成的三元组支持，查询代码：

SELECT ?productid
FROM <product>
WHERE {
  ?productid <status> <Published>;
             <countryCode> <US>.
}

对应的TDE配置：

<?xml version="1.0" encoding="UTF-8"?>
<template xmlns="http://marklogic.com/xdmp/tde">
    <context>product</context>
    <enabled>true</enabled>
    <collections>
        <collection>product</collection>
    </collections>
    <triples>
        <triple>
            <subject>
                <val>sem:iri(productid)</val>
                <invalid-values>ignore</invalid-values>
            </subject>
            <predicate>
                <val>sem:iri(xs:string("languageCode"))</val>
                <invalid-values>ignore</invalid-values>
            </predicate>
            <object>
                <val>sem:iri(languageCode)</val>
                <invalid-values>ignore</invalid-values>
            </object>
        </triple>
        <triple>
            <subject>
                <val>sem:iri(productid)</val>
                <invalid-values>ignore</invalid-values>
            </subject>
            <predicate>
                <val>sem:iri(xs:string("countryCode"))</val>
                <invalid-values>ignore</invalid-values>
            </predicate>
            <object>
                <val>sem:iri(fn:normalize-space(xs:string(countryCode)))</val>
                <invalid-values>ignore</invalid-values>
            </object>
        </triple>
        <triple>
            <subject>
                <val>sem:iri(productid)</val>
                <invalid-values>ignore</invalid-values>
            </subject>
            <predicate>
                <val>sem:iri(xs:string("status"))</val>
                <invalid-values>ignore</invalid-values>
            </predicate>
            <object>
                <val>sem:iri(fn:normalize-space(xs:string(status)))</val>
                <invalid-values>ignore</invalid-values>
            </object>
        </triple>
        <triple>
            <subject>
                <val>sem:iri(productid)</val>
                <invalid-values>ignore</invalid-values>
            </subject>
            <predicate>
                <val>sem:iri(xs:string("created"))</val>
                <invalid-values>ignore</invalid-values>
            </predicate>
            <object>
                <val>sem:iri(audit/created)</val>
                <invalid-values>ignore</invalid-values>
            </object>
        </triple>
        <triple>
            <subject>
                <val>sem:iri(productid)</val>
                <invalid-values>ignore</invalid-values>
            </subject>
            <predicate>
                <val>sem:iri(xs:string("createdBy"))</val>
                <invalid-values>ignore</invalid-values>
            </predicate>
            <object>
                <val>sem:iri(audit/createdBy)</val>
                <invalid-values>ignore</invalid-values>
            </object>
        </triple>
        <triple>
            <subject>
                <val>sem:iri(productid)</val>
                <invalid-values>ignore</invalid-values>
            </subject>
            <predicate>
                <val>sem:iri(xs:string("updated"))</val>
                <invalid-values>ignore</invalid-values>
            </predicate>
            <object>
                <val>sem:iri(audit/updated)</val>
                <invalid-values>ignore</invalid-values>
            </object>
        </triple>
        <triple>
            <subject>
                <val>sem:iri(productid)</val>
                <invalid-values>ignore</invalid-values>
            </subject>
            <predicate>
                <val>sem:iri(xs:string("updatedBy"))</val>
                <invalid-values>ignore</invalid-values>
            </predicate>
            <object>
                <val>sem:iri(audit/updatedBy)</val>
                <invalid-values>ignore</invalid-values>
            </object>
        </triple>
    </triples>
</template>

性能差异的核心原因 #

1. 索引利用效率差异 #

CTS查询直接基于字段索引执行：

• cts:field-values 可以直接从预构建的字段值索引中提取匹配结果，无需遍历文档或三元组存储，多维度匹配规则（大小写不敏感、忽略空格等）已在索引构建阶段处理完成，查询时直接命中索引。
• SPARQL查询依赖三元组存储：即使有三元组索引，引擎仍需遍历匹配两个三元组模式的结果，再执行交集合并操作，索引查询路径比字段索引更复杂。

2. TDE转换带来的隐性开销 #

从TDE配置可见，生成countryCode和status三元组时执行了fn:normalize-space(xs:string(...))转换：

• 这些转换在文档插入时完成，但三元组存储的是处理后的IRI值；SPARQL查询使用<Published>和<US>匹配，若存在隐式格式差异，会额外增加匹配逻辑的耗时。

3. 查询引擎执行模型差异 #

CTS是MarkLogic原生底层查询接口，针对文档索引做了深度优化，执行路径短，直接操作索引数据结构返回结果。
SPARQL引擎需要处理RDF语义模型，包含查询解析、逻辑计划优化、三元组匹配等多步操作，即使简单查询也会有额外的引擎层开销。

4. 结果返回格式的额外开销 #

cts:field-values直接返回匹配的字段值列表，数据结构简单，无需额外转换。
SPARQL查询需要将三元组匹配结果映射为SPARQL绑定变量格式，返回时还要符合SPARQL结果集规范，序列化和格式转换会增加耗时。

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