Question

我正在开发一个应用程序，需要在2小时内获取多达100万份文档

我们正在使用Java客户端API和结构化查询来执行搜索。但是，查询仍然很慢。

代码如下：

def fetchPostMessages(dbParam: DbParam): Page = {

  val queryManager = dbClient.newQueryManager()
  val sqb: StructuredQueryBuilder = queryManager.newStructuredQueryBuilder()

  log.info(s "Fetching post messages from database for params: {}", dbParam)

  val modifiedQueryDef = dbParam.param.map {
    param => {
      sqb.and(
        sqb.word(sqb.jsonProperty(status), toBeReported),
        sqb.word(sqb.jsonProperty(dataCategory), "dataCategory1"),
        sqb.range(sqb.jsonProperty(creationDate), marklogicDateFormat.name, Operator.LE, DateUtil.printFpmlDateTime(param.messagesTime)))

    }
  }.getOrElse(sqb.and(sqb.word(sqb.jsonProperty(status.name), toBeReported.name)))

  modifiedQueryDef.setCollections(XmlConstants.ItracMessageTypes.OUTPUT_MESSAGE.name)
  modifiedQueryDef.setOptionsName(sortOption)
  search(modifiedQueryDef, dbParam.pageNum, dbParam.batchSize)
}


private def search(queryDef: QueryDefinition, startIndex: Int, batchSize: Int): Page = {

  val dataList: ListBuffer[Document] = new ListBuffer()
  val jsonDocManager = dbClient.newJSONDocumentManager()
  jsonDocManager.setMetadataCategories(Metadata.ALL)
  jsonDocManager.setPageLength(
    if (batchSize < pageLength) batchSize
    else pageLength)
  val documentPage = jsonDocManager.search(queryDef, startIndex);
  dataList.++ = (extractContent(documentPage))

  val totalSize = documentPage.getTotalSize
  log.info(s "Total documents to be reported : ${totalSize}")
  var pageSize = documentPage.getPageSize

  while (pageSize < batchSize && pageSize <= totalSize) {

    if (batchSize - pageSize < pageSize)
      jsonDocManager.setPageLength(batchSize - pageSize)

    var newDocPage = jsonDocManager.search(queryDef, pageSize + 1)
    dataList.++ = (extractContent(newDocPage))
    pageSize = pageSize + newDocPage.getPageSize
  }

  log.info("Total messages fetched are : {}", dataList.size)
  Page(startIndex, totalSize - batchSize, dataList.to[collection.immutable.Seq])

}

排序选项包括：

<search:options xmlns:search="http://marklogic.com/appservices/search">
  <search:sort-order type="xs:string" direction="ascending">
    <search:json-property>subdomLvl1</search:json-property>
  </search:sort-order>
  <search:sort-order type="xs:string" direction="ascending">
    <search:json-property>trdId</search:json-property>
  </search:sort-order>
  <search:sort-order type="xs:string" direction="ascending">
    <search:json-property>validStartDate</search:json-property>
  </search:sort-order>
  <search:sort-order type="xs:string" direction="ascending">
    <search:json-property>ver</search:json-property>
  </search:sort-order>
  <search:sort-order type="xs:string" direction="ascending">
    <search:json-property>reportStatus</search:json-property>
  </search:sort-order>
</search:options>

对数据库建立索引如下：

元素范围索引用于 - status，dataCategory和creationDate以及所有排序选项

Answer 1

如果进程不需要文档元数据，请考虑使用jsonDocManager.clearMetadataCategories（）而不是jsonDocManager.setMetadataCategories（Metadata.ALL）进行配置。这种方法将减少服务器和客户端的工作量，并减少传输的数据。

可以通过测试newDocPage.hasNextPage（）来简化循环 - 请参阅：

http://docs.marklogic.com/guide/java/bulk#id_21619

客户端是否可以在文档到达时将文档流式传输到消费流程，而不是在单个列表中累积所有百万个文档？这肯定会提高吞吐量。

您可能还会考虑使用Data Movement SDK来读取多个线程中的文档：

http://docs.marklogic.com/guide/java/data-movement#id_60613

http://docs.marklogic.com/javadoc/client/com/marklogic/client/datamovement/QueryBatcher.html

希望有帮助，

MarkLogic结构化查询性能

1 个答案: