KMongo协程查询环境性能差异问题排查求助

问题概述 #

使用KMongo的CoroutineCollection查询数据库时，本地环境耗时小于0.5秒，但在开发环境中耗时长达10-25秒，性能差异显著。

运行环境与调用栈 #

应用基于Spring Boot Netty（支持Reactor协程）运行，通过DGS框架搭建GraphQL端点，简化调用栈如下：

DGSQuery suspend fun -> suspend fun |
                                    | -> suspend fun KMongo first CoroutineCollection.find()
                                    | -> suspend fun KMongo second CoroutineCollection.find()

关键发现与怀疑点 #

• KMongo默认batchSize为2，导致底层会执行多次查询拉取数据，增加网络交互开销
• 怀疑代码在单线程上运行且该线程被整个应用共享，开发环境因客户端请求和定时任务更多，更容易出现线程拥堵，进而拉长耗时

日志分析 #

从日志可以看到：

[ctor-http-nio-1][][] c.x.d.a.wellreport.***Resolver           : wellReports run
[ctor-http-nio-1][][] c.x.d.d.wellreport.***Service            : all***Provider.invoke() start
[ctor-http-nio-1][][] d.m.w.Mongo***Repository                 : fetchAll started
[tter-2-thread-1][][] org.mongodb.driver.protocol.command      : Sending command '{"find": "***", "filter": {}, "batchSize": 100, "$db": "***", "lsid": {"id": {"$binary": {"base64": "hdGYjs3YSX6K5y2YJzUJkw==", "subType": "04"}}}}' with request id 56 to database *** on connection [connectionId{localValue:22, serverValue:1902}] to server localhost:27017
[ntLoopGroup-3-6][][] org.mongodb.driver.protocol.command      : Execution of command with request id 56 completed successfully in 13.32 ms on connection [connectionId{localValue:22, serverValue:1902}] to server localhost:27017
[ntLoopGroup-3-6][][] org.mongodb.driver.operation             : Received batch of 54 documents with cursorId 0 from server localhost:27017
[ntLoopGroup-3-6][][] d.m.w.Mongo***Repository                 : fetchAll finished
[ntLoopGroup-3-6][][] c.x.d.d.wellreport.***Service            : all***Provider.invoke() finished
[ntLoopGroup-3-6][][] c.x.d.d.wellreport.***Service            : positive***Provider.invoke() start
[ntLoopGroup-3-6][][] w.Mongo***Repository                     : fetchAllPositive start
[tter-2-thread-1][][] org.mongodb.driver.protocol.command      : Sending command '{"find": "***", "filter": {"score": {"$gt": 0.0}}, "batchSize": 2, "$db": "***", "lsid": {"id": {"$binary": {"base64": "hdGYjs3YSX6K5y2YJzUJkw==", "subType": "04"}}}}' with request id 57 to database *** on connection [connectionId{localValue:22, serverValue:1902}] to server localhost:27017
[ntLoopGroup-3-6][][] org.mongodb.driver.protocol.command      : Execution of command with request id 57 completed successfully in 5.26 ms on connection [connectionId{localValue:22, serverValue:1902}] to server localhost:27017
[ntLoopGroup-3-6][][] org.mongodb.driver.operation             : Received batch of 2 documents with cursorId 6831622277590159417 from server localhost:27017

• 首次调用KMongo的find()后，线程从ctor-http-nio-1切换到ntLoopGroup-3-6，后续所有操作都在该线程上执行
• 第二次查询的batchSize为2，仅拉取到2条数据，且返回了非0的cursorId，说明还需要多次请求拉取剩余数据，这会显著增加总耗时

相关代码片段 #

Repository实现类代码：

class MongoXXXScoreRepository(private val collection: CoroutineCollection<MongoXXXScore>) : XXXScoreRepository {
    private val logger by logger()
    override suspend fun fetchAllPositive(): List<XXXScore> {
        logger.info("fetchAllPositive start")
        val result = collection.find(MongoXXXScore::score gt 0.0).toList().map { it.toXXXScore() }
        logger.info("fetchAllPositive start") // 注：此处日志重复，建议改为finished
        return result
    }
}

优化建议 #

• 调整batchSize：显式设置更大的batchSize（如100或根据实际数据量调整），减少网络请求次数，例如：

  collection.find(MongoXXXScore::score gt 0.0).batchSize(100).toList()
  

• 协程调度器优化：确保KMongo使用合适的协程调度器（如Dispatchers.IO）执行IO操作，避免阻塞Netty的事件循环线程。可通过配置KMongo的客户端参数指定调度器，或在调用时用withContext(Dispatchers.IO)包裹查询逻辑
• 并行查询优化：如果两个find查询无依赖关系，可使用async/await并行发起，减少总耗时：

  val allData = async { firstCollection.find().toList() }
  val positiveData = async { secondCollection.find(MongoXXXScore::score gt 0.0).batchSize(100).toList() }
  val result = allData.await() + positiveData.await()
  

• 数据库连接池配置：检查MongoDB连接池的最大连接数设置，确保开发环境有足够的连接资源，避免请求等待连接
• 索引优化：为score字段创建索引，提升查询过滤效率：

  collection.createIndex(Indexes.ascending(MongoXXXScore::score.name))
  

• 线程调度监控：使用协程监控工具（如Kotlinx Coroutines Debug Agent）追踪线程调度情况，确认是否存在线程阻塞或资源竞争问题

内容的提问来源于stack exchange，提问作者bemol