Azure Portal创建Cosmos DB文档失败：ID已存在报错求助

问题描述 #

在Cosmos DB容器中创建新文档时，间歇性触发错误：Entity with the specified id already exists in the system.，确认目标ID不存在于容器中，但首次创建仍失败；问题自行恢复后会复发，严重影响业务。已尝试更换唯一ID、刷新页面、新建标签页、清除浏览器缓存等操作，均未解决。

完整报错信息：

Create document failed
Entity with the specified id already exists in the system., RequestStartTime: 2023-02-15T17:15:23.7571693Z, RequestEndTime: 2023-02-15T17:15:23.7571693Z, Number of regions attempted:1 {"systemHistory":[{"dateUtc":"2023-02-15T17:14:20.3055487Z","cpu":0.537,"memory":494453100.000,"threadInfo":{"isThreadStarving":"False","threadWaitIntervalInMs":0.0377,"availableThreads":32763,"minThreads":52,"maxThreads":32767},"numberOfOpenTcpConnection":506},{"dateUtc":"2023-02-15T17:14:30.3157513Z","cpu":1.468,"memory":495121388.000,"threadInfo":{"isThreadStarving":"False","threadWaitIntervalInMs":0.0246,"availableThreads":32762,"minThreads":52,"maxThreads":32767},"numberOfOpenTcpConnection":506},{"dateUtc":"2023-02-15T17:14:40.3260707Z","cpu":0.472,"memory":495105576.000,"threadInfo":{"isThreadStarving":"False","threadWaitIntervalInMs":0.0132,"availableThreads":32764,"minThreads":52,"maxThreads":32767},"numberOfOpenTcpConnection":505},{"dateUtc":"2023-02-15T17:14:50.3363254Z","cpu":0.555,"memory":495019484.000,"threadInfo":{"isThreadStarving":"False","threadWaitIntervalInMs":0.0282,"availableThreads":32764,"minThreads":52,"maxThreads":32767},"numberOfOpenTcpConnection":505},{"dateUtc":"2023-02-15T17:15:00.3467551Z","cpu":1.111,"memory":495027292.000,"threadInfo":{"isThreadStarving":"False","threadWaitIntervalInMs":0.0255,"availableThreads":32763,"minThreads":52,"maxThreads":32767},"numberOfOpenTcpConnection":505},{"dateUtc":"2023-02-15T17:15:10.3568375Z","cpu":0.559,"memory":495008380.000,"threadInfo":{"isThreadStarving":"False","threadWaitIntervalInMs":0.0155,"availableThreads":32764,"minThreads":52,"maxThreads":32767},"numberOfOpenTcpConnection":503}]} RequestStart: 2023-02-15T17:15:23.7571693Z; ResponseTime: 2023-02-15T17:15:23.7571693Z; StoreResult: StorePhysicalAddress: rntbd://10.0.0.34:14300/apps/c8ad9db4-cea9-48d8-9f4c-a680fc37edf0/services/277b4e29-80c4-4118-8a5e-a57346d6ef1e/partitions/45292f34-171b-40b8-8a5d-40996adabfdf/replicas/133201146384367926p, LSN: 2061, GlobalCommittedLsn: 2061, PartitionKeyRangeId: 0, IsValid: True, StatusCode: 409, SubStatusCode: 0, RequestCharge: 1.24, ItemLSN: -1, SessionToken: -1#2061, UsingLocalLSN: False, TransportException: null, BELatencyMs: 1.63, ActivityId: 3f54660b-6e43-4b71-9a36-125fdb5565d8, RetryAfterInMs: , TransportRequestTimeline: {"requestTimeline":[{"event": "Created", "startTimeUtc": "2023-02-15T17:15:23.7571693Z", "durationInMs": 0.0115},{"event": "ChannelAcquisitionStarted", "startTimeUtc": "2023-02-15T17:15:23.7571808Z", "durationInMs": 0.0141},{"event": "Pipelined", "startTimeUtc": "2023-02-15T17:15:23.7571949Z", "durationInMs": 0.2096},{"event": "Transit Time", "startTimeUtc": "2023-02-15T17:15:23.7574045Z", "durationInMs": 2.1625},{"event": "Received", "startTimeUtc": "2023-02-15T17:15:23.7595670Z", "durationInMs": 0.1415},{"event": "Completed", "startTimeUtc": "2023-02-15T17:15:23.7597085Z", "durationInMs": 0}],"serviceEndpointStats":{"inflightRequests":1,"openConnections":1},"connectionStats":{"waitforConnectionInit":"False","callsPendingReceive":0,"lastSendAttempt":"2023-02-15T17:14:29.9557540Z","lastSend":"2023-02-15T17:14:29.9557540Z","lastReceive":"2023-02-15T17:14:29.9557540Z"},"requestSizeInBytes":1250,"requestBodySizeInBytes":718,"responseMetadataSizeInBytes":179,"responseBodySizeInBytes":49}; ResourceType: Document, OperationType: Create , Microsoft.Azure.Documents.Common/2.14.0

可能原因 #

• 软删除文档残留：Cosmos DB软删除的文档在保留期内，常规查询无法看到，但系统仍会保留ID唯一性约束，导致创建冲突。
• 一致性级别导致的读取延迟：若使用会话/最终一致性，客户端查询可能无法实时获取最新的文档状态（比如刚删除的文档还未同步到查询节点），误以为ID不存在。
• 分区键与ID的组合冲突：Cosmos DB的文档唯一性是ID + 分区键的组合，若同一分区键下存在同名ID（即使其他分区键下没有），也会触发报错。
• 客户端重试逻辑问题：客户端请求超时后重试，但实际服务端已成功处理第一次请求，导致重复创建触发冲突。
• 副本同步延迟：多副本部署场景下，部分副本未同步最新的删除/创建操作，处理请求时判定ID已存在。

解决方案 #

• 排查软删除记录：执行带IsDeleted=true的查询语句，确认是否存在同名ID的软删除文档：

  SELECT * FROM c WHERE c.id = '你的目标ID' AND c._isDeleted = true
  

  若存在，可手动清理软删除文档（需开启软删除清理权限）或等待软删除保留期结束。
• 验证唯一性约束范围：确认容器的分区键配置，检查目标ID是否在同一分区键下已存在（即使其他分区键下无该ID）。
• 调整一致性级别临时排查：临时将容器一致性级别切换为强一致性，验证问题是否消失；若问题解决，说明是一致性延迟导致，可调整客户端查询时的会话令牌或使用分区键精准查询。
• 检查客户端逻辑：排查客户端是否存在重复提交、无幂等性的重试逻辑，添加幂等性控制（比如使用唯一请求ID），避免重复创建请求。
• 查看操作审计日志：通过Cosmos DB诊断日志查看该ID的历史操作记录，确认是否有未察觉的创建/删除操作。
• 强制刷新客户端会话：客户端调用时主动刷新会话令牌，避免使用过期的会话缓存导致的读取不一致。

内容的提问来源于stack exchange，提问作者imsan