Azure CosmosDB执行TransactionalBatch返回Forbidden错误，SubStatusCode为ComputeInternalError：错误解析与修复方法

你遇到的ComputeInternalError是Cosmos DB服务器端在处理TransactionalBatch操作时触发的内部计算异常，这个子状态码比较少见，没有公开的通用文档说明，通常和批量请求的内容、格式或者服务器端临时状态相关。

先从你提供的诊断信息里抓几个关键线索：

• 请求最终返回了MultiStatus主状态码，搭配ComputeInternalError子状态码，说明服务器接收了请求但处理过程中出了内部计算问题
• 请求消耗了1.67 RU，网络传输和客户端状态都正常（没有TransportException、客户端CPU也没过载），排除了网络和本地资源问题

下面是具体的排查和修复步骤：

• 检查批量操作的内容合法性
  • 确认批次里的每个文档都符合Cosmos DB的约束：分区键是否正确设置、单个文档大小不超过2MB上限、JSON格式完全合法（没有语法错误、特殊字符）
  • 如果批次包含更新/删除操作，要确保目标文档存在，或者操作的条件（比如ETag匹配）没有冲突
• 拆分大批次为小批次
  • 过大的批量（比如单次包含数百个操作）可能会触发服务器端的计算负载异常，尝试把批次拆分成更小的单元（比如每次50-100个操作），再执行看看是否恢复正常
• 添加重试逻辑
  • 这个错误大概率是服务器端的临时异常，按照Cosmos DB的最佳实践，给批量操作加上指数退避重试逻辑。你可以依赖SDK自带的RetryHandler，也可以自己实现重试逻辑，等待几秒后再尝试执行
• 升级Cosmos DB SDK版本
  • 你当前使用的是cosmos-netstandard-sdk/3.20.1，这个版本是2021年的旧版本，建议升级到最新的稳定版，新版本会修复很多已知的服务器端交互问题
• 联系Azure官方支持
  • 如果以上步骤都无效，那可能是服务器端的特定集群问题。你可以通过Azure门户提交支持工单，附上完整的Diagnostics信息（尤其是里面的ActivityId），让Azure工程师帮你定位具体的服务器端问题

附上你提供的脱敏诊断信息：

{"name": "ExecuteAsync", "id": "<redacted>", "caller info": {"member": "OperationHelperWithRootTraceAsync", "file": "ClientContextCore.cs", "line": 219}, "start time": "12:47:58:185", "duration in milliseconds": 292.1597, "data": {"Client Configuration": {"Client Created Time Utc": "2021-08-30T12:46:26.5148538Z", "NumberOfClientsCreated": 1, "User Agent": "cosmos-netstandard-sdk/3.20.1|3.19.3|1|X64|Microsoft Windows 10.0.19042|.NET 5.0.9|N|", "ConnectionConfig": {"gw": "(cps:50, urto:10, p:False, httpf: False)", "rntbd": "(cto: 5, icto: -1, mrpc: 30, mcpe: 65535, erd: False, pr: ReuseUnicastPort)", "other": "(ed:False, be:False)"}, "ConsistencyConfig": "(consistency: NotSet, prgns:[])"}}, "children": [{"name": "Execute Next Batch", "id": "<redacted>", "caller info": {"member": "ExecuteAsync", "file": "BatchExecutor.cs", "line": 44}, "start time": "12:47:58:186", "duration in milliseconds": 290.9772, "children": [{"name": "Create Batch Request", "id": "<redacted>", "caller info": {"member": "CreateAsync", "file": "SinglePartitionKeyServerBatchRequest.cs", "line": 52}, "start time": "12:47:58:188", "duration in milliseconds": 101.1877}, {"name": "Execute Batch Request", "id": "<redacted>", "caller info": {"member": "ExecuteServerRequestAsync", "file": "BatchExecutor.cs", "line": 81}, "start time": "12:47:58:290", "duration in milliseconds": 186.8077, "children": [{"name": "Microsoft.Azure.Cosmos.Handlers.RequestInvokerHandler", "id": "<redacted>", "start time": "12:47:58:290", "duration in milliseconds": 168.1416, "children": [{"name": "Microsoft.Azure.Cosmos.Handlers.DiagnosticsHandler", "id": "<redacted>", "start time": "12:47:58:290", "duration in milliseconds": 168.0172, "data": {"CPU Load History": {"CPU History": "(2021-08-30T12:46:26.9586842Z 25.000), (2021-08-30T12:46:56.9678395Z 33.333), (2021-08-30T12:47:16.9629830Z 0.000), (2021-08-30T12:47:26.9605546Z 9.245), (2021-08-30T12:47:36.9619653Z 0.000), (2021-08-30T12:47:58.1734808Z 9.199)"}}, "children": [{"name": "Microsoft.Azure.Cosmos.Handlers.RetryHandler", "id": "<redacted>", "start time": "12:47:58:290", "duration in milliseconds": 168.0093, "children": [{"name": "Microsoft.Azure.Cosmos.Handlers.RouterHandler", "id": "<redacted>", "start time": "12:47:58:290", "duration in milliseconds": 167.9875, "children": [{"name": "Microsoft.Azure.Cosmos.Handlers.TransportHandler", "id": "<redacted>", "start time": "12:47:58:290", "duration in milliseconds": 167.9816, "children": [{"name": "Microsoft.Azure.Documents.ServerStoreModel Transport Request", "id": "<redacted>", "caller info": {"member": "ProcessMessageAsync", "file": "TransportHandler.cs", "line": 109}, "start time": "12:47:58:290", "duration in milliseconds": 167.9315, "data": {"Client Side Request Stats": {"Id": "<redacted>", "ContactedReplicas": [{"Count": 1, "Uri": "rntbd://<redacted>"}, {"Count": 1, "Uri": "rntbd://<redacted>"}, {"Count": 1, "Uri": "rntbd://<redacted>"}], "RegionsContacted": ["https://<redacted>.documents.azure.com/"], "FailedReplicas": [], "AddressResolutionStatistics": [], "StoreResponseStatistics": [{"ResponseTimeUTC": "2021-08-30T12:47:58.4586683Z", "ResourceType": "Document", "OperationType": "Batch", "LocationEndpoint": "https://<redacted>.documents.azure.com/", "StoreResult": {"ActivityId": "<redacted>", "StatusCode": "MultiStatus", "SubStatusCode": "ComputeInternalError", "LSN": 308271, "PartitionKeyRangeId": "0", "GlobalCommittedLSN": 308271, "ItemLSN": -1, "UsingLocalLSN": false, "QuorumAckedLSN": 308271, "SessionToken": "-1#308271", "CurrentWriteQuorum": 3, "CurrentReplicaSetSize": 4, "NumberOfReadRegions": 0, "IsClientCpuOverloaded": false, "IsValid": true, "StorePhysicalAddress": "rntbd://<redacted>", "RequestCharge": 1.67, "BELatencyInMs": "0.726", "RntbdRequestStats": [{"EventName": "Created", "StartTime": "2021-08-30T12:47:58.2909371Z", "DurationInMicroSec": "9"}, {"EventName": "ChannelAcquisitionStarted", "StartTime": "2021-08-30T12:47:58.2909469Z", "DurationInMicroSec": "131386"}, {"EventName": "Pipelined", "StartTime": "2021-08-30T12:47:58.4223335Z", "DurationInMicroSec": "1098"}, {"EventName": "Transit Time", "StartTime": "2021-08-30T12:47:58.4234318Z", "DurationInMicroSec": "34423"}, {"EventName": "Received", "StartTime": "2021-08-30T12:47:58.4578548Z", "DurationInMicroSec": "778"}, {"EventName": "Completed", "StartTime": "2021-08-30T12:47:58.4586336Z", "DurationInMicroSec": "0"}], "TransportException": null}}]}}]}]}]}]}, {"name": "Create Trace", "id": "<redacted>", "caller info": {"member": "FromResponseMessageAsync", "file": "TransactionalBatchResponse.cs", "line": 214}, "start time": "12:47:58:461", "duration in milliseconds": 15.8089}]}]}

内容的提问来源于stack exchange，提问作者Michel