AWS DMS 交换文件占用所有空间

Posted 2023-02-19

【中文标题】AWS DMS 交换文件占用所有空间

【英文标题】：AWS DMS swap files consumes all the space

【发布时间】：2021-05-08 10:33:09

【问题描述】：

我正在迁移许多数据库，但我发现我的大小超过 50GB 的数据库在一段时间后由于存储空间不足而在 CDC 中失败。

我正在使用一个复制实例类dms.r5.large 并且一切运行顺利，直到完全加载完成。 当 CDC 启动时，我收到如下日志消息：

D:  There are 188 swap files of total size 93156 Mb. Left to process 188 of size 93156 Mb

但是交换文件永远不会被删除，实例会不断累积交换文件，最终实例会耗尽存储空间。 需要注意的是，我在监控指标中的交换使用率接近于零。

我已经尝试过dms.r5.xlarge，问题是一样的，这让我觉得内存不是问题。

您知道导致这种行为的原因是什么吗？ 有没有办法调试这个？

谢谢！

更多有用的数据： 复制实例类：dms.r5.large，我已尝试使用 dms.r5.xlarge。40GB 的存储，我已尝试使用 300GB，但最终 CDC 阶段消耗了所有存储。 要迁移的数据库大约是80GB。 任务设置：

    "TargetMetadata": 
        "TargetSchema": "",
        "SupportLobs": true,
        "FullLobMode": false,
        "LobChunkSize": 0,
        "LimitedSizeLobMode": true,
        "LobMaxSize": 32,
        "InlineLobMaxSize": 0,
        "LoadMaxFileSize": 0,
        "ParallelLoadThreads": 0,
        "ParallelLoadBufferSize": 0,
        "BatchApplyEnabled": false,
        "TaskRecoveryTableEnabled": false,
        "ParallelLoadQueuesPerThread": 0,
        "ParallelApplyThreads": 0,
        "ParallelApplyBufferSize": 0,
        "ParallelApplyQueuesPerThread": 0
    ,
    "FullLoadSettings": 
        "TargetTablePrepMode": "DROP_AND_CREATE",
        "CreatePkAfterFullLoad": false,
        "StopTaskCachedChangesApplied": false,
        "StopTaskCachedChangesNotApplied": false,
        "MaxFullLoadSubTasks": 8,
        "TransactionConsistencyTimeout": 600,
        "CommitRate": 10000
    ,
    "Logging": 
        "EnableLogging": true,
        "LogComponents": [
            "Id": "SOURCE_UNLOAD",
            "Severity": "LOGGER_SEVERITY_DEFAULT"
        ,
            "Id": "SOURCE_CAPTURE",
            "Severity": "LOGGER_SEVERITY_DEFAULT"
        ,
            "Id": "TARGET_LOAD",
            "Severity": "LOGGER_SEVERITY_DEFAULT"
        ,
            "Id": "TARGET_APPLY",
            "Severity": "LOGGER_SEVERITY_INFO"
        ,
            "Id": "TASK_MANAGER",
            "Severity": "LOGGER_SEVERITY_DEBUG"
        ]
    ,
    "ControlTablesSettings": 
        "historyTimeslotInMinutes": 5,
        "ControlSchema": "",
        "HistoryTimeslotInMinutes": 5,
        "HistoryTableEnabled": false,
        "SuspendedTablesTableEnabled": false,
        "StatusTableEnabled": false
    ,
    "StreamBufferSettings": 
        "StreamBufferCount": 3,
        "StreamBufferSizeInMB": 8,
        "CtrlStreamBufferSizeInMB": 5
    ,
    "ChangeProcessingDdlHandlingPolicy": 
        "HandleSourceTableDropped": true,
        "HandleSourceTableTruncated": true,
        "HandleSourceTableAltered": true
    ,
    "ErrorBehavior": 
        "DataErrorPolicy": "LOG_ERROR",
        "DataTruncationErrorPolicy": "LOG_ERROR",
        "DataErrorEscalationPolicy": "SUSPEND_TABLE",
        "DataErrorEscalationCount": 0,
        "TableErrorPolicy": "SUSPEND_TABLE",
        "TableErrorEscalationPolicy": "STOP_TASK",
        "TableErrorEscalationCount": 0,
        "RecoverableErrorCount": -1,
        "RecoverableErrorInterval": 5,
        "RecoverableErrorThrottling": true,
        "RecoverableErrorThrottlingMax": 1800,
        "RecoverableErrorStopRetryAfterThrottlingMax": false,
        "ApplyErrorDeletePolicy": "IGNORE_RECORD",
        "ApplyErrorInsertPolicy": "LOG_ERROR",
        "ApplyErrorUpdatePolicy": "LOG_ERROR",
        "ApplyErrorEscalationPolicy": "LOG_ERROR",
        "ApplyErrorEscalationCount": 0,
        "ApplyErrorFailOnTruncationDdl": false,
        "FullLoadIgnoreConflicts": true,
        "FailOnTransactionConsistencyBreached": false,
        "FailOnNoTablesCaptured": false
    ,
    "ChangeProcessingTuning": 
        "BatchApplyPreserveTransaction": true,
        "BatchApplyTimeoutMin": 1,
        "BatchApplyTimeoutMax": 30,
        "BatchApplyMemoryLimit": 500,
        "BatchSplitSize": 0,
        "MinTransactionSize": 1000,
        "CommitTimeout": 1,
        "MemoryLimitTotal": 1024,
        "MemoryKeepTime": 60,
        "StatementCacheSize": 50
    ,
    "ValidationSettings": 
        "EnableValidation": true,
        "ValidationMode": "ROW_LEVEL",
        "ThreadCount": 5,
        "PartitionSize": 10000,
        "FailureMaxCount": 10000,
        "RecordFailureDelayInMinutes": 5,
        "RecordSuspendDelayInMinutes": 30,
        "MaxKeyColumnSize": 8096,
        "TableFailureMaxCount": 1000,
        "ValidationOnly": false,
        "HandleCollationDiff": false,
        "RecordFailureDelayLimitInMinutes": 0,
        "SkipLobColumns": false,
        "ValidationPartialLobSize": 0,
        "ValidationQueryCdcDelaySeconds": 0
    ,
    "PostProcessingRules": null,
    "CharacterSetSettings": null,
    "LoopbackPreventionSettings": null,
    "BeforeImageSettings": null

【参考方案1】：

问题是由于高目标延迟，根本原因是数据库表的结构。

具有大量记录的表缺少primary keys 或unique identifiers 导致全表扫描，更改未应用，然后保存在复制实例存储中。

最终实例将耗尽存储空间。

要解决此问题，您应该运行迁移前评估以检查您的数据库是否申请 DMS 迁移。

解决此问题的另一种方法是在迁移中添加一个额外的列以创建唯一键并在迁移后将其删除。