记一次 .NET 某WMS仓储打单系统 内存暴涨分析

一：背景

1. 讲故事

七月中旬有一位朋友加wx求助，他的程序在生产上跑着跑着内存就飙起来了，貌似没有回头的趋势，询问如何解决，截图如下：

和这位朋友聊下来，感觉像是自己在小县城当了个小老板，规律的生活，有当地资源，各种小关系，有一股财务自由的味道，这也是我一直向往的生活方式 😄😄😄。

既然朋友找到我了，我得想办法给他解决问题，既然是内存暴涨，我就赌一把在托管层面吧，嘿嘿，上windbg说话。

二： windbg 分析

1. 托管还是非托管

一直在追这个系列的朋友应该知道，我无数次的用 !address -summary 和 !eeheap -gc 这两个命令来判断当前内存属于托管层还是非托管层。

0:000> !address -summary

--- State Summary ---------------- RgnCount ----------- Total Size -------- %ofBusy %ofTotal
MEM_FREE                                393     7dfe`f2105000 ( 125.996 TB)           98.43%
MEM_RESERVE                            1691      200`0f1e4000 (   2.000 TB)  99.81%    1.56%
MEM_COMMIT                             6191        0`fed07000 (   3.981 GB)   0.19%    0.00%


0:000> !eeheap -gc
Number of GC Heaps: 1
generation 0 starts at 0x000001D2E572BBC8
generation 1 starts at 0x000001D2E54F70E0
generation 2 starts at 0x000001D252051000
ephemeral segment allocation context: none
         segment             begin         allocated              size
000001D252050000  000001D252051000  000001D26204FFE0  0xfffefe0(268431328)
Large object heap starts at 0x000001D262051000
         segment             begin         allocated              size
000001D262050000  000001D262051000  000001D2655F3F80  0x35a2f80(56242048)
Total Size:              Size: 0xbf4dbf80 (3209543552) bytes.
------------------------------
GC Heap Size:    Size: 0xbf4dbf80 (3209543552) bytes.

卦象上进程指标为 3.98G ，GC堆指标为 3209543552 = 3G ，很显然，本次事故属于 托管层面。

2. 寻找托管层上的大对象

我们都知道C#是托管语言，所以甭管有用没用的对象都逃不出GC堆，言外之意就是看GC堆准没错，挑几个大对象看看。

0:000> !dumpheap -stat
Statistics:
              MT    Count    TotalSize Class Name
00007ff98a68f090   391475     43869284 System.Int32[]
00007ff98b6adfa0  1902760     45666240 System.Collections.ObjectModel.ReadOnlyCollection`1[[System.Linq.Expressions.Expression, System.Linq.Expressions]]
00007ff98b6ac3c0  1951470     46835280 System.Linq.Expressions.ConstantExpression
00007ff98bc452e0  1681178     53797696 System.Linq.Expressions.TypedConstantExpression
00007ff98eacb6b8  1902708     60886656 System.Dynamic.Utils.ListArgumentProvider
00007ff98f236518  1774982     70999280 Microsoft.EntityFrameworkCore.Query.Expressions.ColumnExpression
00007ff98c650c58  1681142     80694816 System.Linq.Expressions.MethodCallExpression3
00007ff98a82bc38  3414094     81938256 System.RuntimeMethodHandle
00007ff98fd96fc0    17750     83936016 System.Collections.Generic.Dictionary`2+Entry[[System.Reflection.MemberInfo, System.Private.CoreLib],[System.Linq.Expressions.Expression, System.Linq.Expressions]][]
00007ff98e5ed5d8    35493    101740504 System.Collections.Generic.Dictionary`2+Entry[[Microsoft.Extensions.DependencyInjection.ServiceLookup.ServiceCacheKey, Microsoft.Extensions.DependencyInjection],[System.Object, System.Private.CoreLib]][]
00007ff98bcff6a8  3639389    116460448 System.Linq.Expressions.PropertyExpression
00007ff98b85cf00  5028347    160907104 System.Reflection.Emit.GenericFieldInfo
00007ff98a671e18  2178117    168395994 System.String
00007ff98a5b6610   160565    171498416 System.Object[]
00007ff98eaa8ab0  4981589    199263560 System.Linq.Expressions.MemberAssignment
00007ff98a672360   398740    391928469 System.Byte[]
00007ff98a746d68   181886    486150592 System.Char[] 

从托管堆上看，System.Linq.Expressions.MemberAssignment 对象高达 498w ，很明显有问题，从类名看可能和 ExpressionTree 有关，那就抽几个对象看看它的引用链上是否有过大的对象。

0:000> !gcroot 000001d25399f690
HandleTable:
    000001D251B715A8 (pinned handle)
    -> 000001D262068CF0 System.Object[]
    -> 000001D2531C3B78 Microsoft.EntityFrameworkCore.Internal.ServiceProviderCache
    -> 000001D25399E3D0 Remotion.Linq.QueryModel
    -> 000001D25399E3B8 Remotion.Linq.Clauses.SelectClause
    -> 000001D25442C068 System.Linq.Expressions.MemberInitExpression
    -> 000001D25442C050 System.Runtime.CompilerServices.TrueReadOnlyCollection`1[[System.Linq.Expressions.MemberBinding, System.Linq.Expressions]]
    -> 000001D2539A0290 System.Linq.Expressions.MemberBinding[]
    -> 000001D25399F690 System.Linq.Expressions.MemberAssignment

引用链特别长，这里我就截取一下，经过一顿排查，我发现大对象居然是 Remotion.Linq.Clauses.SelectClause，objsize 这个对象直接爆掉了，真的很奇葩，如下代码所示：

0:000> !objsize 000001D25399E3B8
sizeof(000001D25399E3B8) = -1187378032 (0xb93a0c90) bytes (Remotion.Linq.Clauses.SelectClause)

有点懵，这个对象居然是罪魁祸首，从引用链看它是 EF 下的一个构建表达式树的小部件，可以肯定的是，朋友在用EF的时候出了什么问题，不过还得硬着头皮继续挖 SelectClause，经过深挖，我发现这个类有大量这样的 char[] 数组，导出来后大概是下面这样。

Logistics.Text30), 
|           Text31 = string TryReadValue(t1.Outer.Outer, 42, WmsOutboundConfirmLogistics.Text31), 
|           Text32 = string TryReadValue(t1.Outer.Outer, 43, WmsOutboundConfirmLogistics.Text32), 
|           Text33 = string TryReadValue(t1.Outer.Outer, 44, WmsOutboundConfirmLogistics.Text33), 
|           Text34 = string TryReadValue(t1.Outer.Outer, 45, WmsOutboundConfirmLogistics.Text34), 
|           Text35 = string TryReadValue(t1.Outer.Outer, 46, WmsOutboundConfirmLogistics.Text35), 
|           IsQueue = Nullable<bool> TryReadValue(t1.Outer.Outer, 47, WmsOutboundConfirmLogistics.IsQueue), 
|           IsStop = Nullable<bool> TryReadValue(t1.Outer.Outer, 48, WmsOutboundConfirmLogistics.IsStop), 
|           CheckCode = string TryReadValue(t1.Outer.Outer, 49, WmsOutboundConfirmLogistics.CheckCode), 
|           ClientCode = string TryReadValue(t1.Outer.Inner, 50, WmsOutboundOrder.ClientCode), 
|           WarehouseCode = string TryReadValue(t1.Outer.Inner, 51, WmsOutboundOrder.WarehouseCode), 
|           ErpNumber = string TryReadValue(t1.Outer.Inner, 52, WmsOutboundOrder.ErpNumber), 
|           OrderCategory = string TryReadValue(t1.Outer.Inner, 53, WmsOutboundOrder.OrderCategory), 
|           OrderStatus = string TryReadValue(t1.Outer.Inner, 54, WmsOutboundOrder.OrderStatus), 
|           OrderType = string TryReadValue(t1.Outer.Inner, 55, WmsOutboundOrder.OrderType), 
|           SendCompany = string TryReadValue(t1.Outer.Inner, 56, WmsOutboundOrder.SendCompany), 
|           SendName = string TryReadValue(t1.Outer.Inner, 57, WmsOutboundOrder.SendName), 
|           SendTel = string TryReadValue(t1.Outer.Inner, 58, WmsOutboundOrder.SendTel), 
|           SendMobile = string TryReadValue(t1.Outer.Inner, 59, WmsOutboundOrder.SendMobile), 
|           SendProvince = string TryReadValue(t1.Outer.Inner, 60, WmsOutboundOrder.SendProvince), 
|           SendCity = string TryReadValue(t1.Outer.Inner, 61, WmsOutboundOrder.SendCity), 
|           SendArea = string TryReadValue(t1.Outer.Inner, 62, WmsOutboundOrder.SendArea), 
|           ...
|           CategoryName = string TryReadValue(t1.Outer.Inner, 88, WmsOutboundOrder.CategoryName), 
|           SourcePlatformCode = string TryReadValue(t1.Outer.Inner, 89, WmsOutboundOrder.SourcePlatformCode), 
|           PayMode = (string)string TryReadValue(t1.Outer.Outer, 90, null), 
|           List = List<WmsOutboundConfirmLogisticsLinesDTO> WmsOutboundConfirmLogisticsBusiness.GetOrderLines(string TryReadValue(t1.Outer.Outer, 5, WmsOutboundConfirmLogistics.OrderNumber)), 
|           ConfirmTime = DateTime TryReadValue(t1.Inner, 91, WmsOutboundOrderConfirmation.CreateTime), 
|           ReturnUrl = (string)string TryReadValue(t1.Outer.Outer, 92, null) 
|       }
|__ ), 
|__ contextType: Core.DataRepository.BaseDbContext, 
|__ logger: DiagnosticsLogger<Query>, 
|__ queryContext: Unhandled parameter: queryContext)                                                      

从内容看，应该是 select 语句的ExpressionTree表示，问了下朋友，说大概是报表业务，不过这些信息给他貌似也没有多大帮助，说实话到这里我其实也不知道怎么继续往下排查了，陷入了绝望。

3. 从绝望中寻找希望

我在想，既然EF构建了大量这样的 ExpressionTree，肯定有问题，但也想不出是什么问题，隔了半天，我突然灵光一现，EF既然构建了树，有可能sql也出来了，对，我何不直接在 heap 上搜索 select 的sql语句。。。。

0:000> !strings /m:*select*
Address            Gen    Length   Value
000001d2e4de64e0    2       1964   SELECT a."Id", a."CreateTime" AS "CreateTime0", a."CreatorId", a."CreatorRealName", a."Deleted", a."OrderNumber", a."CarrierId",...
000001d2e4e11e78    2       1964   SELECT a."Id", a."CreateTime" AS "CreateTime0", a."CreatorId", a."CreatorRealName", a."Deleted", a."OrderNumber", a."CarrierId",...
000001d2e4e3d1f0    2       1964   SELECT a."Id", a."CreateTime" AS "CreateTime0", a."CreatorId", a."CreatorRealName", a."Deleted", a."OrderNumber", a."CarrierId",...
000001d2e4e673c8    2       1964   SELECT a."Id", a."CreateTime" AS "CreateTime0", a."CreatorId", a."CreatorRealName", a."Deleted", a."OrderNumber", a."CarrierId",...
000001d2e4e91760    2       1964   SELECT a."Id", a."CreateTime" AS "CreateTime0", a."CreatorId", a."CreatorRealName", a."Deleted", a."OrderNumber", a."CarrierId",...
000001d2e4ebb2e8    2       1964   SELECT a."Id", a."CreateTime" AS "CreateTime0", a."CreatorId", a."CreatorRealName", a."Deleted", a."OrderNumber", a."CarrierId",...
000001d2e4ee54f8    2       1964   SELECT a."Id", a."CreateTime" AS "CreateTime0", a."CreatorId", a."CreatorRealName", a."Deleted", a."OrderNumber", a."CarrierId",...
000001d2e4f10758    2       1964   SELECT a."Id", a."CreateTime" AS "CreateTime0", a."CreatorId", a."CreatorRealName", a."Deleted", a."OrderNumber", a."CarrierId",...
000001d2e4f398d0    2       1964   SELECT a."Id", a."CreateTime" AS "CreateTime0", a."CreatorId", a."CreatorRealName", a."Deleted", a."OrderNumber", a."CarrierId",...

---------------------------------------
18128 matching strings

果然发现了大量重复的 select 语句，而且从最左边的内存地址看都是非常接近的，也就说明他们是在某一个操作中同时生成的，然后我们导出几个sql语句。

SELECT a."Id", ....
FROM "WmsOutboundConfirmLogistics" AS a
INNER JOIN "WmsOutboundOrder" AS b ON a."OrderNumber" = b."OrderNumber"
INNER JOIN "WmsOutboundOrderConfirmation" AS c ON a."OrderNumber" = c."OrderNumber"
WHERE (a."OrderNumber" = @__pagination_OrderNumber_0) AND (b."FreezeStatus" = FALSE)
ORDER BY a."Id"";

SELECT a."Id", ....
FROM "WmsOutboundConfirmLogistics" AS a
INNER JOIN "WmsOutboundOrder" AS b ON a."OrderNumber" = b."OrderNumber"
INNER JOIN "WmsOutboundOrderConfirmation" AS c ON a."OrderNumber" = c."OrderNumber"
WHERE (a."OrderNumber" = @__pagination_OrderNumber_0) AND (b."FreezeStatus" = FALSE)
ORDER BY a."Id""

拿到这 1.8w 重复的sql 给朋友看，朋友说这是查询报表的sql。

4. 所有线索整合打通

那这里就存在着很大问题，既然是查询报表，为什么会有 1.8w 相同的sql，唯一不同的就是 a."OrderNumber" = @__pagination_OrderNumber_0 中的订单号，难道不应该是 a.OrderNumber in (xxxx) 或者是表关联查询吗？？？ 整理一下就是下面这样的猜想：

-- 理想
select * from a where a.id in (1,2,3)

-- 现实
select * from a where a.id=1;
select * from a where a.id=2;
select * from a where a.id=3;

加上每个sql内存地址相近，再结合爆表的 Remotion.Linq.Clauses.SelectClause 对象，整个流程大概就是：本该表关联或者in操作，结果变成了无数个单条sql语句查询，导致EF底层出现内存爆炸式增长。

三：总结

看了下朋友查询ef的写法，猜测大多都是人肉构建 ExpressionTree 去查询数据库，大写的🐂👃，比如下面的这张图：

解决方案就是让朋友检查下表示式树的写法问题，或者直接灌写好的sql得了，说实话这个dump还是费了九牛二虎之力，本以为很简单，实操起来还是碰到了一点小困难，就当历练成长吧！

更多高质量干货：参见我的 GitHub: dotnetfly