使用EF Core Cosmos时，重复关联EventLocation致首个Event字段为空

我有一个包含Events集合的Conference对象，每个Event关联一个EventLocation，多个Event可以共享同一个EventLocation（比如两场活动都在BallRoomA举办）。但添加和已有Event使用相同EventLocation的新Event时，首个Event的EventLocation字段会变成Null。以下是相关代码和存储结果，需要排查原因并解决。

使用的依赖包 #

<PackageReference Include="Microsoft.AspNetCore.Diagnostics.EntityFrameworkCore" Version="9.0.0-preview.6.24328.4" />
<PackageReference Include="Microsoft.EntityFrameworkCore.Cosmos" Version="9.0.0-preview.6.24327.4" />
<PackageReference Include="Microsoft.EntityFrameworkCore.Tools" Version="9.0.0-preview.6.24327.4">

实体类 #

public class Conference : CosmosBase
{
    public Hotel Hotel { get; set; }
    public ICollection<Event> Events { get; set; } = new List<Event>();
    public string ConferenceName { get; set; }
}
public class Event : CosmosBase
{
    public Conference Conference { get; set; }
    public EventLocation EventLocation { get; set; }
    public string EventName { get; set; }
}
public class EventLocation : CosmosBase
{
    public Hotel Hotel { get; set; }
    public string EventLocationName { get; set; }
}

初始化代码 #

List<Event> CreateEvents(Hotel h, Conference c, List<EventLocation> locs)
{
    var e0 = new Event();
    e0.PK = h.PK;
    e0.ID = e0.GetNewID();
    e0.Conference = c;
    e0.EventName = "Conference Check In";
    e0.EventLocation = locs[0];

    var e1 = new Event();
    e1.PK = h.PK;
    e1.ID = e1.GetNewID();
    e1.Conference = c;
    e1.EventName = "Keynote Session";
    e1.EventLocation = locs[0];

    var events = new List<Event>();
    events.Add(e0);
    events.Add(e1);
    return events;
}

Cosmos DB存储结果 #

{
    "ID": "2516807109275195800",
    "PK": "H-0d5fdf24-845b-4d0b-b92b-1e48979f69aa",
    "ConferenceDescription": "All hotels conference",
    "ConferenceEndDate": "2024-10-10T15:00:00-05:00",
    "ConferenceName": "The Big Conference",
    "ConferenceStartDate": "2024-10-07T09:30:00-05:00",
    "Discriminator": "Conference",
    "HotelID": "2516807109275205674",
    "HotelPK": "H-0d5fdf24-845b-4d0b-b92b-1e48979f69aa",
    "id": "Conference|2516807109275195800",
    "Events": [
        {
            "ID": "2516807109275185830",
            "EventDescription": "Check in to conference",
            "EventEndDate": "2024-10-07T15:30:00-05:00",
            "EventName": "Conference Check In",
            "EventStartDate": "2024-10-07T06:00:00-05:00",
            "PK": "H-0d5fdf24-845b-4d0b-b92b-1e48979f69aa",
            "EventLocation": null
        },
        {
            "ID": "2516807109275183758",
            "EventDescription": "Where the keynote speakers talk",
            "EventEndDate": "2024-10-07T10:30:00-05:00",
            "EventName": "Keynote Session",
            "EventStartDate": "2024-10-07T09:00:00-05:00",
            "PK": "H-0d5fdf24-845b-4d0b-b92b-1e48979f69aa",
            "EventLocation": {
                "EventLocationDescription": "The biggest ballroom we have",
                "EventLocationName": "Grand Ballroom",
                "HotelID": "2516807109275205674",
                "HotelPK": "H-0d5fdf24-845b-4d0b-b92b-1e48979f69aa",
                "ID": "2516807109275190703",
                "PK": "H-0d5fdf24-845b-4d0b-b92b-1e48979f69aa"
            }
        }
    ]
}

原因分析 #

两个Event关联的是同一个ID的EventLocation，但首个Event的该字段存为Null，核心原因是EF Core Cosmos在处理嵌入式重复对象时的序列化逻辑限制：

• 当同一个嵌套对象被多个父对象引用时，EF Core Cosmos默认只会在最后一次出现的位置序列化该对象，前面的引用会被置为Null，这是因为它默认不处理同一文档内的重复嵌套对象引用。
• 当前模型中EventLocation是作为嵌入式对象存储在Conference文档的Events数组里，而非独立文档，这种场景下重复引用同一对象会触发该问题。

解决办法 #

方案1：将EventLocation转为独立文档 #

把EventLocation配置为独立的Cosmos DB文档，通过外键关联而非嵌入式对象：

// 修改Event类，用外键关联EventLocation
public class Event : CosmosBase
{
    public Conference Conference { get; set; }
    public string EventLocationId { get; set; } // 外键字段
    [ForeignKey(nameof(EventLocationId))]
    public EventLocation EventLocation { get; set; }
    public string EventName { get; set; }
}

// 在DbContext中配置EventLocation为独立文档
protected override void OnModelCreating(ModelBuilder modelBuilder)
{
    modelBuilder.Entity<EventLocation>()
        .ToContainer("EventLocations")
        .HasPartitionKey(e => e.PK);
}

这种方式下，EventLocation是独立文档，多个Event通过ID引用，不会出现序列化时的Null问题。

方案2：为每个Event创建EventLocation副本 #

如果必须保持嵌入式结构，不要让多个Event引用同一个EventLocation实例，而是为每个Event创建独立的副本：

// 修改初始化代码，复制EventLocation对象
var e0 = new Event();
// ...其他属性赋值
e0.EventLocation = new EventLocation 
{
    PK = locs[0].PK,
    ID = locs[0].ID,
    Hotel = locs[0].Hotel,
    EventLocationName = locs[0].EventLocationName
};

var e1 = new Event();
// ...其他属性赋值
e1.EventLocation = new EventLocation 
{
    PK = locs[0].PK,
    ID = locs[0].ID,
    Hotel = locs[0].Hotel,
    EventLocationName = locs[0].EventLocationName
};

这样每个Event的EventLocation是独立实例，序列化时都会正常存储。

方案3：配置自定义序列化逻辑 #

配置EF Core使用支持保留对象引用的JSON序列化器，在DbContext的Cosmos选项中设置：

protected override void OnConfiguring(DbContextOptionsBuilder optionsBuilder)
{
    optionsBuilder.UseCosmos(
        "你的Cosmos连接字符串",
        "数据库名称",
        cosmosOptions =>
        {
            cosmosOptions.SerializerOptions = new CosmosSerializationOptions
            {
                IgnoreNullValues = false,
                ReferenceHandler = ReferenceHandler.Preserve
            };
        });
}

注意：该方案会增加文档存储大小，需评估业务场景是否适用。

内容的提问来源于stack exchange，提问作者Terrence