Service Fabric可靠状态与F#默认GetHashCode异常问题求助

I recently updated a child domain object (DataChildObject) in my Service Fabric (SF) stateful service by adding a new tuple field (NewValue composed of two doubles). When attempting to upgrade the service, the upgrade gets stuck on the first upgrade domain with a 'package activation' error. Checking the Event Viewer on the failing node reveals this exception:

Errormsg=TStore.OnApplyAddAsync: Unexpected exception System.NullReferenceException: Object reference not set to an instance of an object.
at xxx.DataChildObject.GetHashCode(IEqualityComparer comp)
at xxx.Data.GetHashCode(IEqualityComparer comp)
at xxx.Data.GetHashCode() in C:\Users\xxx\Source\Repos\xxx\xxx\xxx.Core\Domain.fs:line 17
at System.Fabric.Store.TStore`5.OnApplyAdd(TransactionBase txn, MetadataOperationData metadataOperationData, RedoUndoOperationData operationRedoUndo, Boolean isIdempotent, String applyType)
Assert=System.Exception: at System.Environment.GetStackTrace(Exception e, Boolean needFileInfo)

I understand F# automatically generates equality methods that are triggered during SF state handling, and due to DataContract serialization, fields might be null during GetHashCode checks—but I'm unclear on when and why this check is firing. I tried overriding GetHashCode() for DataChildObject, but it didn't resolve the upgrade error. Here's my current code:

[<DataContract>]
type DataChildObject = class
    [<DataMember(IsRequired=false,Name="Value1")>]
    val mutable Value1 : float * float
    [<DataMember(IsRequired=false,Name="Value2")>]
    val mutable Value2 : float * float
    [<DataMember(IsRequired=false,Name="NewValue")>]
    val mutable NewValue : float * float
    new (v1,v2) = {Value1=v1;Value2=v2;NewValue=1.0,1.0}
    override this.GetHashCode() =
        let value1 = if (box this.Value1 = null) then 1 else this.Value1.GetHashCode()
        let value2 = if (box this.Value2 = null) then 1 else this.Value2.GetHashCode()
        let newValue = if (box this.NewValue = null) then 1 else (this.NewValue).GetHashCode()
        value1+value2+newValue
end

Since these domain objects will likely undergo more changes in the future, I need help understanding and resolving this issue.

Root Cause #

The key issue here is that Service Fabric's TStore is calling the overload GetHashCode(IEqualityComparer comp), not the parameterless GetHashCode() you overridden. F# automatically generates this overload for classes when equality is inferred, and the generated implementation assumes the provided IEqualityComparer is non-null. When TStore passes a null comparer (which can happen during state replication/activation), calling comp.GetHashCode() on your new tuple field triggers a NullReferenceException.

Additionally, your parameterless GetHashCode() includes unnecessary null checks for tuple fields: tuples are value types, so they can never be null—boxing them will always produce a non-null object. These checks don't fix the actual problem and are redundant.

Solution #

To resolve this, you need to override the GetHashCode(IEqualityComparer comp) overload to handle null comparers explicitly, and ensure both overloads compute consistent hash codes. Here's how to adjust your DataChildObject class:

[<DataContract>]
type DataChildObject = class
    [<DataMember(IsRequired=false,Name="Value1")>]
    val mutable Value1 : float * float
    [<DataMember(IsRequired=false,Name="Value2")>]
    val mutable Value2 : float * float
    [<DataMember(IsRequired=false,Name="NewValue")>]
    val mutable NewValue : float * float

    new (v1,v2) = {Value1=v1;Value2=v2;NewValue=(1.0,1.0)}
    
    // Override parameterless GetHashCode() to delegate to the comparer overload
    override this.GetHashCode() =
        this.GetHashCode(EqualityComparer<float*float>.Default)

    // Override the IEqualityComparer overload TStore uses
    override this.GetHashCode(comp: IEqualityComparer) =
        // Fallback to default comparer if the provided one is null
        let safeComp = if comp = null then EqualityComparer<float*float>.Default else comp
        let hash1 = safeComp.GetHashCode(this.Value1)
        let hash2 = safeComp.GetHashCode(this.Value2)
        let hash3 = safeComp.GetHashCode(this.NewValue)
        // Use XOR to combine hashes (reduces collision risk better than addition)
        hash1 ^^^ hash2 ^^^ hash3
end

Key Improvements: #

• Handles Null Comparers: The overload checks if comp is null and uses a default comparer, eliminating the NullReferenceException.
• Consistent Hash Calculation: The parameterless method delegates to the overload, ensuring hash code consistency across SF state operations.
• Better Hash Combination: XOR (^^^) reduces hash collision chances, which is critical for reliable state lookups in TStore.
• Removed Redundant Checks: Tuple fields are value types, so null checks were unnecessary and have been removed.

Additional Recommendations for Future Changes: #

• Override All Equality Methods: When customizing equality logic for SF-facing F# classes, override Equals(obj), Equals(T other), and both GetHashCode() overloads to ensure full compatibility with SF's state store.
• Test Serialization/Deserialization: When adding new fields, verify how old serialized data deserializes into the updated class. Use IsRequired=false for optional fields and ensure default values are properly initialized.
• Prefer Immutable Records: Mutable classes can cause unexpected behavior in SF state replication. F# immutable records automatically generate safe equality/hash code logic that works seamlessly with SF.

内容的提问来源于stack exchange，提问作者Alex Zevenbergen