I’ve been using Workflow Foundation now for over a year and it has become an integral part of our architecture and has by and large been very successful. However it is not without it’s issues, the single most being that official documentation is light at best and relevant blog posts are quite rare.
I recently ran into a fairly serious problem after deploying a new release of our Software.
The Problem
I received this lovely email from our exception tracker:
System.Runtime.DurableInstancing.InstancePersistenceCommandException: The execution of the InstancePersistenceCommand named {urn:schemas-microsoft-com:System.Activities.Persistence/command}LoadWorkflow was interrupted by an error. ---> System.Runtime.Serialization.SerializationException: Deserialized object with reference id '73' not found in stream.
The key part is “Deserialized object with reference id '73' not found in stream.” As I’m sure you can tell this does not really provide any helpful information.
The stack trace does give you some more information as to which property failed, but depending on the object graph the stack trace can be fairly massive.
This is the part which gives you the biggest clue:
at ReadArrayOfPersonFromXml(XmlReaderDelegator , XmlObjectSerializerReadContext , XmlDictionaryString[] , XmlDictionaryString[] )
The Cause
Basically what has happened here is that when trying to load an unloaded workflow instance from the SqlWorkflowInstanceStore the deserializer spits the dummy because a property on the object that existed at the time the Workflow was persisted no longer exists on the class.
Once you understand the cause it does seem reasonable, sort of. In my effort to clean up the solution and get rid of unused properties I had effectively broken every workflow instance which was persisted before the date of the latest release.
As it turns out it’s possible to create this issue by either removing properties or renaming existing properties.
I made the fatal mistake of using passing my Domain Entities into the workflow as an argument.
The Patch
In order to fix this issue I ended up scrambling through the check-in history over the last 6 weeks trying to figure which properties had been removed. Once I figured it out, I added them back and deployed a patch. This solved the problem and the crisis was averted for the short-term.
The Solution
When a workflow is unloaded and persisted to the data store it is serialized and all internal variables & arguments are serialized along with it.
Rules for data objects in workflows:
- Never ever under any circumstances use Domain Entities
- Always use DTO’s for Arguments or Variables in Long Running Workflows
- Set IsRequired=false on the DataMember attribute for all Nullable Properties
- Set the Name property on the DataMember attribute
- Set the Order Property on the DataMember attribute
- Set the Name on the DataContract attribute
This is an example of a class that could create problems in the future.
[DataContract] public class Category { [DataMember] public int Id { get; set; } [DataMember] public List<Product> Products { get; set; } }
Based on the above lessons here is how I would suggest coding this class for use in a Workflow.
[DataContract(Name="Category")] public class Category { [DataMember(IsRequired = false, Name = "Id", Order=0)] public int Id { get; set; } [DataMember(IsRequired = false, Name = "Products")] public List<Product> Products { get; set; } }
By setting the Name property on the DataMember you are then free to change the property name without fear of breaking the deserialization.
Setting IsRequired to False allows you to remove the property in the future without breaking deserialization, obviously you should set this for True for any properties that are truly required by the Workflow to function correctly.
Conclusion
This is just an example of how to deal with data that changes in workflows, in my next post I’m going to cover how you deal with workflows whose logic changes after an instance has been created and persisted.
I hope this helps anyone who’s working with Workflow Foundation or about to start using it.
Till next time.