BlobHelper | consistent storage interface for Microsoft Azure | Storage library

You can tell the immediate cause from the stack trace.

A PDX serialized stream contains a type id which is a reference into a repository of type metadata maintained by a GemFire cluster. In this case, the serialized data of the object contained a typeId that is not in the cluster's metadata repository.

So the question becomes, "what serialized that object and why did it use an invalid type id ?"

The only way I've seen this happen before is when a cluster is fully restarted and the pdx metadata goes away, either because it was not persistent or because it was deleted (by clearing out the locator working directory for example).

GemFire clients cache the mapping between a type and it's type ID. This allows them to quickly serialize objects without continually looking up the type id from the server. Client connections can persist across cluster restarts. When a client reconnects it does not flush the cached information and continues to write objects using its cached type ID.

So the combination of a pdx-metadata losing cluster restart and a client that is not restarted (e.g. an app. server) is the only way I have seen this happen before. Does this match your scenario ?

If so, one of the best ways to avoid this is to persist your pdx metadata and never delete it.

You have a couple of options here, along with a few suggested recommendations.

1) First, I would not use Pivotal GemFire's o.a.g.pdx.ReflectionBasedAutoSerializer. Rather SDG has a much more robust PdxSerializer implementation based on Spring Data's Mapping Infrastructure (i.e. the o.s.d.g.mapping.MappingPdxSerializer).

In addition, SDG's MappingPdxSerializer allows you to register custom PdxSerializer's on an entity field/property case-by-case basis. Imagine if your Customer class has a reference to a complex Address class and that class has special serialization needs.

Furthermore, SDG's MappingPdxSerializer can handle transient and read-only properties.

Finally, you don't have to mess with any fussy/complex Regex to properly identify the application domain model types that need to be serialized.

2) Second, you can leverage Spring's JavaConfig along with SDG's new Annotation-based configuration model to configure Pivotal GemFire PDX Serialization as simply as this...

According to your issue, I assumed that there be connection issues between your client and documentdb endpoint after certain numbers of API call. You could try to create a static instance of DocumentClient and add the retry policy to your DocumentRepository class as follows:

You cannot update a column using the Query service. I recommend that you consider the Function service for transactional consistency. Make the region partitioned and call the function using the .onRegion().withFilter(key).withArgs(columnsAndValuesMap).

Your function will read the object, apply the updates, and put.

In this way your read and update will occur in a single thread on the server, ensuring transactional consistency rather than reading the object on the client, changing a value, doing a put and hoping that nobody else is slipping in underneath you.

Hate to answer my own questions, but here it is... after a while, I found out that if UnitOfWorkManager is not available for some reason, I can instantiate it in the code, by initializing IUnitOfWorkManager in the constructor. Then, you can simply use the following construction in your Save method:

I believe the problem is with how you're converting the binary data to string and then converting it back to byte array.

While converting binary data to string, you're using System.Convert.ToBase64String however while getting the byte array from string you're using Encoding.UTF8.GetBytes. Most likely this is causing the mismatch.

Please change the following line of code:

This is expected behavior. GetBlobReference simply creates an instance of CloudBlob on the client and doesn't make a network request. From the documentation link:

Call this method to return a reference to a blob of any type in this container. Note that this method does not make a request against Blob storage. You can return a reference to the blob whether or not it exists yet.

If you want to get the properties populated, you must call FetchAttributes or use GetBlobReferenceFromServer.

As you says IEnumerable can not add to viewstate and couldn't be a datasource to gridview.

As far as I know, IListBlobItem is an interface.The IEnumerable contains three type of blob items.