Applications that use Caching Application Block may experience exceptions or may receive incorrect data from the cache (884191)

MORE INFORMATION

Exception safety

Data and metadata are processed separately in the old caching block. Data is added to the store synchronously as the user requests the data, but the metadata is added in a background thread. There is no error recovery code that returns the cache to a consistent state if either part of this two-phase process fails.

Thread safety

The original caching block protects itself against users who may call its API-level methods at the same time, but the protection is not extended down to the CacheService. The CacheService is an asynchronous subsystem in the cache that is responsible for all operations on metadata. These operations include storing and retrieving metadata from the cache's in-memory store and performing scavenging and expirations. There is no class that holds or describes the metadata, and the metadata is a series of unrelated data elements that are stored in different collections in the cache service. While the CacheService operates on this data, the CacheService locks each piece of data that it is accessing, but the CacheService never locks the metadata that is associated with a particular piece of cached data. If contention occurs between two or more threads, the data and the metadata that is stored in the cache may be out of sync. This scenario is more likely when the cache is more heavily used. If one cache transaction occurs each second, this scenario is unlikely to occur.

Persistent storage

Metadata is never updated in a persistent way. When the cache operates in a non-persistent mode, the cache starts with an empty cache every time. Data and metadata are added to the cache and are processed together from that point. As items are updated, their last accessed times are kept up-to-date in the scavenging and expiration metadata. If the cache is made persistent, the updates to the metadata are still made to the in-memory metadata and are never written back to the persistent store. If the cache is restarted, the existing data items are read from the persistent store as they are accessed, but metadata is read from the persistent store in its entirety at startup and is stored in-memory. All the timestamps that are used by scavenging and expirations are reset to their original at-creation values, and any indications of updates are lost. Because the indications of updates are lost, data may not expire when it should expire or items that are not correct may be scavenged because of last accessed timestamps that are not correct.

Another problem with persistent storage is that the behavior of the data in the cache is different in this mode when compared to in-memory caches. In-memory caches offer live object references. Therefore, any changes to an item that are made by a user after the data is retrieved from the cache are reflected when the same cache item is retrieved from the cache again. The object is live and changing, and those changed values are available through the cache. If the cache is persistent, a new instance of the object is created and returned every time the user requests the object. This fundamental difference in caching behavior between in-memory caches and persistent storage must be reflected in the user code.