Chapter 23. Cache Writing Modes

23.1. Cache Writing Modes

Red Hat JBoss Data Grid presents configuration options with a single or multiple cache stores. This allows it to store data in a persistent location, for example a shared JDBC database or a local file system. JBoss Data Grid supports two caching modes:

  • Write-Through (Synchronous)
  • Write-Behind (Asynchronous)

23.2. Write-Through Caching

23.2.1. Write-Through Caching

The Write-Through (or Synchronous) mode in Red Hat JBoss Data Grid ensures that when clients update a cache entry (usually via a Cache.put() invocation), the call does not return until JBoss Data Grid has located and updated the underlying cache store. This feature allows updates to the cache store to be concluded within the client thread boundaries.

23.2.2. Write-Through Caching Benefits and Disadvantages

Write-Through Caching Benefits

The primary advantage of the Write-Through mode is that the cache and cache store are updated simultaneously, which ensures that the cache store remains consistent with the cache contents.

Write-Through Caching Disadvantages

Due to the cache store being updated simultaneously with the cache entry, there is a possibility of reduced performance for cache operations that occur concurrently with the cache store accesses and updates.

23.2.3. Write-Through Caching Configuration (Library Mode)

No specific configuration operations are required to configure a Write-Through or synchronous cache store. All cache stores are Write-Through or synchronous unless explicitly marked as Write-Behind or asynchronous. The following procedure demonstrates a sample configuration file of a Write-Through unshared local file cache store.

Configure a Write-Through Local File Cache Store

<local-cache name="persistentCache">
		<persistence>
		    <file-store fetch-state="true"
			    purge="false"
			    shared="false"
			    location="${java.io.tmpdir}"/>
		</persistence>
</local-cache>

  1. The name parameter specifies the name of the local-cache to use.
  2. The fetch-state parameter determines whether the persistent state is fetched when joining a cluster. Set this to true if using a replication and invalidation in a clustered environment. Additionally, if multiple cache stores are chained, only one cache store can have this property enabled. If a shared cache store is used, the cache does not allow a persistent state transfer despite this property being set to true. The fetch-state parameter is false by default.
  3. The purge parameter specifies whether the cache is purged when initially started.
  4. The shared parameter is used when multiple cache instances share a cache store and is now defined at the cache store level. This parameter can be set to prevent multiple cache instances writing the same modification multiple times. Valid values for this parameter are true and false.

23.3. Write-Behind Caching

23.3.1. Write-Behind Caching

In Red Hat JBoss Data Grid’s Write-Behind (Asynchronous) mode, cache updates are asynchronously written to the cache store. Asynchronous updates ensure that cache store updates are carried out by a thread different from the client thread interacting with the cache.

One of the foremost advantages of the Write-Behind mode is that the cache operation performance is not affected by the underlying store update. However, because of the asynchronous updates, for a brief period the cache store contains stale data compared to the cache.

23.3.2. About Unscheduled Write-Behind Strategy

In the Unscheduled Write-Behind Strategy mode, Red Hat JBoss Enterprise Data Grid attempts to store changes as quickly as possible by applying pending changes in parallel. This results in multiple threads waiting for modifications to conclude. Once these modifications are concluded, the threads become available and the modifications are applied to the underlying cache store.

This strategy is ideal for cache stores with low latency and low operational costs. An example of this is a local unshared file based cache store in which the cache store is local to the cache itself. Using this strategy the period of time where an inconsistency exists between the contents of the cache and the contents of the cache store is reduced to the shortest possible interval.

23.3.3. Unscheduled Write-Behind Strategy Configuration (Remote Client-Server Mode)

To set the write-behind strategy in Red Hat JBoss Data Grid’s Remote Client-Server mode, add the write-behind element to the target cache store configuration as follows:

The write-behind Element

<file-store passivation="false"
            path="${PATH}"
            purge="true"
            shared="false">
    <write-behind modification-queue-size="1024"
                  shutdown-timeout="25000"
                  flush-lock-timeout="15000"
                  thread-pool-size="5" />
</file-store>

The write-behind element uses the following configuration parameters:

  1. The modification-queue-size parameter sets the modification queue size for the asynchronous store. If updates occur faster than the cache store can process the queue, the asynchronous store behaves like a synchronous store. The store behavior remains synchronous and blocks elements until the queue is able to accept them, after which the store behavior becomes asynchronous again.
  2. The shutdown-timeout parameter specifies the time in milliseconds after which the cache store is shut down. When the store is stopped some modifications may still need to be applied. Setting a large timeout value will reduce the chance of data loss. The default value for this parameter is 25000.
  3. The flush-lock-timeout parameter specifies the time (in milliseconds) to acquire the lock that guards the state to be periodically flushed. The default value for this parameter is 15000.
  4. The thread-pool-size parameter specifies the size of the thread pool. The threads in this thread pool apply modifications to the cache store. The default value for this parameter is 5.

23.3.4. Unscheduled Write-Behind Strategy Configuration (Library Mode)

To enable the write-behind strategy of the cache entries to a store, add the async element to the store configuration as follows:

The async Element

<persistence>
    <singleFile location="${LOCATION}">
        <async enabled="true"
                    modificationQueueSize="1024"
                    shutdownTimeout="25000"
                    flushLockTimeout="15000"
                    threadPoolSize="5"/>
    </singleFile>
</persistence>

  1. The async element uses the following configuration parameters: . The modificationQueueSize parameter sets the modification queue size for the asynchronous store. If updates occur faster than the cache store can process the queue, the asynchronous store behaves like a synchronous store. The store behavior remains synchronous and blocks elements until the queue is able to accept them, after which the store behavior becomes asynchronous again.
  2. The shutdownTimeout parameter specifies the time in milliseconds after which the cache store is shut down. This provides time for the asynchronous writer to flush data to the store when a cache is shut down. The default value for this parameter is 25000.
  3. The flushLockTimeout parameter specifies the time (in milliseconds) to acquire the lock that guards the state to be periodically flushed. The default value for this parameter is 15000.
  4. The threadPoolSize parameter specifies the number of threads that concurrently apply modifications to the store. The default value for this parameter is 5.