Skip to main content

Toolkit - Tricks with Memory Byte Buffer

This post if part of a series about the File System Toolkit - a custom content delivery API for SDL Tridion.

In previous post Writing My Own Database Engine, I quickly mentioned the use of Memory Byte Buffer from Jana NIO that provides fast access to a file by mapping its content to memory. This post goes into more detail over some tricks that occurred with that implementation.

There is an issue with Memory Byte Buffer. Namely, once it is created by calling FileChannel.map method, it cannot be unmapped, closed or discarded. The byte buffer will exist until it is garbage collected.

From the JavaDoc:

A mapping, once established, is not dependent upon the file channel that was used to create it. Closing the channel, in particular, has no effect upon the validity of the mapping.

A mapped byte buffer and the file mapping that it represents remain valid until the buffer itself is garbage-collected.


The issue is affects Windows OS Java implementations, in the sense that it keep portions of the mapped file open and unavailable to modifications. The index file must be modifiable at any time because of publish/unpublish activities that must be able to modify the index.

Attempting to modify the index file on Windows OS will result in a FileNotFoundException (The requested operation cannot be per formed on a file with a user-mapped section open)

In order to work around this issue, I created a big hack. I created my own factory that creates and destroys MemoryByteBuffers.

public enum MemoryBufferFactory {

    INSTANCE;
    MemoryBufferFactory() {
    }

    public MemoryBuffer getBuffer(FileChannel fileChannel, FileChannel.MapMode mode) throws IOException {
        return new MappedMemoryBuffer(fileChannel, mode);
    }
}

The MemoryBuffer is a interface that defines the operations on a wrapped byte buffer:

public interface MemoryBuffer {
    int capacity();
    byte get(int position);
    void close();
}

The MappedMemoryBuffer implements the MemoryBuffer interface and wraps a java.nio.MappedByteBuffer object. The only trick it does is that in its close() method, it calls an unsupported, private unmap method of the MappedByteBuffer.

public class MappedMemoryBuffer implements MemoryBuffer {

    private final MappedByteBuffer buffer;
    private final FileChannel fileChannel;

    public MappedMemoryBuffer(FileChannel fileChannel, FileChannel.MapMode mode) {
        this.fileChannel = fileChannel;
        buffer = fileChannel.map(mode, 0, fileChannel.size());
    }

    public int capacity() {
        return buffer.capacity();
    }

    public byte get(int position) {
        return buffer.get(position);
    }

    public void close() {
        Class<?> clazz = fileChannel.getClass();
        Method method = clazz.getDeclaredMethod("unmap",
            new Class[]{MappedByteBuffer.class});
        method.setAccessible(true);
        method.invoke(null, new Object[]{buffer});
    }
}



Comments

Popular posts from this blog

Toolkit - Dynamic Content Queries

This post if part of a series about the  File System Toolkit  - a custom content delivery API for SDL Tridion. This post presents the Dynamic Content Query capability. The requirements for the Toolkit API are that it should be able to provide CustomMeta queries, pagination, and sorting -- all on the file system, without the use third party tools (database, search engines, indexers, etc). Therefore I had to implement a simple database engine and indexer -- which is described in more detail in post Writing My Own Database Engine . The querying logic does not make use of cache. This means the query logic is executed every time. When models are requested, the models are however retrieved using the ModelFactory and those are cached. Query Class This is the main class for dynamic content queries. It is the entry point into the execution logic of a query. The class takes as parameter a Criterion (presented below) which triggers the execution of query in all sub-criteria of a Criterio

A DD4T.net Implementation - Custom Binary Publisher

The default way to publish binaries in DD4T is implemented in class DD4T.Templates.Base.Utils.BinaryPublisher and uses method RenderedItem.AddBinary(Component) . This produces binaries that have their TCM URI as suffix in their filename. In my recent project, we had a requirement that binary file names should be clean (without the TCM URI suffix). Therefore, it was time to modify the way DD4T was publishing binaries. The method in charge with publishing binaries is called PublishItem and is defined in class BinaryPublisher . I therefore extended the BinaryPublisher and overrode method PublishItem. public class CustomBinaryPublisher : BinaryPublisher { private Template currentTemplate; private TcmUri structureGroupUri; In its simplest form, method PublishItem just takes the item and passes it to the AddBinary. In order to accomplish the requirement, we must specify a filename while publishing. This is the file name part of the binary path of Component.BinaryConten

Scaling Policies

This post is part of a bigger topic Autoscaling Publishers in AWS . In a previous post we talked about the Auto Scaling Groups , but we didn't go into details on the Scaling Policies. This is the purpose of this blog post. As defined earlier, the Scaling Policies define the rules according to which the group size is increased or decreased. These rules are based on instance metrics (e.g. CPU), CloudWatch custom metrics, or even CloudWatch alarms and their states and values. We defined a Scaling Policy with Steps, called 'increase_group_size', which is triggered first by the CloudWatch Alarm 'Publish_Alarm' defined earlier. Also depending on the size of the monitored CloudWatch custom metric 'Waiting for Publish', the Scaling Policy with Steps can add a difference number of instances to the group. The scaling policy sets the number of instances in group to 1 if there are between 1000 and 2000 items Waiting for Publish in the queue. It also sets the