Netty 4 - Copy specific nr of bytes from ByteBuf into Native ByteBuffer

Question

I've been implementing Netty 4 on my app and I've ran into a limitation which I would like to optimize.

I'm using NioSocketChannel where when I receive data, the final object it needs to end up into is a Direct ByteBuffer Native,created as such: ByteBuffer.allocateDirect(var0).order(ByteOrder.nativeOrder()); This is hard requirement, I can't go around it.

To minimize object allocation, I'm using a ByteBufferPool and my ChannelRead looks something like this:

try{
   receivingBuf.writeBytes((ByteBuf) msg);

   if(receivingBuf.readableBytes() >= nextPacketLength){
      var bb = ByteBufferPoolManager.getByteBuffer();
      receivingBuf.readBytes(bb.array(), 0, nextPacketLength-12);
      bb.position(0);
      bb.limit(nextPacketLength - 12);

      MyCustomMethodToProcessData(bb);
   }
}
finally
{
   ((ByteBuf) msg).release();
}

Please ignore the -12 bytes, that is due to the format of my packets. Also, ByteBufferPoolManager.getByteBuffer() returns a normal heap ByteBuffer.

GOAL: Transfer a specific number of bytes from receivingBuf into a Direct Native Order allocated ByteBuffer.

I tried to do this:

var bb = ByteBufferPoolManager.getNativeByteBuffer();
receivingBuf.readBytes(bb.array(), 0, nextPacketLength-12);
bb.position(0);
bb.limit(nextPacketLength - 12);

MyCustomMethodToProcessData(bb);

But this doesn't work, as you can't access .array() on a Direct ByteBuffer. You get exception:

java.lang.UnsupportedOperationException
    at java.base/java.nio.ByteBuffer.array(ByteBuffer.java:1471)

If I do:

var bb = ByteBufferPoolManager.getNativeByteBuffer();
receivingBuf.readBytes(bb, 0, nextPacketLength-12);
...

This doesn't exist as an available method on ByteBuf.

Additionally, if I do:

var bb = ByteBufferPoolManager.getNativeByteBuffer();
receivingBuf.readBytes(bb);
...

Will cause an exception, as the Native ByteBuffer is massive (1000000 size), and ByteBuf will attempt to copy over 1 million bytes, while the available bytes on the ByteBuf is way less than that.

One option I've considered, is having a byte[] per channel context, which is used exclusively do to this data transfer, but still, I don't think I will have much improvements vs what I'm doing now.

Any ideas on how I can remove this middle hop in ByteBuf -> Heap ByteBuffer -> Native ByteBuffer?

Answer 1

You can modify the limit first

var bb = ByteBufferPoolManager.getNativeByteBuffer();
bb.limit(nextPacketLength-12);
receivingBuf.readBytes(bb);

Or you can duplicate bb and modify its limit. Don't worry about the overhead, because the duplicate shares the original's content, so the overhead is very small.

var bb = ByteBufferPoolManager.getNativeByteBuffer();
bb.duplicate().limit(nextPacketLength-12);
receivingBuf.readBytes(bb);