[jira] [Created] (HBASE-21973) NettyRpcServer performance improve based on Netty

["Nicholas Jiang (JIRA)" <ji...@apache.org>]

Nicholas Jiang created HBASE-21973:
--------------------------------------

             Summary: NettyRpcServer performance improve based on Netty
                 Key: HBASE-21973
                 URL: https://issues.apache.org/jira/browse/HBASE-21973
             Project: HBase
          Issue Type: Improvement
            Reporter: Nicholas Jiang


In NettyRpcServer#NettyRpcServer constructor method, we have the following:

{code:java}
    ServerBootstrap bootstrap = new ServerBootstrap().group(eventLoopGroup).channel(channelClass)
        .childOption(ChannelOption.TCP_NODELAY, tcpNoDelay)
        .childOption(ChannelOption.SO_KEEPALIVE, tcpKeepAlive)
        .childHandler(new ChannelInitializer<Channel>() {

          @Override
          protected void initChannel(Channel ch) throws Exception {
            ChannelPipeline pipeline = ch.pipeline();
            FixedLengthFrameDecoder preambleDecoder = new FixedLengthFrameDecoder(6);
            preambleDecoder.setSingleDecode(true);
            pipeline.addLast("preambleDecoder", preambleDecoder);
            pipeline.addLast("preambleHandler", createNettyRpcServerPreambleHandler());
            pipeline.addLast("frameDecoder", new NettyRpcFrameDecoder(maxRequestSize));
            pipeline.addLast("decoder", new NettyRpcServerRequestDecoder(allChannels, metrics));
            pipeline.addLast("encoder", new NettyRpcServerResponseEncoder(metrics));
          }
        });
    try {
      serverChannel = bootstrap.bind(this.bindAddress).sync().channel();
      LOG.info("Bind to {}", serverChannel.localAddress());
    } catch (InterruptedException e) {
      throw new InterruptedIOException(e.getMessage());
    }
{code}
In build ServerBootstrap, we would configure ServerSocketChannel options and SocketChannel child options to improve rpc perfermance.These options and child options are as follows:

{code:java}
 .option(ChannelOption.SO_BACKLOG, transportConfig.getBacklog())
                .option(ChannelOption.SO_REUSEADDR, transportConfig.isReuseAddr())
                .option(ChannelOption.RCVBUF_ALLOCATOR, NettyHelper.getRecvByteBufAllocator())
                .option(ChannelOption.ALLOCATOR, NettyHelper.getByteBufAllocator())
                .childOption(ChannelOption.SO_KEEPALIVE, transportConfig.isKeepAlive())
                .childOption(ChannelOption.TCP_NODELAY, transportConfig.isTcpNoDelay())
                .childOption(ChannelOption.SO_RCVBUF, 8192 * 128)
                .childOption(ChannelOption.SO_SNDBUF, 8192 * 128)
                .childOption(ChannelOption.ALLOCATOR, NettyHelper.getByteBufAllocator())
                .childOption(ChannelOption.WRITE_BUFFER_WATER_MARK, new WriteBufferWaterMark(
                    transportConfig.getBufferMin(), transportConfig.getBufferMax()))
{code}
What's more,ChannelPipeline includes NettyRpcFrameDecoder,this decorder extends ByteToMessageDecoder.ChannelPipeline is as follows:

{code:java}
 .childHandler(new ChannelInitializer<Channel>() {

          @Override
          protected void initChannel(Channel ch) throws Exception {
            ChannelPipeline pipeline = ch.pipeline();
            FixedLengthFrameDecoder preambleDecoder = new FixedLengthFrameDecoder(6);
            preambleDecoder.setSingleDecode(true);
            pipeline.addLast("preambleDecoder", preambleDecoder);
            pipeline.addLast("preambleHandler", createNettyRpcServerPreambleHandler());
            pipeline.addLast("frameDecoder", new NettyRpcFrameDecoder(maxRequestSize));
            pipeline.addLast("decoder", new NettyRpcServerRequestDecoder(allChannels, metrics));
            pipeline.addLast("encoder", new NettyRpcServerResponseEncoder(metrics));
          }
        });
{code}

Netty provides a convenient decoding tool class ByteToMessageDecoder , this class has accumulate bulk unpacking ability, can read bytes from the socket as much as possible, then synchronously call the decode method, decode the business object, and form a List. Finally, the traversal traverses the List and submits it to ChannelPipeline for processing.

Here we can make a small change, submit the submitted content from a single command to the entire List, which can reduce the number of pipeline execution and increase throughput. This mode has no advantage in low-concurrency scenarios, and has a significant performance boost in boost throughput in high-concurrency scenarios.

Will provide an patch and some perf-comparison for this.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)