泄漏：未调用 ByteBuf.release() - 我们如何解决这个问题？

Question

我们有一个基于网络的网络流量密集型 Java 应用程序/服务器。

旁注：我主要支持这个应用程序，我没有构建它，所以我不了解它。

我们有时会收到如下所示的错误。以前我们经常在服务器启动 3-4 天后收到此错误。现在我注意到我们在重新启动服务器/应用程序后仅 10-15 分钟就收到了这个错误。

我不明白这怎么可能。这个错误是否令人担忧，我们该如何解决？我记得过去对同样的错误进行了广泛的研究，当时我什至尝试升级和修补 netty，但没有任何帮助完全解决问题。

操作系统：Linux
Java 版本：1.8
Netty 版本：netty-all-4.1.30.Final.jar

这是唯一一行特定于应用程序的代码，其他一切都发生在 Netty 中。

com.company.japp.protocol.http.decoders.ConditionalHttpChunkAggregator.channelRead

这是 Netty 本身的某种错误吗？ Netty 升级或任何其他配置调整在这里有帮助吗？

[2020-09-04 08:33:53,072]

ERROR

io.netty.util.ResourceLeakDetector

LEAK: ByteBuf.release() was not called before it's garbage-collected. 

See https://netty.io/wiki/reference-counted-objects.html for more information.

Recent access records: 
Created at:
    io.netty.buffer.AbstractByteBufAllocator.compositeDirectBuffer(AbstractByteBufAllocator.java:221)
    io.netty.buffer.AbstractByteBufAllocator.compositeBuffer(AbstractByteBufAllocator.java:199)
    io.netty.handler.codec.MessageAggregator.decode(MessageAggregator.java:255)
    io.netty.handler.codec.MessageToMessageDecoder.channelRead(MessageToMessageDecoder.java:88)
    com.company.japp.protocol.http.decoders.ConditionalHttpChunkAggregator.channelRead(ConditionalHttpChunkAggregator.java:112)
    io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:362)
    io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:348)
    io.netty.channel.AbstractChannelHandlerContext.fireChannelRead(AbstractChannelHandlerContext.java:340)
    io.netty.handler.codec.ByteToMessageDecoder.fireChannelRead(ByteToMessageDecoder.java:323)
    io.netty.handler.codec.ByteToMessageDecoder.channelRead(ByteToMessageDecoder.java:297)
    io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:362)
    io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:348)
    io.netty.channel.AbstractChannelHandlerContext.fireChannelRead(AbstractChannelHandlerContext.java:340)
    io.netty.handler.timeout.IdleStateHandler.channelRead(IdleStateHandler.java:286)
    io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:362)
    io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:348)
    io.netty.channel.AbstractChannelHandlerContext.fireChannelRead(AbstractChannelHandlerContext.java:340)
    io.netty.channel.DefaultChannelPipeline$HeadContext.channelRead(DefaultChannelPipeline.java:1434)
    io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:362)
    io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:348)
    io.netty.channel.DefaultChannelPipeline.fireChannelRead(DefaultChannelPipeline.java:965)
    io.netty.channel.nio.AbstractNioByteChannel$NioByteUnsafe.read(AbstractNioByteChannel.java:163)
    io.netty.channel.nio.NioEventLoop.processSelectedKey(NioEventLoop.java:644)
    io.netty.channel.nio.NioEventLoop.processSelectedKeysOptimized(NioEventLoop.java:579)
    io.netty.channel.nio.NioEventLoop.processSelectedKeys(NioEventLoop.java:496)
    io.netty.channel.nio.NioEventLoop.run(NioEventLoop.java:458)
    io.netty.util.concurrent.SingleThreadEventExecutor$5.run(SingleThreadEventExecutor.java:897)
    io.netty.util.concurrent.FastThreadLocalRunnable.run(FastThreadLocalRunnable.java:30)
    java.lang.Thread.run(Thread.java:748)

这是ConditionalHttpChunkAggregator的代码。

package com.company.japp.protocol.http.decoders;

import com.company.japp.IHttpProxyServer;
import io.netty.channel.ChannelDuplexHandler;
import io.netty.channel.ChannelHandler;
import io.netty.channel.ChannelHandlerContext;
import io.netty.handler.codec.http.HttpHeaders;
import io.netty.handler.codec.http.HttpMessage;
import io.netty.handler.codec.http.HttpObjectAggregator;
import io.netty.handler.codec.http.HttpResponse;
import io.netty.util.internal.logging.InternalLogger;
import io.netty.util.internal.logging.InternalLoggerFactory;

import java.util.HashSet;

@ChannelHandler.Sharable
public class ConditionalHttpChunkAggregator extends HttpObjectAggregator {
    private static final InternalLogger logger = InternalLoggerFactory.getInstance(ConditionalHttpChunkAggregator.class);

    private volatile boolean        sendaschunked;
    private volatile int        maxContentLength;

    private static IHttpProxyServer         iHttpProxyServer;

    public static void initialize(IHttpProxyServer iHttpProxyServer) {
        ConditionalHttpChunkAggregator.iHttpProxyServer = iHttpProxyServer;
    }

    public ConditionalHttpChunkAggregator(int maxContentLength) {
        super(maxContentLength);
        this.maxContentLength = maxContentLength;
        sendaschunked = false;
    }

    @Override
    public void channelRead(ChannelHandlerContext ctx, Object msg){
        if ((msg instanceof HttpResponse)) {
            HttpResponse response = (HttpResponse)msg;
            if ((msg instanceof HttpMessage)) {
                HttpMessage httpmessage= (HttpMessage)msg;

                try  {
                    // If the content length exceeds the threshhold, then send it as chunked
                    // It's too large to process substitutions
                    Long contentlength = 
                            httpmessage.headers().get(HttpHeaders.Names.CONTENT_LENGTH) != null ? 
                            Long.valueOf(httpmessage.headers().get(HttpHeaders.Names.CONTENT_LENGTH)) : -1;
                    if (contentlength >= maxContentLength) {
                        sendaschunked = true;
                    } else {
                    // Check content types
                         HashSet<String> chunkabletypes = iHttpProxyServer.getConfig().getProperty("chunkabletypes");
                        if (!chunkabletypes.isEmpty() && response.headers().contains(HttpHeaders.Names.CONTENT_TYPE)) {
                            String contentType = response.headers().get(HttpHeaders.Names.CONTENT_TYPE).toLowerCase().trim();
                            if (contentType.length()>0) {
                                sendaschunked = chunkabletypes.contains(contentType);
                                if (!sendaschunked) {
                                    for (String chunkabletype: chunkabletypes) {
                                        // Begins with
                                        if (contentType.indexOf(chunkabletype)==0) {
                                            sendaschunked = true;
                                            break;
                                        }
                                    }
                                }
                            }
                        }
                    }
                    if (sendaschunked) {
                        ctx.fireChannelRead(msg);
                        return;
                    }
                }
                catch(Exception ex) {
                    logger.error("error determining chunkable viability", ex);
                }
            }
        }
        if (sendaschunked) {
            ctx.fireChannelRead(msg);
            return;
        }

        try {
            super.channelRead(ctx, msg);
            
        } catch (Exception e) {
            logger.error("error determining chunkable viability", e);
            e.printStackTrace();
        }
    }
}

这是可分块类型属性的值：

video/x-ms-wvx,video/x-flv,application/x-shockwave-flash,video/quicktime,video/,audio/

编辑

我认为这是 netty 4.1.30 中 io.netty.handler.codec.MessageAggregator 中第 255 行周围的错误。似乎这个 CompositeByteBuf 已分配但未释放。我对吗？我希望得到一些权威的答案来确认或拒绝这个想法。

        // A streamed message - initialize the cumulative buffer, and wait for incoming chunks.
        CompositeByteBuf content = ctx.alloc().compositeBuffer(maxCumulationBufferComponents); // LINE 255 
        if (m instanceof ByteBufHolder) {
            appendPartialContent(content, ((ByteBufHolder) m).content());
        }
        currentMessage = beginAggregation(m, content);

Answer 1

你需要释放你分配的Bytebuf。这不是 Netty 错误。

ByteBuf 位于com.company.japp.protocol.http.decoders.ConditionalHttpChunkAggregator.channelRead(ConditionalHttpChunkAggregator.java:112)

Answer 2

在我们的例子中，响应正文中的问题没有被读取，导致字节 buf 永远不会被释放。

原码：

WebClient client = WebClient.create("some_url");
return client.get().exchange().flatMap(response -> Mono.just(builder.up().build()));

固定代码：

WebClient client = WebClient.create("some_url");
return client.get().retrieve().toBodilessEntity().flatMap(response -> Mono.just(builder.up().build()));

即使是 exchange 的 javadoc 也这么说：

已弃用自 5.3 起，因为可能会泄漏内存和/或连接；请使用 exchangeToMono(Function), exchangeToFlux(Function);还可以考虑使用 retrieve()，它通过 ResponseEntity 提供对响应状态和标头的访问以及错误状态处理。

荣誉：https://stackoverflow.com/a/51321602/3242721

为了在 localhost 上重现问题，我使用了-Dio.netty.leakDetectionLevel=paranoid vm args。