PySpark Standalone：java.lang.IllegalStateException：未读块数据

Question

我对使用 pyspark 相当陌生，我一直在尝试运行一个脚本，该脚本在本地模式下运行良好，包含 1000 行数据子集，但现在在独立模式下对所有数据抛出错误，这是1GB。我认为这会随着更多数据 = 更多问题而发生，但我无法理解导致此问题的原因。以下是我的独立集群的详细信息：

• 3 名执行者
• 每个 20GB 内存
• spark.driver.maxResultSize=1GB（添加了这个，因为我认为这可能是问题，但它没有解决问题）

脚本在我将 spark 数据帧转换为 pandas 数据帧以并行化某些操作的阶段引发错误。我很困惑这会引起问题，因为数据只有 1G 左右，而我的执行程序应该有比这更多的内存。这是我的代码 sn-p - 错误发生在 data = data.toPandas():

def num_cruncher(data, cols=[], target='RETAINED', lvl='univariate'):
    if not cols:
            cols = data.columns
            del cols[data.columns.index(target)]
    data = data.toPandas()
    pop_mean = data.mean()[0]
    if lvl=='univariate':
        cols = sc.parallelize(cols)
        all_df = cols.map(lambda x: calculate([x], data, target)).collect()
    elif lvl=='bivariate':
        cols = sc.parallelize(cols)
        cols = cols.cartesian(cols).filter(lambda x: x[0]<x[1])
        all_df = cols.map(lambda x: calculate(list(x), data, target)).collect()
    elif lvl=='trivariate':
        cols = sc.parallelize(cols)
        cols = cols.cartesian(cols).cartesian(cols).filter(lambda x: x[0][0]<x[0][1] and x[0][0]<x[1] and x[0][1]<x[1]).map(lambda x: (x[0][0],x[0][1],x[1]))
        all_df = cols.map(lambda x: calculate(list(x), data, target)).collect()
    all_df = pd.concat(all_df)
    return all_df, pop_mean

这是错误日志：

    16/07/11 09:49:54 ERROR TransportRequestHandler: Error while invoking RpcHandler#receive() for one-way message.
java.lang.IllegalStateException: unread block data
    at java.io.ObjectInputStream$BlockDataInputStream.setBlockDataMode(ObjectInputStream.java:2424)
    at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1383)
    at java.io.ObjectInputStream.defaultReadFields(ObjectInputStream.java:1993)
    at java.io.ObjectInputStream.readSerialData(ObjectInputStream.java:1918)
    at java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:1801)
    at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1351)
    at java.io.ObjectInputStream.readObject(ObjectInputStream.java:371)
    at org.apache.spark.serializer.JavaDeserializationStream.readObject(JavaSerializer.scala:76)
    at org.apache.spark.serializer.JavaSerializerInstance.deserialize(JavaSerializer.scala:109)
    at org.apache.spark.rpc.netty.NettyRpcEnv$$anonfun$deserialize$1$$anonfun$apply$1.apply(NettyRpcEnv.scala:258)
    at scala.util.DynamicVariable.withValue(DynamicVariable.scala:57)
    at org.apache.spark.rpc.netty.NettyRpcEnv.deserialize(NettyRpcEnv.scala:310)
    at org.apache.spark.rpc.netty.NettyRpcEnv$$anonfun$deserialize$1.apply(NettyRpcEnv.scala:257)
    at scala.util.DynamicVariable.withValue(DynamicVariable.scala:57)
    at org.apache.spark.rpc.netty.NettyRpcEnv.deserialize(NettyRpcEnv.scala:256)
    at org.apache.spark.rpc.netty.NettyRpcHandler.internalReceive(NettyRpcEnv.scala:588)
    at org.apache.spark.rpc.netty.NettyRpcHandler.receive(NettyRpcEnv.scala:577)
    at org.apache.spark.network.server.TransportRequestHandler.processOneWayMessage(TransportRequestHandler.java:170)
    at org.apache.spark.network.server.TransportRequestHandler.handle(TransportRequestHandler.java:104)
    at org.apache.spark.network.server.TransportChannelHandler.channelRead0(TransportChannelHandler.java:104)
    at org.apache.spark.network.server.TransportChannelHandler.channelRead0(TransportChannelHandler.java:51)
    at io.netty.channel.SimpleChannelInboundHandler.channelRead(SimpleChannelInboundHandler.java:105)
    at io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:308)
    at io.netty.channel.AbstractChannelHandlerContext.fireChannelRead(AbstractChannelHandlerContext.java:294)
    at io.netty.handler.timeout.IdleStateHandler.channelRead(IdleStateHandler.java:266)
    at io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:308)
    at io.netty.channel.AbstractChannelHandlerContext.fireChannelRead(AbstractChannelHandlerContext.java:294)
    at io.netty.handler.codec.MessageToMessageDecoder.channelRead(MessageToMessageDecoder.java:103)
    at io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:308)
    at io.netty.channel.AbstractChannelHandlerContext.fireChannelRead(AbstractChannelHandlerContext.java:294)
    at org.apache.spark.network.util.TransportFrameDecoder.channelRead(TransportFrameDecoder.java:86)
    at io.netty.channel.AbstractChannelHandlerContext.invokeChannelRead(AbstractChannelHandlerContext.java:308)
    at io.netty.channel.AbstractChannelHandlerContext.fireChannelRead(AbstractChannelHandlerContext.java:294)
    at io.netty.channel.DefaultChannelPipeline.fireChannelRead(DefaultChannelPipeline.java:846)
    at io.netty.channel.nio.AbstractNioByteChannel$NioByteUnsafe.read(AbstractNioByteChannel.java:131)
    at io.netty.channel.nio.NioEventLoop.processSelectedKey(NioEventLoop.java:511)
    at io.netty.channel.nio.NioEventLoop.processSelectedKeysOptimized(NioEventLoop.java:468)
    at io.netty.channel.nio.NioEventLoop.processSelectedKeys(NioEventLoop.java:382)
    at io.netty.channel.nio.NioEventLoop.run(NioEventLoop.java:354)
    at io.netty.util.concurrent.SingleThreadEventExecutor$2.run(SingleThreadEventExecutor.java:111)
    at java.lang.Thread.run(Thread.java:745)

所以我的问题是：

1. 为什么给工作人员 20GB 的内存不足以容纳这个 1GB 的数据集？
2. 一般来说，像我在这里所做的那样将数据加载到内存中是个好主意，还是有更好的方法来做到这一点？

Answer 1

对于任何可能觉得这篇文章有用的人 - 似乎问题不是为工人/奴隶提供更多内存，而是为驱动程序提供更多内存，正如@KartikKannapur 在 cmets 中提到的那样。所以为了解决这个问题，我设置了：

spark.driver.maxResultSize 3g
spark.driver.memory 8g
spark.executor.memory 4g

可能有点矫枉过正，但它现在就可以了。