如何在多线程模式下读取文件？

Question

I currently have a program that reads file (very huge) in single threaded mode and creates search index but it takes too long to index in single threaded environment.

Now I am trying to make it work in multithreaded mode but not sure the best way to achieve that.

My main program creates a buffered reader and passes the instance to thread and the thread uses the buffered reader instance to read the files.

I don't think this works as expected rather each thread is reading the same line again and again.

Is there a way to make the threads read only the lines that are not read by other thread? Do I need to split the file? Is there a way to implement this without splitting the file?

Sample Main program:

import java.io.BufferedReader;
import java.io.FileNotFoundException;
import java.io.FileReader;
import java.util.ArrayList;

public class TestMTFile {
    public static void main(String args[]) {
        BufferedReader reader = null;
        ArrayList<Thread> threads = new ArrayList<Thread>();
        try {
            reader = new BufferedReader(new FileReader(
                    "test.tsv"));
        } catch (FileNotFoundException e1) {
            e1.printStackTrace();
        }
        for (int i = 0; i <= 10; i++) {
            Runnable task = new ReadFileMT(reader);
            Thread worker = new Thread(task);
            // We can set the name of the thread
            worker.setName(String.valueOf(i));
            // Start the thread, never call method run() direct
            worker.start();
            // Remember the thread for later usage
            threads.add(worker);
        }

        int running = 0;
        int runner1 = 0;
        int runner2 = 0;
        do {
            running = 0;
            for (Thread thread : threads) {
                if (thread.isAlive()) {
                    runner1 = running++;
                }
            }
            if (runner2 != runner1) {
                runner2 = runner1;
                System.out.println("We have " + runner2 + " running threads. ");

            }
        } while (running > 0);

        if (running == 0) {
            System.out.println("Ended");
        }
    }
}

Thread:

import java.io.BufferedReader;
import java.io.IOException;

public class ReadFileMT implements Runnable {
    BufferedReader bReader = null;

    ReadFileMT(BufferedReader reader) {
        this.bReader = reader;
    }

    public synchronized void run() {
        String line;
        try {
            while ((line = bReader.readLine()) != null) {

                try {
                    System.out.println(line);
                } catch (Exception e) {

                }
            }
        } catch (IOException e) {
            // TODO Auto-generated catch block
            e.printStackTrace();
        }
    }
}

Answer 1

Your bottleneck is most likely the indexing, not the file reading. assuming your indexing system supports multiple threads, you probably want a producer/consumer setup with one thread reading the file and pushing each line into a BlockingQueue (the producer), and multiple threads pulling lines from the BlockingQueue and pushing them into the index (the consumers).

Answer 2

看到这个线程 - 如果你的文件都在同一个磁盘上，那么用一个线程读取它们就不会做得更好，尽管一旦你将它们读入主存储器就可以处理多个线程的文件。

Answer 3

If you can use Java 8, you may be able to do this quickly and easily using the Streams API. Read the file into a MappedByteBuffer, which can open a file up to 2GB very quicky, then read the lines out of the buffer (you need to make sure your JVM has enough extra memory to hold the file):

package com.objective.stream;

import java.io.BufferedReader;
import java.io.ByteArrayInputStream;
import java.io.FileInputStream;
import java.io.IOException;
import java.io.InputStreamReader;
import java.nio.MappedByteBuffer;
import java.nio.channels.FileChannel;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.util.stream.Stream;

public class StreamsFileProcessor {
    private MappedByteBuffer buffer;

    public static void main(String[] args){
        if (args[0] != null){
            Path myFile = Paths.get(args[0]);
            StreamsFileProcessor proc = new StreamsFileProcessor();
            try {
                proc.process(myFile);
            } catch (IOException e) {
                e.printStackTrace();
            }   
        }
    }

    public void process(Path file) throws IOException {
        readFileIntoBuffer(file);
        getBufferStream().parallel()
            .forEach(this::doIndex);
    }

    private Stream<String> getBufferStream() throws IOException {
        try (BufferedReader reader = new BufferedReader(new InputStreamReader(new ByteArrayInputStream(buffer.array())))){
            return reader.lines();
        }
    }

    private void readFileIntoBuffer(Path file) throws IOException{
        try(FileInputStream fis = new FileInputStream(file.toFile())){
            FileChannel channel = fis.getChannel();
            buffer = channel.map(FileChannel.MapMode.PRIVATE, 0, channel.size());
        }
    }

    private void doIndex(String s){
        // Do whatever I need to do to index the line here
    }
}

Answer 4

First, I agree with @Zim-Zam that it is the file IO, not the indexing, that is likely the rate determining step. (So I disagree with @jtahlborn). Depends on how complex the indexing is.

Second, in your code, each thread has it's own, independent BufferedReader . Therefore they will all read the entire file. One possible fix is to use a single BufferedReader that they share. And then you need to synchronize the BufferedReader.readLine() method (I think) since the javadocs are silent on whether BufferedReader is thread-safe. And, since I think the IO is the botleneck, this will become the bottleneck and I doubt if multithreading will gain you much. But give it a try, I have been wrong occasionally. :-)

ps I agree with @jtahlmorn that a producer/consumer pattern is better than my share the BufferedReader idea, but that would be much more work for you.