Java PriorityQueue：轮询（）然后再添加（）或窥视（）然后再删除（）是否更好？

Question

I have just finished my networked job queue and I have a little question relating to the performance of a PriortiyQueue in Java.

Take this code:

 private void performJob() {
    lock.lock();
    try {
        NetworkJob job = actions.poll();
        if (job.perform()) {
            return;
        }
        actions.add(job); //Job was a failure, add it back to the queue
    } finally {
        lock.unlock();
    }
}

in the case of the job failing, the job still needs to be in the queue. So, my question: Is it better to poll() and then add() or peek() and then remove()

I am personally leaning towards the code below, but considering that a job shouldn't really be failing (In most cases, assume it was a pass) is it better to just poll() ?

 private void performJob() {
    lock.lock();
    try {
        NetworkJob job = actions.peek();
        if (!job.perform()) {
            return;
        }
        actions.remove(); //Job was a success,  we can remove it from the queue.
    } finally {
        lock.unlock();
    }
}

Totally nitpicking and probably not worth worrying about due to the rarely-used nature of the queue, but it interests me and I'd like to see your reasoning.

Full code:

import android.content.BroadcastReceiver;
import android.content.Context;
import android.content.Intent;
import android.content.IntentFilter;
import android.net.ConnectivityManager;
import android.net.NetworkInfo;
import android.util.Log;

import java.util.PriorityQueue;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.locks.Condition;
import java.util.concurrent.locks.ReentrantLock;

public final class NetworkQueue implements Runnable {

    private final Context context;
    private final AtomicBoolean running = new AtomicBoolean(true);
    private final PriorityQueue<NetworkJob> actions = new PriorityQueue<>(5, new NetworkJobComparator());
    private final ReentrantLock lock = new ReentrantLock();
    private final Condition jobReady = lock.newCondition();
    private final Condition networkUp = lock.newCondition();

    private ConnectionType connection = ConnectionType.NONE;

    public NetworkQueue(Context context) {
        this.context = context;
        context.registerReceiver(new NetworkListener(),
                new IntentFilter(ConnectivityManager.CONNECTIVITY_ACTION));
    }

    @Override
    public void run() {
        try {
            while (running.get()) {
                waitJobAvailable();
                waitNetworkUp();
                performJob();
            }
        } catch (InterruptedException e) {
            e.printStackTrace();
        }
    }

    private void setNetwork(ConnectionType net) {
        lock.lock();
        try {
            connection = net;
            if (connection != ConnectionType.NONE) {
                networkUp.signal();
            }
        } finally {
            lock.unlock();
        }
    }

    private void waitNetworkUp() throws InterruptedException {
        lock.lock();
        try {
            while (connection != ConnectionType.NONE) {
                networkUp.await();
            }
        } finally {
            lock.unlock();
        }
    }


    private void waitJobAvailable() throws InterruptedException {
        lock.lock();
        try {
            while (actions.isEmpty()) {
                jobReady.await();
            }
        } finally {
            lock.unlock();
        }
    }

    private void performJob() {
        lock.lock();
        try {
            NetworkJob job = actions.peek();
            if (!job.perform()) {
                return;
            }
            actions.remove();
        } finally {
            lock.unlock();
        }
    }

    public boolean addJob(NetworkJob job) {
        lock.lock();
        try {
            if (this.actions.contains(job)) {
                return false;
            }
            this.actions.add(job);
            this.jobReady.signal();
            return true;
        } finally {
            lock.unlock();
        }
    }

    public void end() {
        this.running.set(false);
    }

    private class NetworkListener extends BroadcastReceiver {

        ConnectivityManager conn = (ConnectivityManager)
                context.getSystemService(Context.CONNECTIVITY_SERVICE);

        @Override
        public void onReceive(Context context, Intent intent) {
            NetworkInfo networkInfo = conn.getActiveNetworkInfo();
            if (networkInfo == null) {
                setNetwork(ConnectionType.NONE);
                return;
            }
            if (networkInfo.getType() == ConnectivityManager.TYPE_WIFI) {
                setNetwork(ConnectionType.WIFI);
                return;
            }
            setNetwork(ConnectionType.ANY);
        }
    }

}

Answer 1

In standard heap-based PriorityQueue implementation in OpenJDK and OracleJDK the peek() call is extremely fast:

public E peek() {
    return (size == 0) ? null : (E) queue[0];
}

That's because heap root is always the least element. In contrast removing and adding operations can be quite expensive as they may need to restructure the heap. Thus peek/remove solution is likely to be faster.

In my library I have an algorithm to select n least elements from the unsorted input. I implemented it using the PriorityQueue which preserves at most n least elements found so far. First implementation was like add/poll . When I updated to use peek , the performance was drastically improved (up to 10x on some tests).