Windows non-sync writes

[pschichtel]

I've had a case where consecutive add()'s and remove()'s on a QueueFile will not wait for the write to sync to disk before removing()'ing it again.

public class Bench {

    public static void main(String[] args) throws IOException {

        Path tmp = Paths.get("tests-" + System.currentTimeMillis()).toAbsolutePath();
        Files.createDirectories(tmp);

        byte[] payload = new byte[] {0x11, 0x22, 0x33};
        QueueFile test = newTape(tmp, "test");

        long start = System.nanoTime();
        for (int i = 0; i < 1000; ++i) {
            test.add(payload);
            byte[] read = test.peek();
            if (read == null || read.length != payload.length) {
                throw new IllegalStateException();
            }
        }
        long delta = System.nanoTime() - start;
        System.out.println("Time taken: " + delta + "ns");
    }

    public static QueueFile newTape(Path tmp, String name) throws IOException {
        Path tapeFile = tmp.resolve(name + ".tape");
        return new QueueFile.Builder(tapeFile.toFile()).zero(false).build();
    }
}

On Linux it performs as I'd expect: Slower disks take longer then faster disks. On the Windows 10 system I tested this on a pretty slow HDD and it almost outperformed a tmpfs on my Linux system, which suggests that it's not actually writing to disk. I've made a custom build of tape2 where I added raf.getFD().sync(); after every call to raf.write() which brought the runtime back up to where I'd expect it.

[pschichtel]

This was on java 8.