FOP-如何避免大量的页面序列占用大量内存？

Question

How can I avoid FOP to consume a growing amount of memory even when pages do not contain forward-references and < page-sequence> blocks are very small?

Here's a Test java program that feeds FOP with a hand made FO which just repeats over and over the same very basic page-sequence:

Fo2Pdf.java

import java.io.FileNotFoundException;
import java.io.FileOutputStream;
import java.io.OutputStream;
import java.io.PipedInputStream;

import javax.xml.transform.Result;
import javax.xml.transform.Source;
import javax.xml.transform.Transformer;
import javax.xml.transform.TransformerFactory;
import javax.xml.transform.sax.SAXResult;
import javax.xml.transform.stream.StreamSource;

import org.apache.fop.apps.FOUserAgent;
import org.apache.fop.apps.Fop;
import org.apache.fop.apps.FopFactory;
import org.apache.fop.apps.MimeConstants;
import org.xml.sax.helpers.DefaultHandler;

public class Fo2Pdf implements Runnable {

private PipedInputStream in;

public Fo2Pdf(PipedInputStream in)  {
    this.in = in;
}


@Override
public void run() {
    // instantiate Fop factory
    FopFactory fopFactory = FopFactory.newInstance();
    fopFactory.setStrictValidation(false);

    // Setup output
    OutputStream out = null;
    try {
        out = new FileOutputStream("output.pdf");
    } catch (FileNotFoundException e) {
        e.printStackTrace();
    }

    try {
        // Setup user agent
        FOUserAgent userAgent = fopFactory.newFOUserAgent();
        userAgent.setConserveMemoryPolicy(true);

        Fop fop = fopFactory.newFop(MimeConstants.MIME_PDF, userAgent, out);

        // Setup JAXP using identity transformer
        TransformerFactory factory = TransformerFactory.newInstance();
        Transformer transformer = factory.newTransformer(); 

        // Setup input stream
        Source src = new StreamSource(in);

        // Resulting SAX events (the generated FO) must be piped through to FOP
        DefaultHandler defaultHandler = (DefaultHandler) fop.getDefaultHandler();
        Result res = new SAXResult(defaultHandler);

        // Start FOP processing
        transformer.transform(src, res);

    } catch (Exception e) {
        e.printStackTrace();
    }
    }
}

FeedFo.java

import java.io.IOException;
import java.io.PipedInputStream;
import java.io.PipedOutputStream;


public class FeedFo {


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

    // instantiate and connect the pipes
    PipedInputStream in = new PipedInputStream();
    PipedOutputStream out = new PipedOutputStream(in);

    // Fo2Pdf - instantiate and start consuming the stream
    Fo2Pdf fo2Pdf = new Fo2Pdf(in);
    Thread fo2PdfThread = new Thread(fo2Pdf, "Fo2Pdf");
    fo2PdfThread.start();

    /*
     <fo:root xmlns:fo="http://www.w3.org/1999/XSL/Format">
        <fo:layout-master-set>
            <fo:simple-page-master master-name="A4" page-width="210mm" page-height="297mm">
                <fo:region-body/>
            </fo:simple-page-master>
        </fo:layout-master-set>

     */
    out.write(("<fo:root xmlns:fo=\"http://www.w3.org/1999/XSL/Format\"><fo:layout-master-set>" +
            "<fo:simple-page-master master-name=\"A4\" page-width=\"210mm\" page-height=\"297mm\">" +
            "<fo:region-body/></fo:simple-page-master></fo:layout-master-set>").getBytes());


    for(int i=0; i<100000000; i++) {

        // sleep 3 seconds every 50000 page-sequences to make sure the consumer is faster than the producer
        if(i % 50000 == 0) {
            Thread.currentThread().sleep(3000);
        }

        /*
         <fo:page-sequence xmlns:fo="http://www.w3.org/1999/XSL/Format" master-reference="A4">
            <fo:flow flow-name="xsl-region-body">
                <fo:block/>
            </fo:flow>
        </fo:page-sequence>
         */
        out.write(("<fo:page-sequence xmlns:fo=\"http://www.w3.org/1999/XSL/Format\" master-reference=\"A4\"><fo:flow flow-name=\"xsl-region-body\"><fo:block/></fo:flow></fo:page-sequence>").getBytes());
    }

    out.write("</fo:root>".getBytes());
    out.flush();
    out.close();

    fo2PdfThread.join();

    System.out.println("Exit");
}
}

As you notice, FOP writes to disk the PDF as soon as a page-sequence has been closed. This means that pages are (should?) not being kept into memory. But, memory just keeps growing and growing. With a 256MB heap size, generation stops at about 150000 page-sequences.

Why is this happening?

Answer 1

I suspect that, despite your sleep call, your producer is working much faster than your consumer and your piped stream is filling up your memory. Two ways I can think of to fix this:

Option 1 is to use a BlockingQueue instead of a piped stream.

Option 2 is to add a public boolean pipeIsFull() method to Fo2Pdf that returns true if in.available() exceeds, I dunno, 2mb. Then your main for loop will sleep for 500ms or whatever if pipeIsFull() is true.

Also, a way to reduce your memory consumption is

byte[] bytes = ("<fo:page-sequence xmlns:fo=\"http://www.w3.org/1999/XSL/Format\" master-reference=\"A4\"><fo:flow flow-name=\"xsl-region-body\"><fo:block/></fo:flow></fo:page-sequence>").getBytes();
for(int i=0; i<100000000; i++) {
    ...
    out.write(bytes);
}

I don't know how significant of an impact this will have (it'll reduce it by a couple gb, but that's probably peanuts compared to what Fo2Pdf is using), but it can't hurt.