编辑大PDF时iText7（PDFSweep）抛出OutofMemory错误

Question

I was trying to remove some contents from PDF using PDFSweep, below are part of my code, I am using the CompositeCleanupStrategy and adding RegexBasedCleanupStrategy to the strategy:

    CompositeCleanupStrategy strategy = new CompositeCleanupStrategy();
    for (int i = 0; i < keywordlist.size(); i++) {

        String kvalue = keywordlist.get(i);


        Loger.getLogger().info("keyword " + i + "=" + kvalue);

        strategy.add(new RegexBasedCleanupStrategy(kvalue).setRedactionColor(ColorConstants.GRAY));

    }

   try {
        PdfWriter writer = new PdfWriter(dest);
        writer.setCompressionLevel(0);
        PdfDocument pdf = new PdfDocument(new PdfReader(src), writer);
        // sweep
        PdfAutoSweep pdfAutoSweep = new PdfAutoSweep(strategy);

        pdfAutoSweep.cleanUp(pdf);

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

When stragety is small, like there is only one or two , the cleanup is working fine, howere if there are 243 in the keywordlist, the PDF size is about 70 MB, I got following error:

 Exception in thread "main" java.lang.OutOfMemoryError: GC overhead limit 
 exceeded
at java.lang.String.toLowerCase(String.java:2590)
at java.lang.String.toLowerCase(String.java:2670)
at com.itextpdf.io.font.PdfEncodings.convertToString(PdfEncodings.java:287)
at com.itextpdf.kernel.pdf.PdfString.toUnicodeString(PdfString.java:163)
at com.itextpdf.kernel.pdf.canvas.parser.data.TextRenderInfo.getUnscaledBaselineWithOffset(TextRenderInfo.java:425)
at com.itextpdf.kernel.pdf.canvas.parser.data.TextRenderInfo.getBaseline(TextRenderInfo.java:213)
at com.itextpdf.kernel.pdf.canvas.parser.listener.CharacterRenderInfo.<init>(CharacterRenderInfo.java:112)
at com.itextpdf.kernel.pdf.canvas.parser.listener.RegexBasedLocationExtractionStrategy.toCRI(RegexBasedLocationExtractionStrategy.java:156)
at com.itextpdf.kernel.pdf.canvas.parser.listener.RegexBasedLocationExtractionStrategy.eventOccurred(RegexBasedLocationExtractionStrategy.java:135)
at com.itextpdf.pdfcleanup.autosweep.CompositeCleanupStrategy.eventOccurred(CompositeCleanupStrategy.java:115)
at com.itextpdf.kernel.pdf.canvas.parser.PdfCanvasProcessor.eventOccurred(PdfCanvasProcessor.java:534)
at com.itextpdf.kernel.pdf.canvas.parser.PdfCanvasProcessor.displayPdfString(PdfCanvasProcessor.java:549)
at com.itextpdf.kernel.pdf.canvas.parser.PdfCanvasProcessor.access$4700(PdfCanvasProcessor.java:108)
at com.itextpdf.kernel.pdf.canvas.parser.PdfCanvasProcessor$ShowTextArrayOperator.invoke(PdfCanvasProcessor.java:617)
at com.itextpdf.kernel.pdf.canvas.parser.PdfCanvasProcessor.invokeOperator(PdfCanvasProcessor.java:452)
at com.itextpdf.kernel.pdf.canvas.parser.PdfCanvasProcessor.processContent(PdfCanvasProcessor.java:281)
at com.itextpdf.kernel.pdf.canvas.parser.PdfCanvasProcessor.processPageContent(PdfCanvasProcessor.java:302)
at com.itextpdf.kernel.pdf.canvas.parser.PdfDocumentContentParser.processContent(PdfDocumentContentParser.java:77)
at com.itextpdf.kernel.pdf.canvas.parser.PdfDocumentContentParser.processContent(PdfDocumentContentParser.java:90)
at com.itextpdf.pdfcleanup.autosweep.PdfAutoSweep.getPdfCleanUpLocations(PdfAutoSweep.java:130)
at com.itextpdf.pdfcleanup.autosweep.PdfAutoSweep.cleanUp(PdfAutoSweep.java:186)

Answer 1

(Full disclosure: original author of RegexBasedCleanupStrategy here)

RegexBasedCleanupStrategy is not meant to be used like this. You are creating 200 instances of this class, all of which will go over the document to see whether they can match (chunk by chunk) the PDF against the regular expression.

In order to do this, they will store all chunks in the document, sort them, and then loop over them.

So you are duplicating the document 200-something times in memory. That is your bottleneck.

My suggestion: build a better regular expression. You can obviously match keyword a, b, c etc with regex

(a)|(b)|(c)

This would copy the document in memory only once, and then attempt to match the aggregate regex against it.

It has both performance, and memory-footprint benefits.