[ https://issues.apache.org/jira/browse/TIKA-818?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13190895#comment-13190895 ]
Paul Pearcy commented on TIKA-818: ---------------------------------- Hey Nick, Thanks a ton for taking a look! Apologies for the delay in response. The key trigger for PDFBox to use in-memory vs temporary file is the RandomAccess passed to the load method: http://www.jarvana.com/jarvana/view/org/apache/pdfbox/pdfbox/1.6.0/pdfbox-1.6.0-javadoc.jar!/org/apache/pdfbox/pdmodel/PDDocument.html#load(java.io.InputStream, org.apache.pdfbox.io.RandomAccess, boolean) Here is a sample I've been hacking around with: https://gist.github.com/1661161 The code probably isn't the best way to set things up for a couple of reasons: - It'd be nice to allow callers to pick memory or file buffers. Not sure what the correct approach would be to keep Tika interface clean. - I think TikaInputStream has its own temporary file resource management that should probably be used. Haven't figured that out yet. Thanks and Best Regards, Paul > Allow PDFBox to be used with RandomAccessFile vs RandomAccessBuffer to allow > for a memory vs performance tradeoff > ----------------------------------------------------------------------------------------------------------------- > > Key: TIKA-818 > URL: https://issues.apache.org/jira/browse/TIKA-818 > Project: Tika > Issue Type: Improvement > Components: parser > Affects Versions: 0.10, 1.0 > Reporter: Paul Pearcy > Original Estimate: 24h > Remaining Estimate: 24h > > After upgrading to Tika 0.10, began having OOM errors processing large > amounts of PDFs in parallel. The heap dump indicated that all the memory was > getting used up by PDFBox RandomAccessBuffers. After digging around, it looks > like PDFBox now defaults to using RAM vs temporary files for PDF extraction. > This can be overridden to use RandomAccessFiless. > I propose that Tika controls file vs buffer based on the inputstream type > received. If the TikaInputStream is a file, RandomAccessFile should be used > and for other stream types, RandomAccessBuffer can be used. > I believe the code to control this is here: > https://github.com/apache/tika/blob/trunk/tika-parsers/src/main/java/org/apache/tika/parser/pdf/PDFParser.java > At ~line 87: > PDDocument pdfDocument = > PDDocument.load(new CloseShieldInputStream(stream), true); > Not sure if this is the best approach and am curious if there are other ideas > on how to control this and keep the interface clean. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira