I create 50GB files of marcxml all the time. We do NOT put a wrapper element around them, but do put a line feed at the end of each record. Then a trivial line reading loop in java/perl/whatever can read those records individually and process them appropriately.
That turns out the be the right way to do things in Hadoop too. Ralph -----Original Message----- From: Code for Libraries [mailto:CODE4LIB@LISTSERV.ND.EDU] On Behalf Of Kyle Banerjee Sent: Friday, June 08, 2012 2:36 PM To: CODE4LIB@LISTSERV.ND.EDU Subject: Best way to process large XML files I'm working on a script that needs to be able to crosswalk at least a couple hundred XML files regularly, some of which are quite large. I've thought of a number of ways to go about this, but I wanted to bounce this off the list since I'm sure people here deal with this problem all the time. My goal is to make something that's easy to read/maintain without pegging the CPU and consuming too much memory. The performance and load I'm seeing from running the files through LibXML and SimpleXML on the large files is completely unacceptable. SAX is not out of the question, but I'm trying to avoid it if possible to keep the code more compact and easier to read. I'm tempted to streamedit out all line breaks since they occur in unpredictable places and put new ones at the end of each record into a temp file. Then I can read the temp file one line at a time and process using SimpleXML. That way, there's no need to load giant files into memory, create huge arrays, etc and the code would be easy enough for a 6th grader to follow. My proposed method doesn't sound very efficient to me, but it should consume predictable resources which don't increase with file size. How do you guys deal with large XML files? Thanks, kyle <rant>Why the heck does the XML spec require a root element, particularly since large files usually consist of a large number of records/documents? This makes it absolutely impossible to process a file of any size without resorting to SAX or string parsing -- which takes away many of the advantages you'd normally have with an XML structure. </rant> -- ---------------------------------------------------------- Kyle Banerjee Digital Services Program Manager Orbis Cascade Alliance <baner...@uoregon.edu>baner...@orbiscascade.org / 503.999.9787
