[XERCESJ-1653] Memory leak with validating SAX Parser

• Type: Bug
• Status: Open
• Priority: Critical
• Resolution: Unresolved
• Affects Version/s: 2.11.0
• Component/s: SAX
• Environment: Windows 7 Enterprise, JDK 1.8.0_25
• Assignee: Unassigned
• Reporter: Sebastian Millies
• Created: 07/Jan/15 15:00
• Updated: 08/May/17 15:31

Description

I'm parsing a very large XML file with org.apache.xerces.parsers.SAXParser and validation turned on. The file contains 25 million elements of the form specified in the attached DTD's, in total it is ca. 7 GB large.

Heap monitoring with jvisualvm shows millions of QName instances being cached and not being garbage collected.

Turning off validation makes the problem disappear.

I have tested a numer of other parsers (Crimson, Aelfred2, Resin, Woodstox). With Woodstox, for example, I can process my 7 GB file (including validation) with just 64MB of heap. With Xerces, 1024MB of heap do not suffice.

I'll attach a small diagnosis program (SAXMemoryUsage.java) that shows that Xerces heap consumption increases inordinately.