Saxon

Legacy ID: #5247903 Legacy Poster: Alexandre Delarge (adelarge)

Hi, I'm using Saxon v8.7.3, and I'm caching compiled versions of XSL files in order to re-use it across multiple threads inside a web application. My question is related to the contention that happens on some of the underlying Saxon objects. Currently here is how I create the Saxon factory that will generate both my cached XSLT instances, and also standalone Saxon Transformer instances I may use. Note that I store one factory per execution thread: ----- // Create a Saxon Transformer/Templates factory TransformerFactory factory = TransformerFactoryImpl.newInstance(); // Use a distinct NamePool to avoid general contention on the default shared instance when creating Transformer instances through the factory ((TransformerFactoryImpl)factory).getConfiguration().setNamePool(new NamePool()); ----- And here is (basically) how I proceed to create and cache compiled instances: ----- // Get the XSL file as a source compatible object StreamSource xslt = new StreamSource(new FileInputStream(xslFile), xslFile.toURL().toExternalForm()); // Make a template from it using the thread's factory Templates template = factory.newTemplates(xslt); // Cache the template (simplified version...) cache.put(xslFile, template); ----- Now here is how I use the template: ----- // Retrieve the template (simplified version...) Templates template = cache.get(xslFile); // Create transformer Transformer tfm = template.newTransformer(); // Apply XSLT (input and output objects are supplied by the actual caller object) transformer.transform(in, out); ----- With my application running that code under stress conditions, I observe a recurring contention on the method "net.sf.saxon.om.NamePool.allocate". The pointed line in my thread dumps is "NamePool.java:433". I understand that actually all of the threads that use the template do share the same NamePool instance behind the PreparedStylesheet that is created when I compile the XSL file and cache it. As I understand it, the reason for that is the Controller object that is created via the call to "template.newTransformer()" does share its Configuration and Executable instance with the PreparedStylesheet. So I know that I could make my way to the NamePool instance that is used by each Transformer, and replace it with new instances in order to avoid re-using the one from the PreparedStylesheet. Unfortunately things don't seem to be that simple since method "transformDocument" in the Controller will check the same NamePool instance is used by both the input document and the target PreparedStylesheet. So my question is: how do I optimize the sharing of a same PreparedStylesheet across multiple threads ? The only 2 options I see are: - Make each thread use its own NamePool instance, and enforce its usage: is that enough or should I encounter issues in sharing other objects such as the PreparedStylesheet configuration, error listener, etc. ? - Should I perform a deep clone of the PreparedStylesheet, and would it address my need (and be efficient enough not to ruin the caching of the master PreparedStylesheet instance) ? Thanks a lot in advance for any advice on this topic. Regards, Alex.