Memory optimization for JAXB parsing

On some project we had a Java program that needed to parse a lot of small XML messages continuously that uses a lot of memory.

After we made a heap dump and analyzed it we found the problem, a big chunk of the memory was used by Strings from the JAXB Objects. We had over 1000 times the string “de-de”.

So we did a small optimization and reduced the memory usage by 30%.

The solution was to register an XMLAdapter for String objects, that ensured that if a string occurs multiple times only on String instances is used and shared.

Here is a (simplified) implementation:

package de.nilsrudolph.blog;

import jakarta.xml.bind.annotation.adapters.XmlAdapter;

import java.util.concurrent.ConcurrentHashMap;
import java.util.function.Function;

public class CachingXmlAdapter extends XmlAdapter<String, String> {

    private static final ConcurrentHashMap<String, String> STRING_CACHE = new ConcurrentHashMap<>();

    @Override
    public String unmarshal(String s) {
        return STRING_CACHE.computeIfAbsent(s, Function.identity());
    }

    @Override
    public String marshal(String s)  {
        return s;
    }
}

For production use, the STRING_CACHE should probably be replaces by some other caching logic. A normal Map will never evict entries can become very big.

Now you only need to register the XMLAdapter, for example by adding or editing the package-info.java file of the package containing the JAXB classes:

@XmlJavaTypeAdapters({
        @XmlJavaTypeAdapter(value = CachingXmlAdapter.class, type = String.class)
})
package de.nilsrudolph.blog;

import jakarta.xml.bind.annotation.adapters.XmlJavaTypeAdapter;
import jakarta.xml.bind.annotation.adapters.XmlJavaTypeAdapters;