Search code examples
javanashorn

Nashorn OutOfMemoryError when building large js objects (strings)

I'm using Nashorn on 1.8u60 to create model objects to pass back to view tier (thymeleaf). Part of the model object is a somewhat large string (not big enough to cause any issues in plain java) containing HTML. When trying to convert the object back into Java using ScriptObjectMirror methods i'm hitting the following exception. Changing max heap size doesn't seem to have any effect ( changed from 900mb to 1800mb, same error). I couldn't find much online about this, but are there any restrictions that Nashorn places on object sizes? I'm going to try latest 1.8 JDK now.

java.lang.OutOfMemoryError: Java heap space
    at jdk.nashorn.internal.runtime.ConsString.flatten(ConsString.java:105)
    at jdk.nashorn.internal.runtime.ConsString.flattened(ConsString.java:98)
    at jdk.nashorn.internal.runtime.ConsString.toString(ConsString.java:69)
    at jdk.nashorn.api.scripting.ScriptObjectMirror.wrap(ScriptObjectMirror.java:704)
    at jdk.nashorn.api.scripting.ScriptObjectMirror.wrapLikeMe(ScriptObjectMirror.java:721)
    at jdk.nashorn.api.scripting.ScriptObjectMirror.wrapLikeMe(ScriptObjectMirror.java:730)
    at jdk.nashorn.api.scripting.ScriptObjectMirror.access$300(ScriptObjectMirror.java:64)
    at jdk.nashorn.api.scripting.ScriptObjectMirror$13.call(ScriptObjectMirror.java:371)
    at jdk.nashorn.api.scripting.ScriptObjectMirror$13.call(ScriptObjectMirror.java:364)
    at jdk.nashorn.api.scripting.ScriptObjectMirror.inGlobal(ScriptObjectMirror.java:859)
    at jdk.nashorn.api.scripting.ScriptObjectMirror.entrySet(ScriptObjectMirror.java:364)

...

Thanks,Adrian

Solution

  • That line reads

            final char[] chars = new char[length];

    so it appears there's indeed not enough memory for the final string. Nashorn uses ConsString as a way to amortize concatenation costs by delaying concatenation until the result is used (most JS engines use this optimization otherwise e.g. a concatenating lots of strings in a loop will require O(n^2) time).

    This means that you might have a result of many + operators on strings be a tree of ConsString objects that get "flattened" at once. The tradeoff for linearizing the time of the concatenation is the need to keep those ConsStrings around, which'll require more than twice the memory required for the string (more than twice 'cause of the ConsString objects own overhead).

    One way to get around this is to periodically invoke str.toString(). It is seemingly a no-op but internally it forces flattening of the concatenation tree. Try introducing it into your code at some point and see whether it helps.