SavedModelBundle memory leak

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): Yes
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 18.04.5 LTS
• TensorFlow installed from (source or binary): Included in java artifact
• TensorFlow version: 2.4.1

Describe the current behavior
Loading and closing SavedModelBundle in an infinite loop leads to OutOfMemoryError:

Exception in thread "main" java.lang.OutOfMemoryError: Physical memory usage is too high: physicalBytes (7828M) > maxPhysicalBytes (7828M)
	at org.bytedeco.javacpp.Pointer.deallocator(Pointer.java:695)
	at org.tensorflow.internal.c_api.AbstractTF_Status.newStatus(AbstractTF_Status.java:70)
	at org.tensorflow.SavedModelBundle.load(SavedModelBundle.java:401)
	at org.tensorflow.SavedModelBundle.access$000(SavedModelBundle.java:59)
	at org.tensorflow.SavedModelBundle$Loader.load(SavedModelBundle.java:68)
	at org.tensorflow.SavedModelBundle.load(SavedModelBundle.java:242)
	at com.mycompany.app.App.main(App.java:13)

I use 0.3.1 version of the library. It is the only dependency:

<dependency>
  <groupId>org.tensorflow</groupId>
  <artifactId>tensorflow-core-platform</artifactId>
  <version>0.3.1</version>
</dependency>

The memory leak is reproducible with a model from this repository:
https://github.com/tensorflow/java/tree/master/tensorflow-core/tensorflow-core-api/src/test/resources/saved_model

Describe the expected behavior

I assume that the following code should run endlessly

Code to reproduce the issue

package com.mycompany.app;

import org.tensorflow.SavedModelBundle;
import java.net.URISyntaxException;
import java.nio.file.Paths;

public class App {
    public static void main(String[] args) throws URISyntaxException {
        var modelPath = Paths.get(App.class.getResource("/saved_model").toURI()).toString();
        while (true) {
            var model = SavedModelBundle.load(modelPath, "serve");
            model.close();
        }
    }
}

Commencez par la reproduction dans App.main, en vous concentrant sur les appels répétés à SavedModelBundle.load(modelPath, "serve") et model.close(), et utilisez la ressource saved_model mentionnée dans l’issue. Vérifiez que la mémoire continue d’augmenter au fil des itérations et que le chemin de chargement et de fermeture peut être exécuté plusieurs fois jusqu’à son terme sans provoquer d’OutOfMemoryError.