记一次Java full gc的分析过程

有一次在开发环境做一个接口的压测时，结果很不理想。于是用命令查了一下进程的gc情况，每500毫秒查一次，连续查100次

jstat -gc 2847106 500 100

结果发现Full gc次数比Young gc次数多得多，高达1670次，平均每几秒就一次Full gc。

如此频繁的Full gc，可以看出每次Full gc之后都没能清掉堆内的对象释放内存，因此需要打印jvm内存中对象数量来看下是什么对象常驻内存没有释放。执行

 jmap -histo 2847106 | less

结果可以看出是groovy相关的对象占用了大部分的堆空间，问题是工程中并没有直接使用groovy，需要找出是哪个jar间接使用了groovy

原来是工程中使用了json-path来解释接口请求中的json入参（json-path可以使用搜索路径来直接读取json中的某个字段的值），而json-path使用了groovy脚本来读取json数据。

// json-path 将搜索路径都转成了 JSONAssertion
class JSONAssertion implements Assertion {
  String key;
  Map<String, Object> params;

  def Object getResult(object, config) {
    Object result = getAsJsonObject(object)
    return result;
  }

  def getAsJsonObject(object) {
    key = escapePath(key, hyphen(), attributeGetter(), integer(), properties(), classKeyword());
    def result;
    if (key == "\$" || key == "") {
      result = object
    } else {
      def root = 'restAssuredJsonRootObject'
      try {
        def expr;
        if (key =~ /^\[\d+\].*/) {
          expr = "$root$key"
        } else {
          expr = "$root.$key"
        }
        result = eval(root, object, expr)
      } catch (MissingPropertyException e) {
        // This means that a param was used that was not defined
        String error = String.format("The parameter \"%s\" was used but not defined. Define parameters using the JsonPath.params(...) function", e.property);
        throw new IllegalArgumentException(error, e);
      } catch (Exception e) {
        String error = e.getMessage().replace("startup failed:","Invalid JSON expression:").replace("$root.", generateWhitespace(root.length()));
        throw new IllegalArgumentException(error, e);
      }
    }
    return result
  }

  def String description() {
    return "JSON path"
  }

  private def eval(root, object, expr) {
      Map<String, Object> newParams;
      // Create parameters from given ones
      if(params!=null) {
          newParams=new HashMap<>(params);
      } else {
          newParams=new HashMap<>();
      }
      // Add object to evaluate
      newParams.put(root, object);
      // Create shell with variables set
      GroovyShell sh = new GroovyShell(new Binding(newParams));
      // Run
      return sh.evaluate(expr);
  }
}

那为什么使用groovy会有这样的问题呢？

每次groovy都会根据脚本内容生成一个class对象和生成一个新的classloader，由这个classloader去加载class

    public GroovyShell(ClassLoader parent, Binding binding, final CompilerConfiguration config) {
        if (binding == null) {
            throw new IllegalArgumentException("Binding must not be null.");
        }
        if (config == null) {
            throw new IllegalArgumentException("Compiler configuration must not be null.");
        }
        final ClassLoader parentLoader = (parent!=null)?parent:GroovyShell.class.getClassLoader();
        this.loader = AccessController.doPrivileged(new PrivilegedAction<GroovyClassLoader>() {
            public GroovyClassLoader run() {
                return new GroovyClassLoader(parentLoader,config);
            }
        });
        this.context = binding;        
        this.config = config;
    }

继续查看GroovyClassLoader的源码，发现有两个用于缓存的map，sourceCache用来缓存groovy脚本文件名，classCache用来缓存已编译的class。从源码中看出，脚本没有缓存到sourceCache，但会缓存到classCache，导致class对象没有及时被full gc回收。

    /**
     * this cache contains the loaded classes or PARSING, if the class is currently parsed
     */
    protected final Map<String, Class> classCache = new HashMap<String, Class>();

    /**
     * This cache contains mappings of file name to class. It is used
     * to bypass compilation.
     */
    protected final Map<String, Class> sourceCache = new HashMap<String, Class>();

    /**
     * Parses the groovy code contained in codeSource and returns a java class.
     */
    private Class parseClass(final GroovyCodeSource codeSource) throws CompilationFailedException {
        // Don't cache scripts
        return loader.parseClass(codeSource, false);
    }

    /**
     * Parses the given code source into a Java class. If there is a class file
     * for the given code source, then no parsing is done, instead the cached class is returned.
     *
     * @param shouldCacheSource if true then the generated class will be stored in the source cache
     * @return the main class defined in the given script
     */
    public Class parseClass(GroovyCodeSource codeSource, boolean shouldCacheSource) throws CompilationFailedException {
        synchronized (sourceCache) {
            Class answer = sourceCache.get(codeSource.getName());
            if (answer != null) return answer;
            answer = doParseClass(codeSource);
            if (shouldCacheSource) sourceCache.put(codeSource.getName(), answer);
            return answer;
        }
    }

    private Class doParseClass(GroovyCodeSource codeSource) {
        validate(codeSource);
        Class answer;  // Was neither already loaded nor compiling, so compile and add to cache.
        CompilationUnit unit = createCompilationUnit(config, codeSource.getCodeSource());
        if (recompile!=null && recompile || recompile==null && config.getRecompileGroovySource()) {
            unit.addFirstPhaseOperation(TimestampAdder.INSTANCE, CompilePhase.CLASS_GENERATION.getPhaseNumber());
        }
        SourceUnit su = null;
        File file = codeSource.getFile();
        if (file != null) {
            su = unit.addSource(file);
        } else {
            URL url = codeSource.getURL();
            if (url != null) {
                su = unit.addSource(url);
            } else {
                su = unit.addSource(codeSource.getName(), codeSource.getScriptText());
            }
        }

        ClassCollector collector = createCollector(unit, su);
        unit.setClassgenCallback(collector);
        int goalPhase = Phases.CLASS_GENERATION;
        if (config != null && config.getTargetDirectory() != null) goalPhase = Phases.OUTPUT;
        unit.compile(goalPhase);

        answer = collector.generatedClass;
        String mainClass = su.getAST().getMainClassName();
        for (Object o : collector.getLoadedClasses()) {
            Class clazz = (Class) o;
            String clazzName = clazz.getName();
            definePackageInternal(clazzName);
            setClassCacheEntry(clazz);
            if (clazzName.equals(mainClass)) answer = clazz;
        }
        return answer;
    }

找到原因后，项目中就将json-path去掉了。因为json的搜索路径是固定的，之前用json-path就是想少写点代码，现在知道有性能问题只能去掉了。

去掉json-path之后就没有频繁的Full gc了。