Groovy 腳本引發的 Old GC問題

近期上線了一個系統，鑑權部分使用了Groovy腳本，示例代碼如下

ScriptEngineManager factory = new ScriptEngineManager();
ScriptEngine engine = factory.getEngineByName("groovy");

String function = String.format("def getTargetParamValue(%s) {return \"%s\"}", "o", "$o");
engine.eval(function);
Invocable invocable = (Invocable) engine;

Object result = invocable.invokeFunction("getTargetParamValue", "test-string");
System.out.println(result);        

這段代碼定義了一個Groovy的方法，根據傳進去的參數返回對應的值。

由於生產環境流量很大，這段代碼被頻繁執行。測試時的代碼如下

public class ScriptEngineTest {

    public static void main(String[] args) {
        ScriptEngineManager factory = new ScriptEngineManager();
        ScriptEngine engine = factory.getEngineByName("groovy");
        //測試時改爲死循環
        for (int i = 0;; i++) {
            try {

                String function = String.format("def getTargetParamValue(%s) {return \"%s\"}", "o", "$o");
                engine.eval(function);
                Invocable invocable = (Invocable) engine;

                Object result = invocable.invokeFunction("getTargetParamValue", "test-string");
                System.out.println(result);

                TimeUnit.MICROSECONDS.sleep(100);

                System.out.println(new Date().toLocaleString());

            } catch (Exception e) {
                String errorMsg = String.format("異常！%s", e.getMessage());
                System.out.println(errorMsg);
            }

        }
    }
}

模擬生產環境的情況，每秒鐘執行10次。通過VusualVM觀察JVM

• CPU使用情況，可以看到在每次堆內存擴容的時候，CPU使用量會有明顯增加
  

• 堆內存使用情況
  

• metaspace使用量一直在增加
  

• 類加載情況，total loaded classes一直在增加
  

• 線程
  

• dump內存
  

可見，每次循環中生成的 Groovy method在方法執行完成之後並沒有被釋放掉，導致metaspace的使用量一直增加，最終撐爆JVM

針對以上問題，解決方法爲每次將生成的方法緩存下了，下次要執行的時候從緩存中取。

private final ConcurrentHashMap<String, Invocable> concurrentHashMap = new ConcurrentHashMap<>();

private Object getInvokeResult(Object targetParam, String paramName, String expression) throws Exception {
    //targetParamClassName="com.umgsai.web.home.vo.NodeVO"
    String targetParamClassName = targetParam.getClass().getName();
    //expression="$nodeVO.bizOwner"
    //paramName="nodeVO"
    String functionKey = String.format("%s_%s_%s", targetParamClassName, paramName, expression);
    functionKey = StringUtil.replaceChars(functionKey, "$", "");
    functionKey = StringUtil.replaceChars(functionKey, ".", "_");
    //functionKey爲方法的名稱和concurrentHashMap的key，這裏需要去掉特殊字符

    Invocable invocable = concurrentHashMap.get(functionKey);
    if (invocable != null) {
        //如果緩存中有，直接調用
        return invocable.invokeFunction(functionKey, targetParam);
    }

   //如果緩存中沒有，生成方法，並且存到concurrentHashMap
    synchronized (lock) {
        invocable = concurrentHashMap.get(functionKey);
        if (invocable == null) {
            String function = String.format("def %s(%s) {return \"%s\"}", functionKey, paramName, expression);
            engine.eval(function);
            invocable = (Invocable) engine;
            concurrentHashMap.put(functionKey, invocable);
            if (log.isInfoEnabled()) {
                String msg = String.format("Create new Groovy function, functionKey=%s, paramName=%s, expression=%s",
                        functionKey,
                        paramName, expression);
                log.info(msg);
            }
        }
    }

    if (log.isInfoEnabled()) {
        log.info(String.format("Groovy function concurrentHashMap.size=%d", concurrentHashMap.size()));
    }

    return invocable.invokeFunction(functionKey, targetParam);
}

參考 http://www.zgxue.com/120/1204001.html
https://www.cnblogs.com/fourspirit/p/4332154.html
https://www.jianshu.com/p/b1a46cc02377

轉載請註明出處 https://www.cnblogs.com/umgsai/p/10742271.html