注意?。?!HotswapAgent引發(fā)了OOM
這幾天在壓測服務(wù),結(jié)果出現(xiàn)了Metaspace溢出,服務(wù)OOM了。本文從問題開始找到查找問題,到解決問題。
問題現(xiàn)象
服務(wù)在壓測過程中,出現(xiàn)了Metaspace空間不足,服務(wù)OOM的情況,經(jīng)過在線分析,發(fā)現(xiàn)是一段創(chuàng)建任務(wù)的邏輯,在運(yùn)行過程中創(chuàng)建并加載了很多相同的淚。
其中一個線程棧如下:
dynamic-kafka-worker-pool-sync_staging-async_topic-async_topic_c-3" Id=575 BLOCKED on org.springframework.boot.loader.LaunchedURLClassLoader@43e9089 owned by "dynamic-kafka-worker-pool-sync_staging-async_topic-async_topic_c-1" Id=572
at org.springframework.cglib.core.AbstractClassGenerator.generate(AbstractClassGenerator.java:344)
- blocked on org.springframework.boot.loader.LaunchedURLClassLoader@43e9089
at org.springframework.cglib.proxy.Enhancer.generate(Enhancer.java:582)
at org.springframework.cglib.core.AbstractClassGenerator$ClassLoaderData.get(AbstractClassGenerator.java:131)
at org.springframework.cglib.core.AbstractClassGenerator.create(AbstractClassGenerator.java:319)
at org.springframework.cglib.proxy.Enhancer.createHelper(Enhancer.java:569)
at org.springframework.cglib.proxy.Enhancer.createClass(Enhancer.java:416)
at org.springframework.aop.framework.ObjenesisCglibAopProxy.createProxyClassAndInstance(ObjenesisCglibAopProxy.java:57)
at org.springframework.aop.framework.CglibAopProxy.getProxy(CglibAopProxy.java:205)
at org.springframework.aop.framework.ProxyFactory.getProxy(ProxyFactory.java:110)
at org.springframework.context.annotation.ContextAnnotationAutowireCandidateResolver.buildLazyResolutionProxy(ContextAnnotationAutowireCandidateResolver.java:117)
at org.springframework.context.annotation.ContextAnnotationAutowireCandidateResolver.getLazyResolutionProxyIfNecessary(ContextAnnotationAutowireCandidateResolver.java:52)
at org.springframework.beans.factory.support.DefaultListableBeanFactory.resolveDependency(DefaultListableBeanFactory.java:1172)
at org.springframework.context.annotation.CommonAnnotationBeanPostProcessor.autowireResource(CommonAnnotationBeanPostProcessor.java:521)
at org.springframework.context.annotation.CommonAnnotationBeanPostProcessor.getResource(CommonAnnotationBeanPostProcessor.java:497)
at org.springframework.context.annotation.CommonAnnotationBeanPostProcessor$1.getTarget(CommonAnnotationBeanPostProcessor.java:461)
at org.springframework.aop.framework.CglibAopProxy$DynamicAdvisedInterceptor.intercept(CglibAopProxy.java:673)
at cn.howarliu.AsyncTaskProducer$$EnhancerBySpringCGLIB$$ebe8dcee.sendMessage(<generated>)
at cn.howarliu.service.impl.AsyncTaskServiceImpl.create(AsyncTaskServiceImpl.java:88)
at cn.howarliu.service.impl.AsyncTaskServiceImpl$$FastClassBySpringCGLIB$$e8965ed3_3.invoke(<generated>)
at org.springframework.cglib.proxy.MethodProxy.invoke(MethodProxy.java:218)
at org.springframework.aop.framework.CglibAopProxy$CglibMethodInvocation.invokeJoinpoint(CglibAopProxy.java:750)
at org.springframework.aop.framework.ReflectiveMethodInvocation.proceed(ReflectiveMethodInvocation.java:163)
at org.springframework.transaction.interceptor.TransactionInterceptor$$Lambda$3136/0x0000000801913040.proceedWithInvocation(Unknown Source)
at org.springframework.transaction.interceptor.TransactionAspectSupport.invokeWithinTransaction(TransactionAspectSupport.java:295)
at org.springframework.transaction.interceptor.TransactionInterceptor.invoke(TransactionInterceptor.java:98)
at org.springframework.aop.framework.ReflectiveMethodInvocation.proceed(ReflectiveMethodInvocation.java:186)
at org.springframework.aop.framework.CglibAopProxy$DynamicAdvisedInterceptor.intercept(CglibAopProxy.java:689)
at cn.howarliu.service.impl.AsyncTaskServiceImpl$$EnhancerBySpringCGLIB$$f9101846.create(<generated>)
at cn.howarliu.AsyncTaskSupporter.create(AsyncTaskSupporter.java:27)
at cn.howarliu.AsyncTaskSupporter.create(AsyncTaskSupporter.java:35)
分析問題
找代碼
順著線程棧找代碼,經(jīng)過的代碼如下:
- cn.howardliu.AsyncTaskSupporter
public static Long create(AsyncTaskCreateRequestDto request) {
return AsyncTaskSupporter.getBean(AsyncTaskService.class)
.create(request);
}
- cn.howardliu.service.impl.AsyncTaskServiceImpl
@Resource
@Lazy
private AsyncTaskProducer asyncTaskProducer;
@Override
@Transactional(rollbackFor = Exception.class)
public Long create(AsyncTaskCreateRequestDto request) {
// 其他邏輯……
asyncTaskProducer.sendMessage(dataContract);
// 其他邏輯……
}
- cn.howardliu.AsyncTaskKafkaProducer
@Override
@Transactional(rollbackFor = Exception.class)
public void sendMessage(AsyncTaskDataContract contract) {
// 這個方法的邏輯這里不展示了
}
分析原因
如果是有運(yùn)行時生成類的情況,一定是標(biāo)記了@Lazy注解的AsyncTaskProducer和@Transactional。
Spring會在運(yùn)行時查找并創(chuàng)建@Lazy標(biāo)記的bean,如果bean的方法用了@Transactional,會使用Cglib生成動態(tài)代理類。
生成動態(tài)代理類的核心代碼是org.springframework.aop.framework.CglibAopProxy,具體方法如下(可以重點(diǎn)關(guān)注createEnhancer()方法):
@Override
public Object getProxy(@Nullable ClassLoader classLoader) {
if (logger.isTraceEnabled()) {
logger.trace("Creating CGLIB proxy: " + this.advised.getTargetSource());
}
try {
Class<?> rootClass = this.advised.getTargetClass();
Assert.state(rootClass != null, "Target class must be available for creating a CGLIB proxy");
Class<?> proxySuperClass = rootClass;
if (ClassUtils.isCglibProxyClass(rootClass)) {
proxySuperClass = rootClass.getSuperclass();
Class<?>[] additionalInterfaces = rootClass.getInterfaces();
for (Class<?> additionalInterface : additionalInterfaces) {
this.advised.addInterface(additionalInterface);
}
}
// Validate the class, writing log messages as necessary.
validateClassIfNecessary(proxySuperClass, classLoader);
// Configure CGLIB Enhancer...
Enhancer enhancer = createEnhancer();
if (classLoader != null) {
enhancer.setClassLoader(classLoader);
if (classLoader instanceof SmartClassLoader &&
((SmartClassLoader) classLoader).isClassReloadable(proxySuperClass)) {
enhancer.setUseCache(false);
}
}
enhancer.setSuperclass(proxySuperClass);
enhancer.setInterfaces(AopProxyUtils.completeProxiedInterfaces(this.advised));
enhancer.setNamingPolicy(SpringNamingPolicy.INSTANCE);
enhancer.setStrategy(new ClassLoaderAwareUndeclaredThrowableStrategy(classLoader));
Callback[] callbacks = getCallbacks(rootClass);
Class<?>[] types = new Class<?>[callbacks.length];
for (int x = 0; x < types.length; x++) {
types[x] = callbacks[x].getClass();
}
// fixedInterceptorMap only populated at this point, after getCallbacks call above
enhancer.setCallbackFilter(new ProxyCallbackFilter(
this.advised.getConfigurationOnlyCopy(), this.fixedInterceptorMap, this.fixedInterceptorOffset));
enhancer.setCallbackTypes(types);
// Generate the proxy class and create a proxy instance.
return createProxyClassAndInstance(enhancer, callbacks);
}
catch (CodeGenerationException | IllegalArgumentException ex) {
throw new AopConfigException("Could not generate CGLIB subclass of " + this.advised.getTargetClass() +
": Common causes of this problem include using a final class or a non-visible class",
ex);
}
catch (Throwable ex) {
// TargetSource.getTarget() failed
throw new AopConfigException("Unexpected AOP exception", ex);
}
}
protected Enhancer createEnhancer() {
return new Enhancer();
}
看源碼不會有問題
從源碼分析,spring會判斷是否使用class緩存,如果是,不會重復(fù)創(chuàng)建,如果否,會重復(fù)創(chuàng)建。
默認(rèn)會使用緩存,只有兩處邏輯會設(shè)定不用緩存:
- 啟動參數(shù)設(shè)置了cglib.useCache是false;
- classloader用的是SmartClassLoader且類是可重加載的。
檢查第一種情況
先看第一個,從啟動參數(shù)看看有沒有這個變量,使用命令jinfo $pid | grep cglib,發(fā)現(xiàn)沒有相關(guān)參數(shù)。
檢查第二種情況
第二種情況屬于盲區(qū),不知道怎么查。
九年義務(wù)教育告訴我,不會的題先跳過。
檢查其他可能得情況
如果源碼沒有問題,那就是運(yùn)行時有問題,到線上看看運(yùn)行時加載的CglibAopProxy類和我們看到的類是不是一樣的。
借助arthas的jad命令:
有問題的機(jī)器createEnhancer方法如下圖:
沒問題的機(jī)器createEnhancer方法如下圖:
發(fā)現(xiàn)類果然不同,有問題機(jī)器的類主動設(shè)置不適用緩存了。
現(xiàn)象命中,但是為什么?
可能性1:jar包版本不一致
unzip解壓fat-jar,找到spring-aop的包,發(fā)現(xiàn)版本是相同的。
其實(shí)這一個已經(jīng)排除,但是還是不死心,萬一雖然版本相同,但是包內(nèi)容不同呢?
再次解壓spring-aop的包,找到CglibAopProxy.class,通過javap -c -v反編譯:
protected org.springframework.cglib.proxy.Enhancer createEnhancer();
descriptor: ()Lorg/springframework/cglib/proxy/Enhancer;
flags: (0x0004) ACC_PROTECTED
Code:
stack=2, locals=1, args_size=1
0: new #77 // class org/springframework/cglib/proxy/Enhancer
3: dup
4: invokespecial #78 // Method org/springframework/cglib/proxy/Enhancer."<init>":()V
7: areturn
LineNumberTable:
line 231: 0
LocalVariableTable:
Start Length Slot Name Signature
0 8 0 this Lorg/springframework/aop/framework/CglibAopProxy;
能夠看到,反編譯后的邏輯是,先調(diào)用了Enhancer的init方法,即構(gòu)造函數(shù),然后就調(diào)用了areturn指令返回了結(jié)果。
所以,無論是jar的版本,還是jar中的內(nèi)容,都是原來的return new Enhancer();。
可能性2:從別的jar包加載的CglibAopProxy
有問題的服務(wù)取到的類是別的包的,用arthas的sc命令看下:
[arthas@]$ sc -d org.springframework.aop.framework.CglibAopProxy
class-info org.springframework.aop.framework.CglibAopProxy
code-source file:/path/to/runner.jar!/BOOT-INF/lib/spring-aop-5.1.10.jar!/
name org.springframework.aop.framework.CglibAopProxy
isInterface false
isAnnotation false
isEnum false
isAnonymousClass false
isArray false
isLocalClass false
isMemberClass false
isPrimitive false
isSynthetic false
simple-name CglibAopProxy
modifier
annotation
interfaces org.springframework.aop.framework.AopProxy,java.io.Serializable
super-class +-java.lang.Object
class-loader +-org.springframework.boot.loader.LaunchedURLClassLoader@4aa21f9d
+-jdk.internal.loader.ClassLoaders$AppClassLoader@6a6824be
+-jdk.internal.loader.ClassLoaders$PlatformClassLoader@5ee0a0ae
classLoaderHash 4aa21f9d
都是 spring-aop-5.1.10.jar 中類。
可能性3:加載類時代碼被改了
其實(shí)排除了上面兩種可能,就只剩下這一種可能了。
有問題的服務(wù)使用了HotswapAgent,找到HotswapAgent的源碼分析,終于找到了SpringPlugin類,其中有一段改寫createEnhancer方法的代碼:
@OnClassLoadEvent(classNameRegexp = "org.springframework.aop.framework.CglibAopProxy")
public static void cglibAopProxyDisableCache(CtClass ctClass) throws NotFoundException, CannotCompileException {
CtMethod method = ctClass.getDeclaredMethod("createEnhancer");
method.setBody("{org.springframework.cglib.proxy.Enhancer enhancer = new org.springframework.cglib.proxy.Enhancer();enhancer.setUseCache(false);return enhancer;}");
LOGGER.debug("org.springframework.aop.framework.CglibAopProxy - cglib Enhancer cache disabled", new Object[0]);
}
找到這里,基本上可以確定就是這個插件的問題了。
解決問題
很多時候,提出問題、找到問題、分析問題是難點(diǎn),解決問題反而是比較簡答的。比如上面的問題,我們有兩種解法:
- 方法一:既然是運(yùn)行時加載bean出現(xiàn)了重復(fù)創(chuàng)建Cglib代理類的問題,那就直接刪除@Lazy注解,啟動服務(wù)時創(chuàng)建并注入bean,運(yùn)行時就不存在創(chuàng)建bean的問題了;
- 方法二:刪除有問題的組件,等不再壓測的時候再加回來。
實(shí)際工作中也是,我是先把@Lazy注解去掉了,然后又耐心找問題。
這算是一個技巧,先解問題,不要成為阻塞點(diǎn),然后找到問題原因,提升能力。換句話就是,先做該做的,再做想做的。
文末總結(jié)
HotswapAgent在Spring啟動時改寫了CglibAopProxy的createEnhancer()方法,使用Cglib生成代理類時,不會使用緩存class,會重新生成class。 在高并發(fā)場景中,請求中如果需要生成代理類,會重復(fù)生成class,造成Metaspace溢出,出現(xiàn)OOM。