[BUG]关于SpringValueRegistry.scanAndClean方法报并发修改异常的疑问 #5296
Description
- I have checked the discussions
- I have searched the issues of this repository and believe that this is not a duplicate.
- I have checked the FAQ of this repository and believe that this is not a duplicate.
Describe the bug
最近使用旧版apollo时遇到prototype-bean导致内存被打满的问题，查找issue找到#1670
在查看相关方法时，找到了遇到并发问题并解决的：#2281
但是查看SpringValueRegistry类的issue还是发现了：#4355、#4356、#4800、#4925、#5169
问题主要集中于scanAndClean还是有报并发错误，我查看synchronizedListMultimap方法的注释时，发现有相关叙述：
/**
* Returns a synchronized (thread-safe) multimap backed by the specified multimap. In order to
* guarantee serial access, it is critical that <b>all</b> access to the backing multimap is
* accomplished through the returned multimap.
*
* <p>It is imperative that the user manually synchronize on the returned multimap when accessing
* any of its collection views:
*
* <pre>{@code
* Multimap<K, V> multimap = Multimaps.synchronizedMultimap(
* HashMultimap.<K, V>create());
* ...
* Collection<V> values = multimap.get(key); // Needn't be in synchronized block
* ...
* synchronized (multimap) { // Synchronizing on multimap, not values!
* Iterator<V> i = values.iterator(); // Must be in synchronized block
* while (i.hasNext()) {
* foo(i.next());
* }
* }
* }</pre>
*
* <p>Failure to follow this advice may result in non-deterministic behavior.
*
* <p>Note that the generated multimap's {@link Multimap#removeAll} and {@link
* Multimap#replaceValues} methods return collections that aren't synchronized.
*
* <p>The returned multimap will be serializable if the specified multimap is serializable.
*
* @param multimap the multimap to be wrapped in a synchronized view
* @return a synchronized view of the specified multimap
*/
其中It is imperative that the user manually synchronize on the returned multimap when accessing any of its collection views:
Failure to follow this advice may result in non-deterministic behavior.
是否说明synchronizedMultimap的iterator实际上是非线程安全的，于是我就尝试自己写了一个简单的复现代码，并获得了其他issue中提到的错误
To Reproduce
复现这个synchronizedMultimap并发写入并定期清理导致报错的场景，我复现出其中一种，即输入固定量的k-v，定时清除，在写入停止后，因为剩余值(registry.entries().size())为负,在获取iterator时会报错：negative size
@Test
public void testConfig() {
Multimap<String, String> registry = Multimaps.synchronizedListMultimap(LinkedListMultimap.create());
ExecutorService executorService = Executors.newFixedThreadPool(10);
ScheduledExecutorService scheduledExecutorService = Executors.newSingleThreadScheduledExecutor();
AtomicInteger removeTotal = new AtomicInteger();
for(int i=0;i<10;i++){
executorService.submit(()->{
try {
for(int j = 0;j<100000;j++){
Thread.sleep(1);
registry.put("aa",System.currentTimeMillis()+"");
}
System.out.println("done");
}catch (Exception e){
e.printStackTrace();
}
});
}
scheduledExecutorService.scheduleWithFixedDelay(()->{
try {
long start = System.currentTimeMillis();
Iterator<Map.Entry<String, String>> ss = registry.entries().iterator();
while (ss.hasNext()) {
Map.Entry<String, String> entry = ss.next();
if (!StringUtils.isNullOrEmpty(entry.getValue())) {
ss.remove();
removeTotal.incrementAndGet();
}
}
System.out.println("cost:"+(System.currentTimeMillis()-start)+"ms");
System.out.println("removed total: "+removeTotal.get());
System.out.println("remain: "+registry.entries().size());
}catch (Exception e){
System.out.println(e.getMessage());
e.printStackTrace();
}
},5,5,TimeUnit.SECONDS);
try {
Thread.sleep(1000000000);
} catch (InterruptedException e) {
e.printStackTrace();
}
}
Additional Details & Logs
当我为清理代码增加synchronized后该问题解决：
synchronized (registry) {
Iterator<Map.Entry<String, String>> ss = registry.entries().iterator();
while (ss.hasNext()) {
Map.Entry<String, String> entry = ss.next();
if (!StringUtils.isNullOrEmpty(entry.getValue())) {
ss.remove();
removeTotal.incrementAndGet();
}
}
}