最近发现线上监控有个SQL调用量很大,但是方法的调用量不是很大,查看接口实现,发现接口是做了缓存操作的,使用Spring cache缓存注解结合tair实现缓存操作。但是为啥SQL调用量这么大,难道缓存没有生效。测试发现缓存是正常的,分析了代码发现,代码存在缓存穿透的风险。具体注解是这样的:
@Cacheable(value = "storeDeliveryCoverage", key = "#sellerId + '|' + #cityCode", unless = "#result == null")
unless = "#result == null"表明接口返回值不为空的时候才缓存,如果线上有大量不合法的请求参数过来,由于为空的不会缓存起来,每次请求都打到DB上,导致DB的sql调用量巨大,给了黑客可乘之机,风险还是很大的。找到原因之后就修改,查询结果为空的时候兜底一个null,把这句unless = "#result == null"条件去掉测试了一下,发现为空的话还是不会缓存。于是debug分析了一波源码,终于发现原来是tair的问题。
由于tair自身的特性,无法缓存null。既然无法缓存null,那我们就兜底一个空对象进去,取出来的时候把空对象转化为null。基于这个思路,我们把Cache的实现改造了一下:
@Override
public void put(Object key, Object value) {
if (value == null) {
// 为空的话,兜底一个空对象,防止缓存穿透(由于tair自身特性不允许缓存null对象的原因,这里缓存一个空对象)
value = new Nil();
}
if (value instanceof Serializable) {
final String tairKey = String.format("%s:%s", this.name, key);
final ResultCode resultCode = this.tairManager.put(
this.namespace,
tairKey,
(Serializable) value,
0,
this.timeout
);
if (resultCode != ResultCode.SUCCESS) {
TairSpringCache.log.error(
String.format(
"[CachePut]: unable to put %s => %s into tair due to: %s",
key,
value,
resultCode.getMessage()
)
);
}
} else {
throw new RuntimeException(
String.format(
"[CachePut]: value %s is not Serializable",
value
)
);
}
}
Nil类默认是一个空对象,这里给了个内部类:
static class Nil implements Serializable {
private static final long serialVersionUID = -9138993336039047508L;
}
取缓存的get方法实现:
@Override
public ValueWrapper get(Object key) {
final String tairKey = String.format("%s:%s", this.name, key);
final Result result = this.tairManager.get(this.namespace, tairKey);
if (result.isSuccess() && (result.getRc() == ResultCode.SUCCESS)) {
final Object obj = result.getValue().getValue();
// 缓存为空兜底的是Nil对象,这里返回的时候需要转为null
if (obj instanceof Nil) {
return null;
}
return () -> obj;
}
return null;
}
改好了之后,测试一下,结果发现还是没有生效,缓存没有兜底,请求都打到DB上了。debug走一遍,看了下Cache的源码,终于发现关键问题所在(具体实现流程参考上一篇:Spring Cache- 缓存拦截器( CacheInterceptor)):
private Object execute(final CacheOperationInvoker invoker, Method method, CacheOperationContexts contexts) {
// Special handling of synchronized invocation
if (contexts.isSynchronized()) {
CacheOperationContext context = contexts.get(CacheableOperation.class).iterator().next();
if (isConditionPassing(context, CacheOperationExpressionEvaluator.NO_RESULT)) {
Object key = generateKey(context, CacheOperationExpressionEvaluator.NO_RESULT);
Cache cache = context.getCaches().iterator().next();
try {
return wrapCacheValue(method, cache.get(key, new Callable
根据key从缓存中查找,返回的结果是ValueWrapper,它是返回结果的包装器:
private Cache.ValueWrapper findCachedItem(Collection contexts) {
Object result = CacheOperationExpressionEvaluator.NO_RESULT;
for (CacheOperationContext context : contexts) {
if (isConditionPassing(context, result)) {
Object key = generateKey(context, result);
Cache.ValueWrapper cached = findInCaches(context, key);
if (cached != null) {
return cached;
}
else {
if (logger.isTraceEnabled()) {
logger.trace("No cache entry for key '" + key + "' in cache(s) " + context.getCacheNames());
}
}
}
}
return null;
}
private Cache.ValueWrapper findInCaches(CacheOperationContext context, Object key) {
for (Cache cache : context.getCaches()) {
Cache.ValueWrapper wrapper = doGet(cache, key);
if (wrapper != null) {
if (logger.isTraceEnabled()) {
logger.trace("Cache entry for key '" + key + "' found in cache '" + cache.getName() + "'");
}
return wrapper;
}
}
return null;
}
这里判断缓存是否命中的逻辑是根据cacheHit是否为空,而cacheHit是ValueWrapper类型,查看ValueWrapper是一个接口,它的实现类是SimpleValueWrapper,这是一个包装器,将缓存的结果包装起来了。而我们前面的get方法取缓存的时候如果为Nil对象,返回的是null,这样缓存判断出来是没有命中,即cacheHit==null,就会去执行具体方法朔源。所以到这里已经很清晰了,关键问题是get取缓存的结果如果是兜底的Nil对象,应该返回new SimpleValueWrapper(null)。应该返回包装器,包装的是缓存的对象为null。测试了一下,发现ok了。具体源码如下:
/**
* 基于tair的缓存,适配spring缓存框架
*/
public class TairSpringCache implements Cache {
private static final Logger log = LoggerFactory.getLogger(TairSpringCache.class);
private TairManager tairManager;
private final String name;
private int namespace;
private int timeout;
public TairSpringCache(String name, TairManager tairManager, int namespace) {
this(name, tairManager, namespace, 0);
}
public TairSpringCache(String name, TairManager tairManager, int namespace, int timeout) {
this.name = name;
this.tairManager = tairManager;
this.namespace = namespace;
this.timeout = timeout;
}
@Override
public String getName() {
return this.name;
}
@Override
public Object getNativeCache() {
return this.tairManager;
}
@Override
public ValueWrapper get(Object key) {
final String tairKey = String.format("%s:%s", this.name, key);
final Result result = this.tairManager.get(this.namespace, tairKey);
if (result.isSuccess() && (result.getRc() == ResultCode.SUCCESS)) {
final Object obj = result.getValue().getValue();
// 缓存为空兜底的是Nil对象,这里返回的时候需要转为null
if (obj instanceof Nil) {
return () -> null;
}
return () -> obj;
}
return null;
}
@Override
public T get(Object key, Class type) {
return (T) this.get(key).get();
}
public T get(Object o, Callable callable) {
return null;
}
@Override
public void put(Object key, Object value) {
if (value == null) {
// 为空的话,兜底一个空对象,防止缓存穿透(由于tair自身特性不允许缓存null对象的原因,这里缓存一个空对象)
value = new Nil();
}
if (value instanceof Serializable) {
final String tairKey = String.format("%s:%s", this.name, key);
final ResultCode resultCode = this.tairManager.put(
this.namespace,
tairKey,
(Serializable) value,
0,
this.timeout
);
if (resultCode != ResultCode.SUCCESS) {
TairSpringCache.log.error(
String.format(
"[CachePut]: unable to put %s => %s into tair due to: %s",
key,
value,
resultCode.getMessage()
)
);
}
} else {
throw new RuntimeException(
String.format(
"[CachePut]: value %s is not Serializable",
value
)
);
}
}
public ValueWrapper putIfAbsent(Object key, Object value) {
final ValueWrapper vw = this.get(key);
if (vw.get() == null) {
this.put(key, value);
}
return vw;
}
@Override
public void evict(Object key) {
final String tairKey = String.format("%s:%s", this.name, key);
final ResultCode resultCode = this.tairManager.delete(this.namespace, tairKey);
if ((resultCode == ResultCode.SUCCESS)
|| (resultCode == ResultCode.DATANOTEXSITS)
|| (resultCode == ResultCode.DATAEXPIRED)) {
return;
}
else {
final String errMsg = String.format(
"[CacheDelete]: unable to evict key %s, resultCode: %s",
key,
resultCode
);
TairSpringCache.log.error(errMsg);
throw new RuntimeException(errMsg);
}
}
@Override
public void clear() {
//TODO fgz: implement here later
}
public void setTairManager(TairManager tairManager) {
this.tairManager = tairManager;
}
public void setNamespace(int namespace) {
this.namespace = namespace;
}
public void setTimeout(int timeout) {
this.timeout = timeout;
}
static class Nil implements Serializable {
private static final long serialVersionUID = -9138993336039047508L;
}
}
测试用例就不贴了。