想根据redis keyPattern进行统计，以评估每个key的访问量和rt，用来获取热key等信息，以便进行优化
lettuce自带metrics，但只统计到command的纬度，没有具体的key tag
- 按照keyPattern统计
- tag至少含command,key,exception
- 尽可能的无侵入式
例如keyPattern = user:%s，具体的key为user:1、user:2 ....，如果直接将具体的key统计成tag，会导致tag数量异常的多，将会引发outOfMemory，所以要按照keyPattern进行聚合
- RedisKeyPatternParse: 用于将原始key解析成keyPattern
- JdkRegexpRedisKeyPatternParse: Jdk正则转换
示例：
定义key pattern常量类
public class RedisKeyConstantTest {
public static String USER_KEY = "user:%s";
}
初始化RedisKeyPatternParse，解析原始key对应的keyPattern
protected RedisKeyPatternParse redisPatternParse = new JdkRegexpRedisKeyPatternParse(RedisKeyConstantTest.class);
@Test
void exist_parse() {
Optional<String> keyPattern = redisPatternParse.parse("user:11");
assertEquals("user:.*", keyPattern.orElse(null));
}
现有项目都是通过redisTemplate操作redis
redisTemplate.opsForValue().set("user:1","aaaaaa")
redisTemplate源码
public class RedisTemplate<K, V> extends RedisAccessor implements RedisOperations<K, V>, BeanClassLoaderAware {
private final ValueOperations<K, V> valueOps = new DefaultValueOperations<>(this);
@Override
public ValueOperations<K, V> opsForValue() {
return valueOps;
}
思考： 想无侵入式的对redisTemplate进行增强，可以生成ValueOperations的代理对象，redisTemplate.opsForValue()返回代理对象， 但valueOps被private final修饰，无法替换
所以换个思路：
- 创建redisTemplate代理对象，拦截opsForValue方法，返回代理的ValueOperations
- valueOperations也是代理对象，方法执行前后会调用redisMetrics进行统计
- redisMetrics会调用RedisKeyPatternParse解析keyPattern
定义RedisTemplateAspect，拦截opsFor方法，创建对应对代理对象
@Aspect
public class RedisTemplateAspect {
...
@Pointcut("execution(public * org.springframework.data.redis.core.RedisTemplate.opsFor*())")
public void operationPointCut() {
}
@Around("operationPointCut()")
public Object aroundOpsFor(ProceedingJoinPoint joinPoint) throws Throwable {
Object result = joinPoint.proceed();
if (needProxy(result)) {
return operationProxyCache.computeIfAbsent(result, (re) -> {
ProxyFactory proxyFactory = new ProxyFactory(result);
proxyFactory.addAdvice(redisOperationMethodInterceptor);
return proxyFactory.getProxy();
});
}
return result;
}
...
创建ValueOperations的代理，ValueOperations执行前后会通过redisMetrics进行统计
public class RedisOperationMethodInterceptor implements MethodInterceptor {
private static final Logger LOG = LoggerFactory.getLogger(RedisOperationMethodInterceptor.class);
private final RedisMetrics redisMetrics;
public RedisOperationMethodInterceptor(RedisMetrics redisMetrics) {
this.redisMetrics = redisMetrics;
}
@Override
public Object invoke(MethodInvocation invocation) throws Throwable {
Object[] arguments = invocation.getArguments();
boolean needMetrics = arguments.length > 0 && arguments[0] instanceof String;
if (!needMetrics) {
return invocation.proceed();
}
long start = System.currentTimeMillis();
Object result;
String key = (String) arguments[0];
String command = invocation.getMethod().getName();
try {
result = invocation.proceed();
} catch (Exception e) {
LOG.warn("redis execute error", e);
redisMetrics.metrics(key, command, System.currentTimeMillis() - start, e);
throw e;
}
redisMetrics.metrics(key, command, System.currentTimeMillis() - start, null);
return result;
}
}
RedisMetrics通过redisPatternParse提取keyPattern，进行统计
public class RedisMetrics {
...
public void metrics(String originalKey, String command, long cost, Exception e) {
Optional<String> parseResultOptional = redisPatternParse.parse(originalKey);
if (parseResultOptional.isPresent()) {
doMetrics(parseResultOptional.get(), command, cost, e);
} else {
doMetrics("UNKNOWN", command, cost, e);
LOG.debug(
"not found key, originalKey:{}, redisPatternParse:{}",
originalKey,
redisPatternParse.getClass().getSimpleName());
}
}
redis_seconds_count{command="set",exception="None",key="user:.*",} 1.0
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent