Skip to content

Commit

Permalink
将synchonized和AtomicLong直接改为无锁，牺牲一些统计数据准确性带来性能提升和减低内存消耗
Browse files Browse the repository at this point in the history
  • Loading branch information
丁建水 committed Aug 25, 2021
1 parent fc2e2bb commit 0746419
Show file tree
Hide file tree
Showing 2 changed files with 28 additions and 36 deletions.
50 changes: 21 additions & 29 deletions src/com/wenshuo/agent/log/ExecuteLogUtils.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,6 @@
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ScheduledThreadPoolExecutor;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicLong;

import com.wenshuo.agent.AgentUtils;
import com.wenshuo.agent.ConfigUtils;
Expand All @@ -44,7 +43,7 @@ public class ExecuteLogUtils {

private static ScheduledThreadPoolExecutor counterLogExecutor;

private static Map<String, Map<String, AtomicLong[]>> executeCounterMap;
private static ConcurrentHashMap<String, ConcurrentHashMap<String, long[]>> executeCounterMap;

private static final Object executeCounterLock = new Object();

Expand Down Expand Up @@ -82,7 +81,7 @@ public static synchronized void init() {
}
setNextDateStartTimeMillis();
initWriter();
executeCounterMap = new ConcurrentHashMap<String, Map<String, AtomicLong[]>>();
executeCounterMap = new ConcurrentHashMap<String, ConcurrentHashMap<String, long[]>>();
startTimeMillis = System.currentTimeMillis();
counterLogExecutor = new ScheduledThreadPoolExecutor(1, new NamedThreadFactory("pool-thread-agent-log", true));
counterLogExecutor.scheduleWithFixedDelay(new OutputLogRunnable(), interval, interval, TimeUnit.SECONDS);
Expand All @@ -99,26 +98,26 @@ public static void log(String className, String methodName, long executeTime) {
* @author dingjsh
* @time 2015-7-28下午01:35:25
*/
synchronized static void outputCounterLog() throws IOException {
Map<String, Map<String, AtomicLong[]>> executeCounterMapInner = executeCounterMap;
executeCounterMap = new ConcurrentHashMap<String, Map<String, AtomicLong[]>>();
synchronized static void outputCounterLog() throws IOException {
ConcurrentHashMap<String, ConcurrentHashMap<String, long[]>> executeCounterMapInner = executeCounterMap;
executeCounterMap = new ConcurrentHashMap<String, ConcurrentHashMap<String, long[]>>();
String startTime = formatTimeMillis(startTimeMillis);
String endTime = formatTimeMillis(System.currentTimeMillis());
long byteLength = removeJSONArrayEndBracket();
if (0 == byteLength) {
writeLog("[",true, startTimeMillis);
writeLog("[", true, startTimeMillis);
}

Set<Map.Entry<String, Map<String, AtomicLong[]>>> entrySet = executeCounterMapInner.entrySet();
Iterator<Map.Entry<String, Map<String, AtomicLong[]>>> ite = entrySet.iterator();
Set<Map.Entry<String, ConcurrentHashMap<String, long[]>>> entrySet = executeCounterMapInner.entrySet();
Iterator<Map.Entry<String, ConcurrentHashMap<String, long[]>>> ite = entrySet.iterator();
int length = entrySet.size();
if (length > 0 && byteLength > 10) { // 说明文件不只是[],json数组中已经有内容
writeLog(",", startTimeMillis);
}
for (int index = 0; ite.hasNext(); index++) {
Map.Entry<String, Map<String, AtomicLong[]>> entry = ite.next();
Map.Entry<String, ConcurrentHashMap<String, long[]>> entry = ite.next();
String className = entry.getKey();
Map<String, AtomicLong[]> method2ExecuteMap = entry.getValue();
Map<String, long[]> method2ExecuteMap = entry.getValue();
String methodExecuteJson = MethodExecuteJSONformatter.getMethodExecuteJSON(className, method2ExecuteMap,
startTime, endTime, isUsingNanoTime, logAvgExecuteTime);
writeLog(methodExecuteJson, startTimeMillis);
Expand All @@ -132,42 +131,35 @@ synchronized static void outputCounterLog() throws IOException {
}

private static void logExecuteCounter(String className, String methodName, long executeTime) {
Map<String, AtomicLong[]> methodCounterMap = getOrCreateClassExecutesMapping(className);
AtomicLong[] counter = methodCounterMap.get(methodName);
// dingjs modfied in 20210825 原来是用锁和Atomiclong来避免线程安全问题，现在去掉这些实现。原因是此处线程安全问题顶多导致
// 数据有一些误差，这个误差对于agent数据监控并没有什么影响，牺牲一定的准确性带来性能的提升是有必要的
ConcurrentHashMap<String, long[]> methodCounterMap = getOrCreateClassExecutesMapping(className);
long[] counter = methodCounterMap.get(methodName);
if (null == counter) {
synchronized (methodCounterMap) {
counter = methodCounterMap.get(methodName);
if (null == counter) {
methodCounterMap.put(methodName, new AtomicLong[] {new AtomicLong(1), new AtomicLong(executeTime)});
} else {
counter[0].incrementAndGet();
counter[1].addAndGet(executeTime);
}
}
methodCounterMap.put(methodName, new long[]{1, executeTime});
} else {
counter[0].incrementAndGet();
counter[1].addAndGet(executeTime);
counter[0]++;
counter[1] = executeTime + counter[1];
}
}

/**
*
* ExecuteLogUtils
*
* @description 获得class和它的调用次数映射关系，如果exeuteCounterMap中没有，则创建一个并放入
* @param className 类名
* @return class和它的调用次数映射关系
* @description 获得class和它的调用次数映射关系，如果exeuteCounterMap中没有，则创建一个并放入
* @author dingjsh
* @date 2018年5月25日 下午4:40:33
* @version 1.2.0
*/
private static Map<String, AtomicLong[]> getOrCreateClassExecutesMapping(String className) {
Map<String, AtomicLong[]> methodCounterMap = executeCounterMap.get(className);
private static ConcurrentHashMap<String, long[]> getOrCreateClassExecutesMapping(String className) {
ConcurrentHashMap<String, long[]> methodCounterMap = executeCounterMap.get(className);
if (null == methodCounterMap) {
synchronized (executeCounterLock) {
methodCounterMap = executeCounterMap.get(className);
if (null == methodCounterMap) {
methodCounterMap = new ConcurrentHashMap<String, AtomicLong[]>();
methodCounterMap = new ConcurrentHashMap<String, long[]>();
executeCounterMap.put(className, methodCounterMap);
}
}
Expand Down
14 changes: 7 additions & 7 deletions src/com/wenshuo/agent/log/MethodExecuteJSONformatter.java
View file
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
package com.wenshuo.agent.log;

import java.util.Map;
import java.util.concurrent.atomic.AtomicLong;


/**
* MethodExecuteJSONformatter
Expand All @@ -17,7 +17,7 @@ class MethodExecuteJSONformatter {
* MethodExecuteJSONformatter
*
* @param className 类名
* @param method2ExecuteMap Map<方法名, AtomicLong[]> 数组第一个是执行次数，第二个是执行时间
* @param method2ExecuteMap Map<方法名, Long[]> 数组第一个是执行次数，第二个是执行时间
* @param startTime 统计周期的开始时间，格式为 yyyy-MM-dd HH:mm:ss
* @param endTime 统计周期的结束时间，格式为 yyyy-MM-dd HH:mm:ss
* @return 格式化后的json string
Expand All @@ -26,19 +26,19 @@ class MethodExecuteJSONformatter {
* @date 2018年8月22日 下午8:49:18
* @version 2.0.0
*/
static String getMethodExecuteJSON(String className, Map<String, AtomicLong[]> method2ExecuteMap,
static String getMethodExecuteJSON(String className, Map<String, long[]> method2ExecuteMap,
String startTime, String endTime, boolean isUsingNanoTime, boolean logAvgExecuteTime) {
StringBuilder json = new StringBuilder("{");
appendString(json, "class", className).append(",");
appendString(json, "start", startTime).append(",");
appendString(json, "end", endTime).append(",");
appendKey(json, "methods").append("[");
for (Map.Entry<String, AtomicLong[]> methodEntry : method2ExecuteMap.entrySet()) {
for (Map.Entry<String, long[]> methodEntry : method2ExecuteMap.entrySet()) {
String methodName = methodEntry.getKey();
AtomicLong[] executeCounter = methodEntry.getValue();
long counter = executeCounter[0].longValue();
long[] executeCounter = methodEntry.getValue();
long counter = executeCounter[0];
long timeInMillis
= isUsingNanoTime ? executeCounter[1].longValue() / 1000000 : executeCounter[1].longValue();
= isUsingNanoTime ? executeCounter[1] / 1000000 : executeCounter[1];
json.append("{");
appendString(json, "name", methodName).append(",");
appendLong(json, "counter", counter).append(",");
Expand Down

0 comments on commit 0746419

Please sign in to comment.