对于一些专业的 Java 程序员来说,这可能是一个愚蠢的问题,但我现在要疯了,所以我还在问。请指导我正确的方向有人。
如何分析哪个方法/partOfMethod 在我的 java 程序中消耗更多时间?
(我正在使用 Eclipse 和 Junit)
使用 jvisualvm。它现在与 JDK 捆绑在一起,但也有一个独立版本,它是最新的。通常,当您启动它时,您可以选择要连接到哪个正在运行的 java 进程(这很可能是我们正在运行的单元测试)。您可以指定需要跟踪的合格类名过滤器。通常,会出现一些类检测,您将能够跟踪分配给每个方法的处理时间(以及累积时间)。
您需要获取Java 分析器。其中一些在 Eclipse 中集成得很好。
除此之外,您还可以使用自定义分析器类(如果您没有太多的分析工作要做并且您已经怀疑存在一些瓶颈)。
这是一个简单的类来做到这一点:
/**
* Small helper class to profile the code, take timing, ...
*
* To use this, simply call the start method with an identifier. When you want to measure the time, call the stop method
* with the same identifier. To output statistics, simply call the toString method or the toCsv method to create a CSV
* file with the profiler information.
*
* @author Vincent Prat @ MarvinLabs
*/
public class Profiler {
private static final int THEORETICAL_MAX_NAME_LENGTH = 50;
private static Profiler singletonInstance = null;
private Map<String, Profile> profiles; // Fast access to profiles by name
private List<Profile> profilesStack; // Profiles as created chronologically
/**
* Get access to the singleton instance (create it if necessary)
*/
public static Profiler getInstance() {
if (singletonInstance == null) {
singletonInstance = new Profiler();
}
return singletonInstance;
}
/**
* Protected constructor for singleton
*/
protected Profiler() {
profiles = new HashMap<String, Profiler.Profile>();
profilesStack = new ArrayList<Profile>();
}
/**
* Start a profile. If the profile does not exist, it will be created. If it exists, a new round of measure is
* taken.
*
* @param name
* The name of the profile. If possible, less than Profiler.THEORETICAL_MAX_NAME_LENGTH characters
*
* @see Profiler.THEORETICAL_MAX_NAME_LENGTH
*/
public void start(String name) {
Profile p = profiles.get(name);
if (p == null) {
p = new Profile(name);
profiles.put(name, p);
profilesStack.add(p);
}
p.start();
}
/**
* Stop a profile and compute some statistics about it.
*
* @param name
* The name of the profile as declared in the corresponding start method
*/
public void stop(String name) {
Profile p = profiles.get(name);
if (p == null) {
throw new RuntimeException("The profile " + name + " has not been created by a call to the start() method!");
}
p.stop();
}
/**
* Clear all the current measures. Not to be called within any start/stop pair.
*/
public void reset() {
profiles.clear();
}
/**
* Build a string containing all the information about the measures we have taken so far.
*/
@Override
public String toString() {
final StringBuffer sb = new StringBuffer();
for (Profile p : profilesStack) {
sb.append(p.toString());
sb.append("\n");
}
return sb.toString();
}
/**
* Output the measures to an output string
*/
public void toCsvFile(OutputStream os) throws IOException {
Profile.writeCsvHeader(os);
for (Profile p : profilesStack) {
p.writeCsvLine(os);
}
}
/**
* Profile information. It stores statistics per named profile.
*
* @author Vincent Prat @ MarvinLabs
*/
private static class Profile {
private static final String CSV_HEADERS = "Name, Call Count, Total Time (ms), Average Time (ms), Min Time (ms), Max Time (ms), Delta Time (ms), Delta Ratio (%)\n";
private static final String FORMAT_STRING = "%-" + THEORETICAL_MAX_NAME_LENGTH + "."
+ THEORETICAL_MAX_NAME_LENGTH
+ "s: %3d calls, total %5d ms, avg %5d ms, min %5d ms, max %5d ms, delta %5d ms (%d%%)";
private static final String CSV_FORMAT_STRING = "%s,%d,%d,%d,%d,%d,%d,%d\n";
private String name;
private long startTime;
private long callCount;
private long totalTime;
private long minTime;
private long maxTime;
public Profile(String name) {
this.name = name;
this.callCount = 0;
this.totalTime = 0;
this.startTime = 0;
this.minTime = Long.MAX_VALUE;
this.maxTime = Long.MIN_VALUE;
}
public void start() {
startTime = System.currentTimeMillis();
}
public void stop() {
final long elapsed = (System.currentTimeMillis() - startTime);
if (elapsed < minTime) minTime = elapsed;
if (elapsed > maxTime) maxTime = elapsed;
totalTime += elapsed;
callCount++;
}
private String getFormattedStats(String format) {
final long avgTime = callCount == 0 ? 0 : (long) totalTime / callCount;
final long delta = maxTime - minTime;
final double deltaRatio = avgTime == 0 ? 0 : 100.0 * ((double) 0.5 * delta / (double) avgTime);
return String
.format(format, name, callCount, totalTime, avgTime, minTime, maxTime, delta, (int) deltaRatio);
}
@Override
public String toString() {
return getFormattedStats(FORMAT_STRING);
}
public static void writeCsvHeader(OutputStream os) throws IOException {
os.write(CSV_HEADERS.getBytes());
}
public void writeCsvLine(OutputStream os) throws IOException {
os.write(getFormattedStats(CSV_FORMAT_STRING).getBytes());
}
}
}
和示例用法:
Profiler.getInstance().start("marker1");
// Do something...
Profiler.getInstance().start("marker2");
// Something else...
Profiler.getInstance().stop("marker2");
// And some more...
Profiler.getInstance().stop("marker1");
// Output the profiling result
System.out.println(Profiler.getInstance().toString());
你可以看看Yourkit(商业软件),它可以监控内存、CPU等等。它有特殊的视图来显示方法中花费了多少时间(例如,您可以看到 40% 的执行时间花费在方法中xyz()
)。
称我为老式,但这是我认为最简单的方法:
long a, b, c, d;
a = System.currentTimeMillis();
// some code 1
b = System.currentTimeMillis();
// some code 2
c = System.currentTimeMillis();
// some code 3
d = System.currentTimeMillis();
System.out.println("Some code 1 took "+(b-a)+"mil to execute. ("+((b-a)/1000)+" seconds)");
System.out.println("Some code 2 took "+(c-b)+"mil to execute. ("+((c-b)/1000)+" seconds)");
System.out.println("Some code 3 took "+(d-c)+"mil to execute. ("+((d-c)/1000)+" seconds)");
希望这可以帮助 :)
测量程序的哪一部分需要多少运行时间的过程称为“分析”。
eclipse 有很多分析插件。一个在这里描述。
有一些像IdeOne这样的在线工具可以让你花时间执行代码块。试试看!