arthas/arthas-tutorials-en/command/trace/trace.md

Trace method calling path, and output the time cost for each node in the path.trace

trace{{}} can track the calling path specified by class-pattern{{}} / method-pattern{{}} , and calculate the time cost on the whole path.

Parameters

Name	Specification
class-pattern	pattern for the class name
method-pattern	pattern for the method name
condition-express	condition expression
[E]{{}}	enable regex match, the default behavior is wildcards match
[n:]{{}}	execution times
#cost	time cost

trace command Docs

There's one thing worthy noting here is observation expression. The observation expression supports OGNL grammar, for example, you can come up a expression like this "{params,returnObj}"{{}} . All OGNL expressions are supported as long as they are legal to the grammar.

Thanks for advice{{}} 's data structure, it is possible to observe from varieties of different angles. Inside advice{{}} parameter, all necessary information for notification can be found.

Pls. refer to core parameters in expression for more details.

• Pls. also refer to https://github.com/alibaba/arthas/issues/71 for more advanced usage
• OGNL official site: https://commons.apache.org/proper/commons-ognl/language-guide.html

Many times what we are interested is the exact trace result when the method call takes time over one particular period. It is possible to achieve this in Arthas, for example: trace *StringUtils isBlank '#cost>100'{{}} means trace result will only be output when the executing time exceeds 100ms.

watch{{}} /stack{{}} /trace{{}} , these three commands all support #cost{{}} .

Notice

trace{{}} is handy to help discovering and locating the performance flaws in your system, but pls. note Arthas can only trace the first level method call each time.

After version 3.3.0, you can use the Dynamic Trace feature to add new matching classes/methods, see the following example.

Usage

Trace method

trace demo.MathGame run{{execute T2}}

Press Q{{exec interrupt}} or Ctrl+C{{exec interrupt}} to abort

Trace times limit

If the method invoked many times, use -n{{}} options to specify trace times. For example, the command will exit when received a trace result.

trace demo.MathGame run -n 1{{execute T2}}

Include jdk method

• --skipJDKMethod <value> {{}} skip jdk method trace, default value true.

trace --skipJDKMethod false demo.MathGame run{{execute T2}}

Press Q{{execute T2}} or Ctrl+C{{}} to abort

Filtering by cost

trace demo.MathGame run '#cost > 10'{{execute T2}}

Press Q{{execute T2}} or Ctrl+C{{}} to abort

Only the call path which's time cost is higher than 10ms{{}} will be shown. This feature is handy to focus on what's needed to focus when troubleshoot.

• Here Arthas provides the similar functionality JProfile and other commercial software provide. Compared to these professional softwares, Arthas doesn't deduce the time cost trace{{}} itself takes, therefore it is not as accurate as these softwares offer. More classes and methods on the calling path, more inaccurate trace{{}} output is, but it is still helpful for diagnostics where the bottleneck is.
• "[12.033735ms]" means the method on the node takes 12.033735{{}} ms.
• "[min=0.005428ms,max=0.094064ms,total=0.105228ms,count=3] demo:call()" means aggregating all same method calls into one single line. The minimum time cost is 0.005428{{}} ms, the maximum time cost is 0.094064{{}} ms, and the total time cost for all method calls (3{{}} times in total) to "demo:call()" is 0.105228ms{{}} . If "throws Exception" appears in this line, it means some exceptions have been thrown from this method calls.
• The total time cost may not equal to the sum of the time costs each sub method call takes, this is because Arthas instrumented code takes time too.

Trace multiple classes or multiple methods

The trace command will only trace the subcalls in the method to the trace, and will not trace down multiple layers. Because traces are expensive, multi-layer traces can lead to a lot of classes and methods that ultimately have to be traced.

You can use the regular expression to match multiple classes and methods on the path to achieve a multi-layer trace effect to some extent.

Trace -E com.test.ClassA|org.test.ClassB method1|method2|method3

Dynamic trace

Supported since version 3.3.0.

Open terminal 1, trace the run{{}} method, and you can see the printout listenerId: 1{{}} .

trace demo.MathGame run{{execute T2}}

Press Q{{execute T2}} or Ctrl+C{{}} to abort

Now to drill down into the sub method primeFactors{{}} , you can open a new terminal 2 and use the telnet localhost 3658{{}} connects to the arthas, then trace primeFactors{{}} with the specify listenerId{{}} .

trace demo.MathGame primeFactors --listenerId 1{{execute T2}}

Press Q{{execute T2}} or Ctrl+C{{}} to abort

At terminal 2 prints the results, indicating that a method has been enhanced: Affect(class count: 1 , method count: 1){{}} , but no more results are printed.

At terminal 1, you can see that the trace result has increased by one layer:

Dynamic trace by specifying listenerId{{}} , you can go deeper and deeper. In addition, commands such as watch{{}} /tt{{}} /monitor{{}} also support similar functionality.