Using python to implement high-performance testing tools (2)

In the previous article "Using python to implement high-performance testing tools (1)" we optimized performance by changing the python parser, but it is still far from actual needs. This article introduces optimized code for automated testing.

Option 2: Optimize the code

          If a worker wants to do his job well, he must first sharpen his tools. To optimize the code, you must first find the bottleneck of the code. The easiest way is to add log or print. It needs to be deleted after debugging is completed, which is more troublesome. Python also provides many profile tools: profile, cProfile, hotshot, pystats, but the results provided by these tools are not very readable. It is not intuitive enough to see which function or line takes up the most time at a glance. python line_profiler provides such a function. You can intuitively see which line takes up the most time. It can be said to be "fast, accurate and ruthless". Download address: http://pythonhosted.org/line_profiler/

After installing line_profiler Finally, there will be a kernprof.py in the C:\Python27\Lib\site-packages directory. Add @profile on the functions that may have bottlenecks, such as the following example:

    @profile
    def create_msg2(self,H,msg):
        li = msg.keys()
        msg_type=li[0]
        ULR_avps=[]
        ULR=HDRItem()
        ULR.cmd=self.dia.dictCOMMANDname2code(self.dia.MSG_TERM[msg_type])

        if msg_type[-1]=='A':
            msg=msg[msg_type]
            self.dia.setAVPs_by_dic(msg_type,msg,ULR_avps)
            ULR.appId=H.appId
            ULR.EndToEnd=H.EndToEnd
            ULR.HopByHop=H.HopByHop
            msg=self.dia.createRes(ULR,ULR_avps)

        else:
            self.dia.setAVPs(msg_type,msg,ULR_avps)
            ULR.appId=self.dia.APPID
            self.dia.initializeHops(ULR)
            msg=self.dia.createReq(ULR,ULR_avps)

        return msg

Run this file: kernprof. py -l -v D:\project\mp\src\protocols\libdiametermt.py, get the following results. From this picture, you can intuitively see that the setAVPS method takes up 96.6% of the time. Then locate this function and add the @proflie modifier again (Profile can be added to multiple functions at once). You can further see setAVPS The ratio of time taken by each line of code in the function.

Through step-by-step analysis, we can see that in the open source protocol library, in the setAVPS method, the attribute of finding avp is searched from a loop of 3000 , each AVP needs to be cycled 3000 times, there are at least 10 avps in a diameter message, and each time encoding an avp needs to be cycled 30,000 times. Our initial solution was to delete many avps that were not used in our performance testing (there is no way, test development resources are limited, and many times there is no good design, first make something that meets the needs.), but it only improved At about 150, it is still far from the demand. So we changed AVP to dictionary mode, so that we can quickly find the attributes of AVP based on the name.

In addition to code optimization, the number of encoding avp threads is also increased. The following chapters will talk about the impact of multi-threading and multi-process on performance. to be continued. . . .

[Recommended course: Python video course]

The above is the detailed content of Using python to implement high-performance testing tools (2). For more information, please follow other related articles on the PHP Chinese website!