Contents of this test car is a half-price spike activity
After landing, if you have an appointment before the scheduled event succeeded, then it is eligible for buying activities, if not before the activities involved in the reservation, you will be prompted: you do not buy no appointment qualifications.
Before 12W appointments have people booking successful, the spike activities began when the rule is: The first person able to request a response in the places, all the other would prompt: buying failed
In the middle there is a rule: After clicking the buy button will pop up a verification code, verification code to enter the correct verification OK , before we actually submit snapped up post
So the question arises: complicated by the presence of 2 places:
1, codes
2, snapping Interface post Interface
Analysis of pressure:
Because not previously accumulated data online, so I do not know how much there will be real pressure all parties to a goal not only toward the ultimate application of the unit under pressure concurrently measuring machine
Because of 12W people have successfully reserved, then spike Offers up to 5W high, the intensity of the major events can be described as increasing the difficulty to measure pressure, estimated that buying activity 10 minutes, and there may be a 12W request, and concurrent activities reached a unknown relatively large value, it is unknown why say, because not everyone gave this data, light chasing me to the limit, I was speechless. Limits exist only under certain circumstances.
My analysis:
1, using LR , the maximum number of users 10 000 , the set point setting 10000 , regarded as the limit. This limit does not limit my application, I concurrency utilities limits, LR maximum number of users allowed is 10000
2, the machine will not only 1 table, even if a station did not test the significance of the line for at least 2 or more units. Because the pre-test verification code when performance is poor, so I simply used a 7 back-end
Script analysis:
1, codes
(1) code, used only in the script to check for out verification picture, on the line. Figure get out to prove that it is not loaded. It is also problematic. To count the error rate.
2, buy post Interface
(1) large concurrent, business logic validation, confirmation message interface has returned, there is no "campaign has not started," "no reserve", only one successful, the other full of failure prompted panic buying
(2) large complicated, snapping successful uid actual storage UID consistent amount of data and the like.
(3) large concurrency, see the error rate per unit time
LR solve some of the tools configuration errors:
1、captcha_info.c(6): Continuing after Error -27780: [GENERAL_MSG_CAT_SSL_ERROR]connect to host "t-captcha.pcauto.com.cn" failed: [10054] Connection reset by peer
captcha_info.c(6): Continuing after Error -27776: Server "t-captcha.pcauto.com.cn" shut connection during attempt to negotiate SSL session
The reason is that LR does not support SSL , open the recording options in the script -options-network-port Mapping-add entry
Certificate PEM , under normal circumstances exist in Tengines read SSL under the certificate directory, export locally, then LR can be uploaded configuration.
note:
Remove
Remove
2、The load generator is currently running the maximum number of Vusers of this type
To buy an example script analysis:
This interface returns the data is json content and app pop-up prompts
Because I want to check the activities underway, under a large amount of concurrency, various return results right, and the results confirm the correctness.
script:
Rendezvous Settings:
Hello scenario run this action can
Policy settings:
Pressure results:
Pressure analysis results:
Tools concurrent 1W , the request can be confirmed 1W / s while removing hair, but because by the network speed, response speed, Resin allow the maximum number of threads configured limit (resin configuration 256 , over 256 requests simultaneously over time, we will wait, it will not completely concurrent, although the tool made a request, but not the actual request to the application layer ) , then in order to check the real concurrency. I accesslog analysis per second, specifically how many logs over. After analysis,
The actual per concurrent processing only the highest 5200+
Can be obtained, per 1W request / S , actually only got the highest response 5K / s