Shulou(Shulou.com)06/03 Report--

1 what is full-link pressure test?

Based on the actual production business scenario and system environment, simulate the massive user requests and data to stress test the whole business chain, and continuously tune the process.

2 what problem does the full-link pressure test solve?

Verify the stability and scalability of the service under peak traffic

Verify the stability of the newly launched function

Conduct downgrade, alarm and other fault drills

More accurate capacity assessment of online services

Evaluation of supporting capacity of computer room failure and sudden increase of traffic

3 when do you need it?

The speed of business development (with the continuous development of business, the number of dependent modules is increasing. Need to find out the deficiency to solve)

Within a predictable period of time (preferably half a year, a quarter is a little late), the business will develop rapidly and the capacity of online machines will have to be greatly expanded.

Sometimes the expansion is not linear. If you expand from two to four, your service capacity may be increased by twice as much.

If you continue to expand the service capacity, you may not be able to improve it, because it will be limited by other modules, such as DB, public composition, middleware, etc.

Less and less confident about the results of stand-alone pressure test.

3.1 A good indicator, generally speaking, we will press our own module.

3.2 the stress test of a single machine does not represent the real online scene, and the heart will become more and more virtual. At this time, it is necessary to consider the full link.

4 how many stages is the stress testing method divided into?

Make a service call to an online stand-alone or cluster

Record the online traffic and play it back on a single machine

Drainage manometry is carried out by modifying weight

Full-link pressure test (the difficulty is to sort out the boundaries of the link)

Full-link pressure testing is aimed at modern more and more complex business scenarios and full-link system dependence. Therefore, we should first split the core business and non-core business, identify which business scenarios and modules are targeted at the traffic peak, and prepare for capacity expansion. Instead of expanding all system service clusters dozens of times in order to solve the impact of massive traffic, this will result in unnecessary cost investment.

5 what are the core functions of full-link pressure testing?

Data construction

Pressure test isolation

Service isolation

Data isolation

6 what work needs to be done in the stress test?

Data model construction

Construction of pressure testing platform

6.1 the machine data in the pressure measuring machine can be collected and viewed in real time, the pressure testing can be stopped at any time, and the real-time error rate reaches the threshold within a certain period of time.

System code modification (pressure test mark, mock, business logic)

6.2. write requests are written to shadow areas (such as marking in header headers, storing, caching, message, log, etc.), dependent external services do mock processing (SMS, email, push, etc.), and multi-thread transformation

Real flow reservoir, released in batches (playback of traffic generated during peak business periods)

Step by step pressure test

7 Why do you need capacity planning?

When to increase or decrease machines, ensure system stability and save costs

8 what indicators does the stress test focus on?

Application level

Server resources

Basic services: middleware and database

Core indicators: (response time does not use average response time, focus on 95 lines; throughput is linked to response time; throughput is linked to success rate)

System:

System capacity

Service performance

Infrastructure bottleneck

Middleware bottleneck

The direct dependency influence of the system

9 how to obtain the service capability of a single machine

The simulation request achieves the purpose of stress testing by making a simulation request call to a machine in the production environment.

Request forwarding forwards requests from multiple machines in a distributed environment to one machine

Adjust the load balancer to modify the weight of the load balancer device, so that the stress-testing machine assigns more requests.

9.1 while carrying out the pressure test, the system load of the pressure test machine is detected in real time, and the pressure test is stopped as soon as the system load reaches the preset threshold. At the same time, outputting a pressure test report, the single machine service capacity value obtained through the stand-alone pressure test is also a very important reference for capacity planning.

Minimum number of machines = estimated business visits / stand-alone capacity

9.2 when planning the capacity of a single system, the capacity of all the dependent links is unlimited, which makes the value of the stand-alone capacity we obtain is optimistic; 9.3 when using single-system planning, there is no guarantee that all systems are in place in one step. Most efforts are focused on core a few core systems; 9.4 some of the problems will only be exposed under really large traffic, such as network bandwidth, and so on.

The focus of full-link voltage test transformation: https://www.codercto.com/a/55270.html

Meituan quake pressure test: https://tech.meituan.com/2018/09/27/quake-introduction.html

Meituan's practice of automatic pressure Test in the whole Link

Ali full-link pressure test

Full-link pressure test with praise

JD.com full-link pressure test

Ele.me full link pressure test

The solution of Didi full-link pressure test

Practice of logical thinking in full-link pressure testing

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.