Pressure sensing full link analysis

Software testing architect from the original club, the authors cctester

Recently network transmission, micro-channel pay collapse, Haro is a problem, some companies Performance Test Architect Job began to hot up, now called the Link-pressure measurement, what is the full link voltage measured it, with the traditional press the difference is measured Shane? A Liti full link was first out, in 2012, when 11 double, zero, the system transaction success rate of less than 50%, under a single error, error shopping cart, and accompanied by a large oversold, and later made a full link pressure measurement, this article will talk about the key points to link the whole pressure measured.

Many companies interviewed, there are a lot of performance test form, the company is also a general tool stage, do simple monitoring, and the report ended, the basic approach is to take such a form, and some relative development team is responsible for, will process pressure measured in assisting diagnosis to see if there optimization points, in general, how many will find some problems, it will have some effect, but often big promotion, would give rise to other problems, leader asked not done pressure test yet? What do you think have done, but if they were not doing enough .....

1. What is the full link online performance test:

Based on real user scenarios, the actual online environment, in accordance with established traffic, links to various business processes stress test.

2. Why do the whole performance test link:

Many cable companies under the performance test, why do we need to do the whole link, can solve the problem in general performance test it? I think to do performance testing in each environment are complementary processes, performance testing under online, because the machine monitoring, as well as rapid deployment of sufficient appropriate permissions, we can quickly locate some performance bug, such as memory leaks, deadlocks, ultra selling and other issues, but the machine under the line of indicators can not achieve accurate feedback to the actual situation of the line, we can not simply filled with a lot of experience through some formulas to calculate the value of such results and racking our brains also lacks much difference , coupled with the on-line environment under the most points to the link, module-based pressure measurement, pressure measurement so full link in this context was born, our premise is already online module pressure measurement is completed, no bottleneck situation under initiate, carry out adequately simulate the pressure measurement on the link online.

3. What is the core link the whole pressure is measured?

Whatever the test, the core of things must needs analysis, that the whole point is to link performance requirements analysis performance at Shane, and traditional line test so what difference does it make?

Request Source:

In the traditional performance test line, generally we get started debugging interface parameters, write the script, tested in accordance with the scene, and we need to online statistics based on actual data source, contains the web end, app end, the small end of the program, etc., this is our client data sources, as well as the distribution of our carrier bandwidth consumption, cdn node, thus involves measuring the pressure outside the network, the difference in pressure measurement strategy and the details of the network outside the network is still relatively large, This article does not make specific discussion.

Architecture topology analysis:

Deployment structure lines tend to be more complex than many of our test environment, test environment is often a small branch line, dependent on a variety of micro cluster online services, middleware, db needs research is very clear, how many servers, server on the deployment instance, every detail will affect the results of measurements of pressure, as well as the accuracy of the analysis.

data analysis:

Data analysis can be divided into many levels, in general the performance of the pressure test, we usually focus on parametric data and db data, measured in full-Link pressure, also we need to focus on, redis data, mq accumulation, as well as key size of the actual bandwidth the effects of these are related to middleware-related, if there are problems, the impact on the site is often devastating, this often under-bandwidth WLAN test is not covered, the line will cross the room to call, so the need to focus on this particular .

Monitoring and analysis:

Under most circumstances, we will do the monitoring hardware layer include cpu, bandwidth, memory, disk, etc., then the client data acquisition, indicators generally measured by the pressure data collection, but these links in the whole pressure measured in still seems also there are basic, we need to go through more dimensions server monitoring business metrics data contains the cluster service, real-time single-layer data db, container-level resource monitoring data and other content, as well as combined health of indicators, measuring the pressure required on-line set thresholds, to avoid the risk as much as possible online, prevent churn.

Pressure measurement goal setting:

Many online companies when we measured under pressure because there was no reference data, possibly down to the inflection point as the preferred target, and mature Internet companies will do to assess the capacity of the line, usually a combination based on past performance and flow, there will be a certain percentage estimated growth, as well as to evaluate by pushing conversion rates, personal feel that we can do long-term model to the data accumulated, reach the reference value of experience.

Traffic Playback:

First, the company can do very little traffic playback, this involves the transformation of the system, the key is this data processing, playback can achieve a lot less traffic, the test will be a lot of preparatory work, but the task of transformation to develop early also very heavy, closed for three months only transform a prototype version of Ali is also a development team, any company can refer to a technology type, but do shades will be very different.

Share this article from the public micro-channel number - the road to self-improvement engineer (Engineer_Way)

In the original source and reprint information see article details, if infringement, please contact [email protected] deleted.

Guess you like

Origin www.cnblogs.com/hellolonewoif/p/12168416.html