Chengdu Railway Station, DevOps architecture and operation summit, you may miss the dry cargo (with PPT)


youkedeyunjisuan· 2016-05-19 21:35:46

the scene of the audience from the Internet Co online, including Changhong, Jingdong, Wenxuan, Hong Wei, HUAWEI, tap4fun, Sichuan mobile, sanlingsheng'an, Wanda, who fly immediately Consumer finance, more life, imprisoned, number, testbird, testin, camera360, plump, Venus, digital and other, active scene question more highlights the audience focus on content.

: Xiao Li, Kingsoft Xishanju system operation manager

Kingsoft Xishanju system operation manager Xiao Li, he brought entitled" virtual maintenance practice "project I share, from the beginning of the project evaluation, how to Their business migration to virtualization or cloud, and describes how to do the monitoring and alarm, disaster recovery, continuous operation and maintenance.

: Tianle, UCloud director

UCloud comprehensive development the director brought Tianle speech theme is" design and practice "account billing system with high availability architecture, we focus on shared billing system architecture design and practice of UCloud accounts, and accumulated experience in the process of evolution, the main contents are: the current account billing system and the problems, in-depth account billing system architecture evolution, brief introduction summarize the experience, gray switching scheme. He believes that the architecture is not the best, only the most suitable for the current business development architecture.

the following compiled Kingsoft Xishanju system operation manager Xiao Li and UCloud D director Tianle share content.

Xiao Li: my virtual project management practice of

Xiao Li, KVM maintenance expert. Kingsoft system operation and maintenance manager, former grand game researcher. 15 years of work experience, 10 years of game industry operation and maintenance experience, 5 years KVM virtualization operation and maintenance experience maintenance micro channel subscription number: "KVM virtualization practice" is the author of "the depth of practice KVM," a book.

the share mainly four parts: non technology factor and the pressure of the business model construction, technology and experience sharing, monitoring, disaster preparedness, emergency response, case and public cloud operation and maintenance experience.

this paper mainly intercept non technology factors, pressure of the business model, how the business migration to the virtual environment, software and hardware selection and disaster recovery plan to organize.

, non technology factors

2015 years is cloud computing a large outbreak of a year, the cloud computing virtualization intuitive feeling is gradually increase the talent demand is growing, professional treatment, cloud computing recruitment of micro channel group is unusually active. The second feeling is to do cloud computing friends in the company more and more attention.

1. cloud is a more advanced way of resource utilization;

2. cloud service deployment more efficient and convenient

3. with several years of development, has become the infrastructure and cloud ecosystem really in that big data, video, education, medicine Treatment and other aspects of the application.

. < strong > has from the enterprises shall not cloud the issue into for how to cloud the issue of < / strong >. Enterprise cloud can choose to use public cloud, you can choose to build private cloud, you can choose to use a hybrid cloud, most of the way to use the cloud should be mixed cloud.

calculation on the operation and maintenance of the human impact of the most obvious is the division of the more professional, more high < / strong > requirements. For example, the original you in the interior of a company may be calculated or business operation and maintenance of; in the cloud may the systems, the underlying aspects of the operation and maintenance work to the cloud, which will focus more attention to business, the business better, more professional. If you go to do private cloud, that is, the operation and maintenance of IaaS layer, including data centers, network, security, and so on, only in the presence of large enterprises,. In addition, because the cloud computing platform has a large number of API, if you make good use of these API, can be achieved from the bottom to the upper level of the full open, operation and maintenance aspects of the trend is more emphasis on automation .

second, business how to migrate to the virtual environment

the first step / span> to < strong > persuade the boss and colleagues support virtualization. with the popularity of cloud computing virtualization concept, a lot of people on the cloud is no longer as the beginning of exclusion, but to do the first project, be sure to ensure its success, set a good example.

the second step, how to select potential shares

how to ensure the first project successful virtualization?? We must choose potential , find a good success in virtualization project, it has many features:

  • , now CPU is multi core, single process can be very easy to do; "

  • followed by Microsoft YaHei; font-size: 14px; line-height: INH Erit; "> the utilization rate is not high business , such as the perennial utilization rate of only 20~30% business, through several business will merge into a host, thereby increasing its utilization

  • < / strong>, very often love to do virtualization, because of the rapid deployment of virtualization advice, solve business frequent changes this pain point; span>

  • non core business , if one started with the core business virtualization, if there are problems, will face a great pressure, even will affect the entire company confidence in virtualization, so the first virtual items from the non core business began to.

also not all business when do virtualization, business in the physical pressure on the already very high, it is difficult to by virtualization to do the integration.

the third step, the virtualization project implementation cycle. < / strong > implementation of virtualization should generally follow a kind of following process: < strong > business performance needs assessment < / strong > and the < strong > pressure model design a virtualization solution, build test environment, test system, test business, small scale deployment, the full deployment, the full deployment of the final virtual maintenance.

and solve problems in the implementation of. < / strong > in the implementation process has some problems need to pay attention to, primary concern < strong > virtualization layer stability < / strong > and < strong > virtual machine fast automatic management maintenance < / strong >, then solve the binding < / strong > < strong > and the business more closely, the most important is need to possess a set of monitoring, health, alarm, emergency Xi response plan.

, pressure of the business model analysis

construction business pressure model, how to concretely. The first to be familiar with the business architecture, the how the best map "color: out to be aware of, clear role relationship between .

and then the performance data collection and analysis, there are two kinds of methods:

  • < span style=" color: RGB (136, 136, 136); >" font-family: Microsoft YaHei; font-size: 14px; line-height: 1.5em; "> collection of each project the number of servers and the role of , see the monitoring data and long-term CPU memory pressure situation, general observation two months;

  • two is through the existing script collection server performance, the main data in order to collect more fine;

  • by collecting the pressure data, the pressure according to the model, pressure model, determine the proportion of


software, "color: for the production environment we generally must choose stable version. However, based on the stable version, memory version is good, why? There is a data, the data for a long time, the contrast of the same configuration CentOS 6.1 and CentOS 5.6 CPU power, CentOS 6.1 to than CentOS 5.6. 9% is the kernel version is higher, its CPU interrupt and context switch optimization better, also network IO, disk IO also optimization better.

hardware. < / strong > as a start configuration to be a little bit better, because the configuration was more strongly defended, you can more virtual machine virtual, and you will eventually save the cost; in addition, memory is a little bigger, because after running your host for a period of time, more often than not, you will found that memory is not enough, when the time comes to add memory. Finally, as far as possible to choose the mainstream brand.

application layer (online backup transfer is not

there are two ways:

  • application layer Disaster recovery, basically as same as the original physical machine, you in a physical machine how to do disaster recovery, in the virtual machine with the same method do disaster recovery;

  • virtual disaster recovery, snapshot, buying the mirror copy.

general recommendations on the application level disaster recovery, because in the application layer do disaster recovery consumption of resources is much less. That is, to disaster recovery exercise regularly, everyone familiar with the process, again to verify this in the end is not effective disaster recovery mechanism, can be summarized as two points:

  • all XML virtual machine description files should be regularly cross backup;

  • of regular exercise and ourselves to be familiar with the process and related business also need to allow them to rehearse, problems arise when we can recover soon.

the first cloud is the trend of virtualization is the first step in ; then in the production environment, we try to choose the plan the < /strong> mature technology and perfect, because the production environment to virtualization is the basic orientation; IT skills , no matter what the original aspect of the operation and maintenance, may be more or less the use of virtual operation and maintenance. In addition, we recommend the virtual enterprise internal, word of mouth is also very important, once the problem will affect our reputation to push the need for virtualization.

< P style=" guests "margin-top: 0.2em; padding-right: 0.5em; padding-bottom: 5px; padding-left: 0.5em; color: RGB (33, 33, 34); font-size: 1.8em; line-height: 1; border-bottom-width: 1px; border-bottom-style: solid; border-color: RGB (33, 33, 34); white-space: normal;" > Tianle, UCloud comprehensive development center director is responsible for the overall management of the UCloud data analysis, account billing system, network, multi account platform development project. With more than 7 years of Internet and it R & D related experience, worked in the business school, Royal online, Shengtai Yun, served as senior engineer, head of the Shengtai Yun monitoring and charging product development.

the main interception account billing system architecture evolution process of finishing.

14px; color: RGB (136, 136, 136); "

service architecture" white-space: normal; line-height: 1.5em; >

UCloud service architecture mainly through the following six a stage:" /span>

  • single mode; "white-space: normal;"

  • has released

  • service transformation;

  • style= "according to SET

Room in the deployment of set, press set were across the room heat by disaster recovery .

1. single mode architecture on-line service system

UCloud service architecture is mainly divided into three parts:

  • PHP Web Conosle, responsible for all front-end interactive, Background service logic assembly;

  • platform service, public service accounts, billing, monitoring, name services;

  • business system divided into access data center services in the background layer.

PHP web console, business systems division data center service, on-line platform service composition, web console by Protobuf and all the back-end services for communication.


to solve the problem facing the front of it, we first need to support the Web layer gray gray method contains the following release:

  • Useless households behavior in accordance with the single IP am; IP (area) am; to IP modulo gradually grayscale control sphere of influence;

  • useful households behavior in accordance with the single internal users (Development Account) am; internal test account am; user classification gradually gray published control effect.


  • Gateway layer to the development of API manage the back-end API registration and management, authorization management, flow control;

  • development API layer and solve the interaction layer, requires the integration of cross system logic calls into question, front-end designed injection product interaction and user experience;

  • unified development of single point landing token, system and front end realize cross domain API calls to the front-end code can be completely static.

at this stage, the complete front-end display can be independently controlled release, the full realization of the front end and the back end of decoupling, API protocol assurance forward compatible, web client can arbitrarily reconstructed interactive optimization front-end architecture, realizes cross domain independent deployment, independent of gray strategy between each other without

The lastest articles of youkedeyunjisuan

What kind of sparks will be brushed off by the financial and IT's innovative...

Chengdu Railway Station, DevOps architecture and operation summit, you may...

All the technical details are here (four).

How to build cloud platform architecture? [with PPT]

All the technical details are here (one).

Launch UMCloud engine OpenStack to build a competitive cloud services

Following Ali cloud, UCloud officially joined the CSA Cloud Security Alliance

AlphaGo can play chess, but he can sell? Labs Comet retail challenge with UCloud