19-02 Evolution of architecture technology selection based on service level

From Scratch - Single Service Application

Single application technology selection

• (GitHub, Gitee...) Search for products with threads
• Use the most familiar technology and go online at the fastest speed
• If you have the money: consider a commercial solution

How to choose a personal small program technology

• Search for software to quickly build programs
• Technology Selection
  • Back-end technology selection
    • Use the most familiar technology to achieve the fastest development speed
    • MySQL、MyBatis、SoringBoot
  • Front-end technology selection
    • 

Separation of application services and data services

• The start-up stage of enterprise application
• two purposes
  • Choose the right technology to implement the project
  • Set the tone for the future direction of the project
  • 

issues to consider

• Choice of operating platform
• Choosing a Commercial or Open Source Solution
• Determine the R&D model of the project
• Determine the specific technology used

Get twice the result with half the effort - introduce the cache system

issues to consider

• where to use the cache
• What type of cache to use
• adopt that caching mode
• What cache component to use

cache location

• client cache
  • cache in browser
  • app cache
• caching in the network
  • proxy server cache
  • CDN cache
• server cache

How to determine where the cache is located

• If there is no performance bottleneck, don’t consider it, if it is slow, use the cache there

cache type

• memory cache
  • Very fast, data may be lost
  • Applicable to scenarios with very high speed requirements and data loss tolerance
  • heap cache
    • advantage:
      • No need for serialization, deserialization
      • good performance
    • shortcoming:
      • Will affect GC
      • Capacity is limited by the size of the heap memory
      • Generally stored as soft references or weak references
    • Applicable scenarios for in-heap caching
      • Store very hot data
• disk cache
  • Performance is worse than memory cache, data will not be lost
  • Applicable to scenarios that require persistence

load balancing

DNS-based load balancing

• Configure the same resolution record for multiple addresses on the DNS server
• advantage:
  • The work of load balancing is handed over to the DNS server, reducing the maintenance work of website management
  • The technical implementation is relatively flexible, convenient, simple and easy to implement, and the cost is low
  • Wide range of applications, suitable for most TCP/IP applications
• shortcoming:
  • Generally does not reflect the current running status of the server
  • After a server goes offline, even if the DNS record is modified, it may take a long time for the record to take effect
  • Ensure that the target address pointed to by DNS resolution is highly available, and the address will not be changed frequently
  • Generally speaking, large websites will use DNS as a first-level load balancing
    • The IP pointed to by DNS does not correspond to a machine, but a highly available server cluster
    • The dig command can query records

Load balancing based on reverse proxy

• The request passes through the reverse proxy, and the load balancing algorithm is provided by the reverse proxy component, and a server address is calculated and returned
• represent realization
  • NGINX
  • HAProxy
  • Apache

Typical process of Internet project load balancer evolution (experience)

• Early project: NGINX
• Mid-term: High availability of NGINX combined with Keepalived
• After that, install LVS or F5 to expand multiple NGINX
• If an LVS cluster is too hungry, it will be combined with DNS to expand LVS

Stateful vs Stateless

• Status: Whether the server should store the user's login status
• Whether the server should maintain the user's session

Stateful

sticky session

• After the client logs in on a Web Server, subsequent requests will be bound to the Web Server instance
• 
• Advantages and disadvantages
  • No need to introduce additional components
  • easy to implement
  • There is a single point of problem: additional failover is required
  • There may be an imbalance

session sharing

• Use session to keep the session, and store multiple application instances in a central storage
• Advantages and disadvantages
  • Additional components need to be introduced, even if any web application crashes, it can still be used
  • But once the Session Store crashes, all sessions will be lost

session replication

• Replicate sessions between Web Server instances
• Advantages and disadvantages
  • No need for failover
  • No need to introduce additional components
  • Session replication consumes bandwidth and memory

no status

• The server does not record the user's login status: the server no longer maintains the session
• When the user logs in, a token is provided, which is usually encrypted
• After that, each request will bring this token (passed in header, URL parameter, and cookie)
• Stateful Cons = Stateless Pros
• Disadvantages of statelessness:
  • Once the token is issued to the user, it is difficult to control its offline time
    

Stateful vs Stateless

read-write separation

CDN

• Static files: works really well
• Dynamic data: not working well
• dynamic content static
• static and dynamic separation

Principles of CDN composition

Issues to be considered in CDN technology selection

• self built or commercial
  • Prioritize the use of commercial CDN
  • When the commercial CDN can't meet the business needs and can't stand it, then consider self-built
  • Reason: Self-built is not cost-effective
  • Xiaomi, Kuaishou, Dianping => are all commercial CDNs
• How to choose a CDN
  • speed
    • Number of nodes
    • bandwidth capacity
    • node distribution
  • Function

    • speed up optimization

      • DNS optimization
      • 

    • monitoring statistics

      • Real-time monitoring: click rate, hit rate, occupied traffic
      • Access log monitoring

    • safety

      • Anti-leech
      • IP black and white list
  • price
    • Billing by bandwidth peak
    • Pay by traffic

Full Text Search

• Reduced query pressure on the database
• Improve application performance and improve user experience
• Five Paths to Realize Full-text Search
  • Use the database's built-in full-text search capabilities
  • Full-text search using deep database integration
  • Use an open source full-text search engine
  • Use commercial full-text search engines
  • Self-developed full-text search engine

Use the database's built-in full-text search capabilities

• advantage
  • low learning cost
  • convenient
  • Maintains architectural simplicity without introducing new components
• shortcoming
  • The search engine is not separated from the database, and the responsibility of the database is not single enough
  • The ability of full-text search is limited by the query ability of the database
• Not recommended for use

Full-text search using deep database integration

• Representative implementation: Sphinx
  • can run independently
  • Can be deeply integrated with MySQL and PostgreSQL
• Selection suggestion
  • Sphinx is very popular in the industry
  • If you are interested in the cost of operation and maintenance and do not have high requirements for scalability, you can try
• Tips
  • Compared with other full-text search engines, Sphinx is not very powerful
  • There are not many domestic documents
  • More and more enterprises are migrating from Sphinx to the third mode

Use an open source full-text search engine

• Currently the most mainstream way
  • rich choice
  • Very good scalability
• shortcoming
  • Requires independent deployment of search engines
  • Need to use the search engine specific syntax to operate the search engine
    • Both disadvantages and advantages, the disadvantage is that there is a learning cost, the advantage is that it can support more complex queries; in addition, search engines such as Elasticsearch support both DSL and SQL
  • Search engine and database are two independent software, data consistency needs to be considered
    • Synchronize with plugins like logstash-input-jdbc
    • When the application writes to the database, it also writes to the search engine
• Selection suggestion: you can use it with confidence

Use commercial full-text search engines

• Alibaba Cloud Open Search
• Microsoft Azure Search
• Not many success stories
• Selection suggestion
  • Commercial products have good service, save trouble and worry, and can be used with confidence
  • Consider the impact of closedness of commercial products: For example, it is difficult for manufacturers to customize according to your special needs

Self-developed full-text search engine

• Ali, Suning, Wind Information, Oriental Fortune
• Advantages: The core technology is in your own hands, with better flexibility, and you can adjust and optimize from the bottom when encountering problems
• Disadvantages: The technical requirements of the team will be relatively high, and the cost will be very high