Thinking about SLA of cloud computing

One SLA（ Service level agreement ） What is it? ？

Cloud computing SLA It is part of the user agreement of the cloud platform signed by users , It is often thought that when using cloud services SLA And take effect ,SLA Generally, the agreement for the availability of cloud services ：

You can refer to ：https://www.huaweicloud.com/declaration/sla.html

1 Definition

• service
• Service cycle
• Total service cycle time
• Service not available *（ Specific business can define the judgment basis ）

The judgment of service unavailability is business related , There are several common ways to judge , It is considered that the service is unavailable during this time

1） Within unit time , Because the server access failed 10 Time

2） Unit time , Failed to perform a task xx Time

• Service unavailable time
• Service availability = （ Total cycle time - Unavailable time ）/ Total cycle time * 100%
• Monthly service fee

2 compensate

• The way

Service availability	Add
99.9%≤ Service availability <99.99%	monthly 10%
95%≤ Service availability <99.9%	Monthly service fee 30%
Service availability <95%	Monthly service fee 50%

• Application time limit
• Application method

3 Limit （ Exclude compensation conditions ）

Two What is service unavailable ？

1 Server perspective

basis ： monitor + Warning indicators , The regulated end similar to mode modulation

defects ： Not accurate enough , The problem of not accessing the layer through the system will be ignored , But it really affects user business

reflection ：

• The access layer needs to be thinner , There are higher requirements for availability
• abnormal （ System internal error 、 Business error 、 Time delay exceeds expectation ） Reasonable handling of can also achieve usability
• The observability of the system is good , Do a good job of error code specification
• More accurately judge the availability of the user side

2 Client perspective

basis ： Real service results , Similar to the main mode of modulation

defects ： How to collect data from the service side

reflection ：

• client-sdk Asynchronous aggregation reporting ？
• SLA It should be service log statistics + Client burying point 2 The two methods are implemented independently , Learn from others' strong points and close the gap