Recently in thinking how to do small flow, search the Internet, summed up such as the following:
1, the premise, the site PV has reached a certain scale, for example, millions of PV, do not make a small flow update function, may bring large area flow loss. A small flow test is required under such conditions.
2, what is small flow? A small portion of the PV total is used for new functions, while the remaining users still use the original functionality
3, how to do?
3.1. Program preparation (4 steps required)
-Deployment and configuration. You need to deploy the server cluster for the small stray labs separately and deploy the new functionality to these clusters. Httpserver configuration based on the forwarding of cookies
-cookie interception and forwarding processing. Block all requests in filter, add the cookie flag for HTTP streaming, and then 302 to the URL of the current request. Don't forget to erase the cookie (assuming you don't need a cookie)
-code management. New function Pull branch, test finished after the merge to the trunk
-Log monitoring. Good PV statistics, good source statistics
3.2, Methodology (more virtual, first beat the head to think of one)
The amount of-PV to be done
-How much PV the experimental group needs, based on what theory determines
-Set how many comparison groups
-How does confidence count?
-what metrics are monitored (PV?) Uv? Order quantity? )
PS: The individual thinks the methodology is not universal, it is good for his own