The classification of functions (group_list.json) is manual for Node Margin Strategy

[ema-pe]

The models in Node Margin Strategy require the number of requests per second expected for each feature group (high, medium, and low usage). This makes it possible to not train the models for each specific function, which is why there is a group_list.json file: it contains the list of functions for each group. The data is then used to calculate the number of predicted requests for each group, starting with the number of requests per second for each function in execution.

The current implementation is limited: the DFaaS agent is not responsible for classifying functions, but reads (see groupsreader.go the function groups from a static file (group_list.json). The file path can be specified using the AGENT_GROUP_LIST_FILE_NAME environment variable. This means that after the agent starts, the user can add (or remove) a function by editing the file (since the agent reads the contents each time GetFuncsGroups is called) and manually classifying the function.

We should improve this by automating the classification of a function and automatically updating this file when a new function is deployed.