Model-Based Analysis of Serverless Applications

Abstract

Serverless computing is a relatively new execution model where the cloud platform provider manages the allocation of resources for containerized functions dynamically. This evolving paradigm is called Function as a Service (FaaS). The statelessness of these functions enables the application to be scaled up elastically in the case of peak loads. They can be tested easily in isolation, but the behavior arising by integrating them to an application is both hard to predict and test. The parallel execution of the functions and the shift of its state to data storages can cause several workflows accessing the same data. These workflows are hard to detect, particularly for complex applications. Therefore, we suggest an approach for modelling an existing serverless application based on a specialized graph holding all relevant features. Our serverless-specific model can be applied during the whole life cycle of a complex application and offers a good basis for this specific class of applications. It helps to optimize an existing system by identifying hot spots, supports the generation of test cases and can be used to monitor an existing system. Furthermore, we show how the generation of the model can be automated by realizing a tool supporting Amazon's AWS Lambda.