Always-On Recording Framework for Serverless Computations: Opportunities and Challenges
Shreyas Kharbanda, Pedro Fonseca · 2023
Serverless computing simplifies cloud programming by managing infrastructure and providing basic primitives for building distributed components. However, these efforts tend to increase complexity as developers create large, interdependent applications. The highly distributed nature of serverless computations makes it challenging for developers to comprehend the granular behavior of their applications. To address this problem, we present our vision for an always-on recording framework for serverless computations. Our goal is to efficiently capture low-level details of serverless computations by refining the ideas of traditional record-replay approaches while exploiting the properties of serverless environments. This envisioned approach is poised to provide greater serverless computation visibility for developers to diagnose faults, understand interactions between components, conduct performance analysis, and ensure security and efficiency. Fully tapping this potential will require exploring several research directions, ranging from low-cost recording, storage, and data management systems to practical replay analysis tools.