Abstract

Modern cloud environments, characterized as resource-dynamic with new developments, see prevalence in ephemeral resources. Such resources can be unstable in resource availability, experiencing both anticipated and unforeseen terminations during utilization. Their prices, although attractive, can be fluctuating over time. The presence and prevalence of ephemeral resources in cloud environments pose a challenge to current cloud-native databases and workloads, which requires a rethink of design principles and necessitates the new primitives: query preemption, resource arbitration, and cost tolerance. In this paper, we design Ratchet, a resourceadaptive query execution framework, to realize the identified primitives. Ratchet enables adaptive query suspension and resumption at various granularities, resource arbitration for complex and heterogeneous workloads, and a fine-grained pricing model to utilize dynamic cloud resources without the risk of unexpectedly high prices. We also explore emerging directions to inspire future research.