FaaSScout: Fast and Full Lifecycle RCA for FaaS Applications Using Salient Feature Mining

Serverless has emerged as a popular paradigm for cloud-native services. However, serverless applications' complexity and dynamic nature pose significant challenges in ensuring system availability and performance. Many root cause analysis (RCA) methods exist for microservice systems, but they are unsuitable for precisely modeling serverless applications. This is because: (1) Compared to microservices, serverless applications exhibit a highly dynamic nature. They have a short lifecycle and only generate instantaneous pulse-like data, lacking long-term continuous information. (2) Existing methods primarily concentrate on the runtime stage, neglecting other stages and failing to account for the complete lifecycle of serverless applications. (3) Current methods exhibit inadequate comprehension and optimization of multi-modal features, leading to lower model accuracy and exorbitant training costs. To address these limitations, we propose <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FaaSScout</i>, a fast and complete lifecycle root cause analysis method for FaaS applications. It integrates multi-modal observability data generated from the platform and application side using <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Global Call Graph</i>. We employ feature reduction techniques to enhance feature efficacy and train a Graph Attention Network (GAT) based graph auto-encoder to compute reconstruction scores for the nodes in the global call graph. Based on the scores, we incorporate historical node stability and transition probabilities to determine the root cause at the granularity of the lifecycle stage of serverless functions. We evaluate on two serverless benchmarks, the results show that <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FaaSScout</i> outperforms other baseline methods with a top-k precision improvement ranging from 21.28% to 82.91%.