Adaptive Tracing and Fault Injection based Fault Diagnosis for Open Source Server Software

The high overhead of tracing, the amount of up-front effort required to select trace points, and the lack of effective data analysis model are the significant barriers to the adoption of intra-component tracing for fault diagnosis today. This paper introduces a novel method for fault diagnosis by combining function level adaptive tracing, fault injection, and graph convolutional network. In order to implement this method, we introduce techniques for (i) selecting function level trace points, (ii) constructing approximate function call trees for programs when using adaptive tracing, and (iii) constructing graph convolutional network with fault injection campaign. We evaluate our method on four widely used open source server software: Redis, Nginx, Httpd, and SQlite. The experimental results show that our method outperforms log-based method, full tracing method, and Gaussian influence method in terms of accuracy, efficiency, and performance impact on the diagnosis target.