Abstract: Cloud applications are often deployed in shared data centers to optimize re-source allocation and improve management efficiency. However, since a cloud application often has a large amount of different microservices, it is difficult for operators to analyze these microservices with a unified model. To deal with the above problem, this paper proposes a sequential trace-based fault diagnosis framework called as Midiag by mining the patterns of microservices' system call sequences. Midiag collects system calls with a non-invasive lightweight tool, and then uses k-means to cluster system call se-quences as patterns with the longest common subsequence. The GRU-based neural network is employed to model the patterns of system call sequences to predict the next system call, and thus Midiag diagnoses faults by comparing the predicted system call and the actual one in a specific pattern. We have validated Midiag with many different types of applications deployed in containers. The results demonstrate that Midiag can well classify these applications as different types and accurately diagnose the applications injected with faults.
Authors: Yao Sun (Nanjing Institute of Big Date, Jinling Institute of Technology, China); Lun Meng (Hohai University, China); Shudong Zhang (Capital Normal University, China)
Email: firstname.lastname@example.org, email@example.com, firstname.lastname@example.org