Distributed Log Search Based on Time Series Access and Service Relations

Abstract

Distributed tracing helps administrators to analyze root causes of microservices under system failure. It enables tracking procedures by log messages. Distributed trace log searches require short response times. Therefore, this study proposes a log search method with fast response time to search queries. Log messages are stored on several nodes as blocks grouped by date/time and service-name. The search method focuses on time-series access patterns and service relations. It decreases the number of accessed log messages per query on search. Experiment results show that the proposed method is maximally 0.91 s faster than the parallel method ‘all parallel’.