Practical and accurate pinpointing of configuration errors using static analysis

Zhen Dong, Artur Andrzejak, Kun Shao

Abstract: Software misconfigurations are responsible for a substantial part of today's system failures, causing about one-quarter of all customer-reported issues. Identifying their root causes can be costly in terms of time and human resources. We present an approach to automatically pinpoint such defects without error reproduction. It uses static analysis to infer the correlation degree between each configuration option and program sites affected by an exception. The only run-time information required by our approach is the stack trace of a failure. This is an essential advantage compared to existing approaches which require to reproduce errors or to provide testing oracles. We evaluate our approach on 29 errors from 4 configurable software programs, namely JChord, Randoop, Hadoop, and Hbase. Our approach can successfully diagnose 27 out of 29 errors. For 20 errors, the failure-inducing configuration option is ranked first.