EBugs in Cloud Systems

This repository contains a dataset of 210 exception-related bugs, also denoted as eBugs, curated from six widely deployed cloud systems, including Cassandra, HBase, HDFS, Hadoop MapReduce, YARN, and ZooKeeper. The repository also contains a list of inaccurate exceptions* detected by DIET, a static analysis tool that detects inaccurate exceptions by finding inconsistency between an exception’s class and its error message.

* Please refer to our paper (see below) for the definition of inaccurate exceptions and other types of eBugs.

The eBugs dataset

The eBugs dataset contains 210 exception-related bugs. These bugs are selected from more than 60,000 JIRA issues, which were submitted for the six studied cloud systems from 2007 to 2018. Every eBug is studied from multiple perspectives to answer the following key research questions:

• RQ1: How are eBugs triggered in cloud systems?
• RQ2: What is the relation between the triggering conditions and the root causes of eBugs?
• RQ3: What are the impacts of eBugs on cloud systems?

Please check out our paper (see below) for detailed statistics of each perspective, as well as the findings and implications on them.

DIET - Detecting Inaccurate Exceptions using triggering condition Types

Based on the findings on the eBugs dataset, we design and implement a static analysis tool, called DIET, to detect an important class of eBug, i.e., inaccurate exception. From a high level perspective, DIET detects inaccurate exceptions by understanding an exception’s class and error message. If the class and the error message imply different types of triggering conditions, DIET will flag the exception as an inaccurate one.

Please refer to our paper for more details about how DIET works.

Inaccurate exceptions detected by DIET

We apply DIET to the latest versions of the studied systems, and find 31 inaccurate exceptions. Two of them are eBugs that cause failure symptoms. The remaining 29 inaccurate exceptions are bad practices, where their classes and error messages are indeed inconsistent. We submit them to the developers of these systems in the form of JIRA issues. We group the eBugs and bad practices that are in the same file into the same JIRA issue. At the time of writing, developers have confirmed the two eBugs and 21 bad practices. Moreover, all of the confirmed inaccurate exceptions are previously-unknown. Below is the list of all the confirmed eBugs and bad practices. We only list one instance of HDFS-14467 and one instance of HDFS-14468 because all the bugs in these two issues have the same exception class and error message. You can find more details about the detected eBugs and bad practices in both the dataset and our paper.

Publication

Understanding Exception-Related Bugs in Large-Scale Cloud Systems [preprint] [slides]
Haicheng Chen, Wensheng Dou, Yanyan Jiang, Feng Qin
34th IEEE/ACM International Conference on Automated Software Engineering (ASE 2019).