Memory Leak Detection Using Transient Workload Detection And Clustering

摘要

Methods and apparatus for memory leak detection using clustering and trend detection are disclosed. Performance metrics are collected from an executing process. A first statistical analysis of at least one metric is used to identify trending and non-trending workload periods for the process. A second statistical analysis on the metrics for the non-trending workload periods is used to determine clusters of metrics corresponding to stable workload levels. A third statistical analysis is performed on each of the clusters to determine whether an upward trend in memory usage occurred. If an upward trend in memory usage is detected, a notification of a potential memory leak is generated.

主权项

1. A method, comprising:
performing, by one or more computing devices:
classifying, based at least in part on a first statistical analysis, one or more performance metric collections corresponding to an observation period of an executing process into a first set of metrics representing one or more sub-periods during which a workload of the executing process exhibited a change trend, and a second set of metrics representing one or more sub-periods during which the workload did not exhibit a change trend;grouping, based at least in part on a second statistical analysis, the second set of metrics into one or more metric clusters, wherein each metric cluster of the one or more metric clusters represents a respective statistically stable workload level;detecting, based at least in part on a third statistical analysis of a particular performance metric corresponding to the one or more metric clusters, whether an upward trend in memory usage at the executing process during the observation time period is indicated; andin response to detecting that an upward trend in memory usage at the executing process during the observation time period is indicated, providing a notification of a potential memory leak at the executing process.