Anomaly Detection Of Container Cloud Platform Based On IForest

With the maturity and popularity of technologies such as Docker,Kubernetes and Istio,the container cloud platform is becoming the transformation direction of more and more enterprise IT infrastructure architectures.In particular,Internet companies have chosen to run online service applications on the container cloud platform to provide services to customers.In the container cloud platform,online service resource usage has been constantly changing due to real-time fluctuations in the number of service requests.The method to manually configure alarm rules is too dependent on the experience of Ops,and configuring a highly available alarm rule often requires multiple practices.Besides,the normal jitter of user visits will cover up some online service abnormal events,resulting in too long response time when business problems occur.Online services also have the characteristics of diversity,and different services have different preferences for various resource indexes.So it is difficult to detect all service states using a fixed standard algorithm uniformly.To resolve these issues,this thesis proposes an anomaly detection method based on i Forest.This method focuses on the status of online services running on the container cloud platform.Without the need of setting an alarm threshold,this method can detect the service anomaly with a short sequence of time series data.Considering the volatility of online service resource consumption,calculated a set of stable resource index optimization data through the service request number indicator data.Comprehensively analyze the resource bias of different services,the random selection of features is changed to weighted selection in i Forest.and the accuracy of anomaly detection is improved by increasing the weight coefficient of service-biased resources.The method optimizes the online service resource index data firstly,then uses the semi-fixed sliding window model to select the data set,and the isolated forest will be constructed based on the dynamic resource feature weighting method.Finally,the anomaly score of online service in the stage to be tested will be calculated.This is the anomaly detection process.The experimental results demonstrate that this method can accurately detect the anomaly status of online services running on the container cloud platform.This thesis designs and implements an anomaly detection system for the container cloud platform.The experimental results demonstrate that this system can collect monitoring data of each dimension of online service in real-time,and ensure stable operation while performing automatic anomaly detection and alarm.