How to Improve Your Kubernetes Experience with Service Mesh and MLOps

Conference: KubeCon + CloudNativeCon North America 2021

2021-10-13

Authors: Maksim Chudnovskii, Igor Gustomyasov

Summary

The presentation discusses various methods and technologies used in machine learning cooperation systems for anomaly detection, root cause analysis, and predictive auto-scaling in Kubernetes clusters.

The system is divided into two main parts: preparation and evaluation of models, and real-time execution of trend models
Istio is used as a service mesh to collect service mesh metrics, and Permittelsa is used as a data layer to collect time series data
Combining workloads in schedule groups can reduce network resource consumption and optimize overall latency
Anomaly detection methods can significantly reduce the flow of notifications and automate the process of establishing monitoring thresholds
Predictive auto-scaling can proactively predict the required number of service ports using time series data and feature generation

In the presentation, the speaker discussed how slow start of applications became a problem for colleagues who write code in Java. To address this issue, they tried to scale the application proactively using predictive auto-scaling, which always has the necessary amount of replicas at hand based on predicted values of a specific metric. This method can be useful for achieving elasticity in Kubernetes clusters.

Abstract

In this session, speakers will talk about using a machine learning approach to optimize application performance in Kubernetes clusters in a large private cloud (50+ On-Premise Kubernetes Clusters in a Private Cloud, 500+ Compute Nodes, 10+ Istio Meshes. The speech from Sberbank will cover concrete practical cases and tell in detail about the experience of using machine learning models, consider in detail the architecture of the models, as well as the process of preparing training data, which is based on service mesh telemetry.

Materials:

Tags: