To view the Chinese version of this schedule please go here. 请点击此处查看中文版本。
Please note: This schedule is automatically displayed in China Standard Time (CST). To see the schedule in your preferred timezone, please select from the drop-down menu to the right, above "Filter by Date."
With the continuous development of cloud native ecosystem, more and more services run their systems on Kubernetes. Baidu supports the creation and destruction of a large number of K8S clusters, elastic scaling, and configuration changes,These operations are cumbersome、complicated and dangerous, HOW to build a stable multi-cluster management system?
Baidu learnt from the declarative way of cluster-api, use CRD to abstract various resources in the cluster to solve the problems like machine heterogeneity, plug-in diversification, and idempotent operation. The administrator k8s cluster can manage 300+ user k8s clusters and 4000+ heterogeneous nodes to ensure that the cluster can be maintained, observed, and traceable. This talk will share the experience of refactoring the traditional cluster management systems into declarative way and how declarative APIs fit into our requirements.
Yasong Xu is a senior software engineer at Baidu Cloud Native Team, responsible for development of container cloud platform, focusing on monitoring, reliability and performance of multiple Kubernetes clusters.
Cloud native service is much more elastic and customizable than traditional cloud service. Computing power, storage and network capability should be realtime allocated on demand, metrics for metering and parameters for billing are complicated, product pricing strategies will count on many parameters not only include resource metrics, and different resource providers will use different procedure on resource creation and releasing, so hard-coded metering/billing system can not meet fast-growing requirements. A universal model for pricing, standard API for metering and billing is a strong requirement in cloud-native era. Also, the metering/billing system itself should be realtime, robust, scalable and secure. In this topic, a metering/billing model is proposed to fullfill all these requirements, and a working system is implemented according the model.
Dan Ma is a Senior Software Engineer of QingCloud who is responsible for development of kubernetes observability products and metering/billing system. He focuses on Kubernetes, Big Data, Metering/Billing models and AI technologies. He is interested in open source technologies and... Read More →
Anne Song is a Product Manager of QingCloud, who engages in the functional definition, product plan and design for the platform of metering and billing system based on cloud; collects user requirements, information of competitive products, responsible for requirements analysis, competitive... Read More →
Abnormal configuration detection is a very important activity in cluster operation, especially for cluster upgrade. However, with the evolution of Kubernetes, it has become very difficult to detect cluster abnormal configurations. This proposal presents a DSL framework for detecting Kubernetes abnormal configurations. The framework is made up of the reporter builder and the reporter execution. The reporter builder allows users to freely combine configurations which to be detected. And the reporter execution allows users to define the way to detect configurations using DSL. This framework supports a variety of data sources and collection methods. It can also be used with OpenKruise Broadcast Job to issue detection tasks, and also can be used with NPD. The DSL framework supports many kubernetes-based products on Alibaba and guarantees tens of thousands of clusters upgrade successfully.
Jing Gu is an engineer on Kubernetes Service team at Alibaba Cloud and is a member of Kubernetes. She primarily works on Kubernetes AIOps and cloud controller manager for Alibaba Cloud.
Jun Deng is a senior engineer at Alibaba Cloud. He works on container service products, mainly focusing on cloud native applications and automated troubleshooting service for Kubernetes clusters.