Ali sister REVIEW: Ali cloud has succeeded in large-scale landing cloud native, KubeCon General Assembly on the 26th on, CNCF TOC and Ali cloud senior technical expert Li Xiang keynote speaker, shared Alibaba scale expansion, reliability, development efficiency, migration experience in strategy, and to explore cloud native landing and respond to a number of technical challenges.
Why do cloud native? What can cloud native what value? From the initial groping alone now embrace open source community feedback, Ali Baba gone through what kind of native cloud journey? What are the technical experience there? Today, all share it.
Years of precipitation, adhere to explore with both landing
Alibaba From 2011 technical native container system by cloud practice in the industry are not great examples for reference back any territory, and gradually worked out a set of container par tier global technology and service companies throughout the Ali Group infrastructure architecture. While exploring this lonely journey, but was consistently adhere to so far. It is at this technique to explore and forge ahead in desperate process, Alibaba's technical team of experienced complete all the critical nodes of the original cloud technology wave in the raw, not only the technological revolution has become an important witness, becoming China one promoter and leader worthy of the native cloud technology system.
Ali's body mass, business complex, pushing cloud native to find the right entry point. Driven by two-eleven cost pressures, resource efficiency and cost optimization has become the starting point cloud Ali native.
Ali Starting from the container, research and scheduling of low-cost virtualization technology: to provide a flexible, standard deployment unit; replace the static resource allocation is dynamic scheduling on-demand, to further enhance the efficiency of the deployment, solve the problem of resource fragmentation, improve deployment densities; by storing virtual network computing and storage separation techniques to enhance the mission's mobility can further improve the reliability of resources, reducing resource costs.
Driven by resource costs, Ali completed a comprehensive container, resource allocation is also efficient dispatching platform to take over. Ali cloud native did not stop there. Improve R & D efficiency and speed up the iterative cycle is the secret weapon to promote Ali business enhancement. Ali hopes native cloud lets developers more efficient.
In order to reduce application deployment easier and increase the degree of automation to deploy, Ali began to use as a container Kubernetes orchestration platform, and continue to promote the performance and scalability of Kubernetes. Specifically Kubernetes, Ali continued to develop, deploy process improvement. In order to build a more cloud original biochemical CI / CD, further to achieve standardization and automation, from development to on-line processes, Ali introduced applications such as standardized management Helm also tried GitOps deployment process like this, also contributed for the PaaS layer automation of the final state. Meanwhile, Ali also begun to explore the service grid, we are committed to further improve the standard of universal service governance, reduce developers to adopt the threshold, to further promote the popularity of micro-services in multiple languages and environments.
This year, Ali has launched a cloud over the whole station. After exploring the transformation of cloud native, Ali infrastructure system is modernized and standardized. The use of container technologies, applications and host complete runtime decoupling; use Kubernetes abstract Pod and Volume, etc., completed a variety of resources to achieve harmonization; through intelligent scheduling and PaaS platform to automatically migrate applications, do not fix stabilizing factor as possible, by Ali cloud native technology greatly reduces the difficulty of the cloud.
This increase in resources and personnel efficiency of the process, Alibaba's entire infrastructure has become more open, open source ecosystem connectivity, constantly absorbing and contribute good ideas, technology and ideas in the exchange interaction. Today, Ali cloud not only supports China's largest cloud-native applications double 11, and has the largest public cloud clustering and mirroring warehouse. As the only public cloud competition of container services firms selected to Gartner, Ali cloud also accumulated the most abundant and valuable customer practice.
Pursuit of perfection, optimization and scalability scale
Flexibility and scale, this is a complex scene and traffic peaks key factors supporting various types of Ali Baba.
After continuous grinding, Alibaba made on Kubernetes scale and performance remarkable results: the number of storage object 25 times improvement, the number of nodes supported upgrade from 5000 to tens of thousands, from end to end scheduling delays 100ms becomes 5s, etc. Wait. Many of which work together to carry out in Alibaba and communities, and these developments have contributed to the community, and we expect other companies and developers can also enjoy Alibaba scale technology brought dividends.
Ali Baba continued to optimize performance, it can be divided into the following four dimensions: workload tracking, performance analysis, customized scheduling, large-scale distribution mirrors. First, a complete track, the playback mechanism for workload scheduling, will be followed by a detailed analysis of all performance problems, one by one to overcome technical bottlenecks. Kubernetes itself is highly customizable, Alibaba for their own business scenarios precipitated customized scheduling capabilities and mirroring distribution system. Dragonfly open source project born out of the two-eleven, with strong distribution capabilities mirror. Dozens of super-clusters, each cluster has a super tens of thousands of nodes, millions of containers.
Alibaba landing Kubernetes can be divided into three stages: first, to provide resources supplied by Kubernetes, but more interference but operation and maintenance processes, which is rich container vessel system, the mirror standardization and lightweight virtualization capabilities to the above PaaS platform . The second step, operation and maintenance of the transformation process PaaS platform in the form of Kubernetes controller, PaaS to bring more automation capabilities for the end state. Finally, the operating environment and other traditional heavy models into native container and pod lightweight mode, while PaaS ability to fully handed over to Kubernetes controller, to form a complete system architecture cloud native.
How to solve the key difficulties in cloud native
Alibaba Cloud explore native, since the inquiry started in the container and scheduling system, to now embrace open source technology standardization. For the moment the developer of advice: If you want to build cloud-native architecture, it is recommended to start directly from Kubernetes. On the one hand, Kubernetes born as a platform builders, has become the mainstay of cloud native ecology, which not only shields the low-level details down, up and support a variety of surrounding business ecosystems; on the other hand, is more important is that the community has increasingly the more open source projects Kubernetes around the building, such as Service Mesh, Kubeflow.
So as someone who, what Ali "to avoid the pit Guide" have it?
Native cloud technology architecture evolution in the most difficult challenge, in fact, comes from the management Kubernetes itself. Because Kubernetes relatively young, its own operation and maintenance of ecological management system is not perfect. For Ali, the tens of thousands of cluster management is critical that we explore and summarize the four methods: Kubernetes on Kubernetes, use K8s to manage K8s itself; node issues a rollback strategy required by the rules of gray release; the image segmentation environment, divided into analog and production environments; and the lower side of the foot in the monitoring effort, the cartridge Kubernetes whiter and transparency, early detection, prevention and solve problems.
Another key technical problem is Kubernetes multi-tenant management. Compared to the poor scalability limitations namespace and naming conflicts, you can create a virtual cluster on Kubernetes. While improving scalability, isolation can achieve strong API level, virtual clusters and clusters by real syncer link to add agent in the node, to achieve a better and more rental management and better utilization.
The KubeCon conference, Ali cloud heavy announced two projects: Cloud Native App Hub - Application Center for Kubernetes manage all developers, OpenKruise - Kubernetes automation open source project sets from the world's top Internet scene.
OpenKruise Open Source Address
https://github.com/openkruise/kruise
Cloud Native App Hub
https://developer.aliyun.com/hub
Cloud native application center (Cloud Native App Hub), can be simply understood as Helm application China mirror sites, user-friendly obtain application resources, and greatly simplifies the steps Kubernetes deploy the installation of an application; OpenKruise / Kruise project aims to become a "cloud native applications automation engine "to solve large-scale application operation and maintenance of many pain points in the scene. The salon debut, developers have experienced rapid application center cloud-native applications downloaded from, and pod-place upgrade with state, sidecar vessel injection, the three warm-up mirror node scenario, the actual experience of a powerful automated operation and Kruise peacekeeping capacity.
It is worth mentioning that, OpenKruise project comes from large-scale application deployment Alibaba economy in the past years, publishing best practices and management; container platform team from Group operation and maintenance of large-scale applications, large-scale establishment of capacity; Ali cloud Kubernetes demand from service thousands of customers precipitation. To solve the problem on Kubernetes automation applications from different dimensions, including deployment, upgrade, elastic scaling capacity, QoS regulation, health checks, migration, restoration and so on.
Original release time: 2019-06-27
author: Future has come
from The Yunqi community partners, " Ali technology " for information may concern " Ali technology ."