Author | Gao Xianglin, Chen Jun, Sun Jianbo
Industry news
Recently, Gartner, a well-known international research organization, released the 2020 report on the competitive landscape of container public cloud, and Alibaba Cloud has once again become the only domestic selected vendor. The Gartner report shows that Alibaba Cloud container service has a strong performance in the Chinese market, and has a rich product form. It has a good technology development strategy in areas such as Serverless containers, service grids, security sandbox containers, hybrid cloud and edge.
On April 9th, the Cloud Native Computing Foundation (CNCF) Technical Supervision Committee voted that Dragonfly was officially promoted to the CNCF incubation level escrow project.
On April 16th (Thursday) at 19:00, the second phase of the SIG Cloud Provider Alibaba webinar will be held. This webinar is a self-run live broadcast. There are two main forms of live broadcast: video live broadcast + text live broadcast. Video live broadcasting will be carried out in Yunqi community, developer community, and developer group. Text live broadcasting will be combined with DockOne community for live text broadcasting in 10+ WeChat groups.
April 8 is the fifth anniversary of the establishment of PingCAP, and its development team also released the first RC version of the landmark TiDB 4.0 on that day.
Important upstream progress
Resolved the bug that when the client's watch timed out, it would relist all resources instead of rebuilding the connection.
Upgrade PodInfo to a first-class object of the scheduler framework through a series of PRs, which can improve performance and observability.
Resolved a bug in the CRD with scale subresource regarding the default value of the replicas field.
Add etcd WithRequireLeader option in api watches . Without this option, when the etcd cluster election fails, the watches facing etcd in the apiserver will always live.
Instead of KEP, it is proposed to use GRPC as a probing protocol, because grpc provides an official health check method.
Open source project recommendation
The Corefile supported by different versions of CoreDNS may be different. This project can help developers migrate corefile versions. It can be used in binary mode or introduced as a tripartite library.
This project is a Go language tripartite library, which mainly provides functions similar to commands in shell scripts.
In simple terms, it is based on a tool similar to helm template made by Kustomize, which includes many functions such as parameter setting, packaging, pipelined serial rendering, and custom function (mirror) rendering. Naturally complementary to OAM, both of which are direct products for the concept of Infrastructure as Data, kpt is the manipulation tool of Data, and OAM is the format of Data. Kpt can be used as a command line tool for rendering, packaging, and distributing parameters before the final OAM yaml of the output of the application, and can be used in combination with the GitOps mode.
krustlet is a kubelet project written in rust language.
Reading recommendation this week
The purpose of OAM and DAPR is for developers, especially enterprise-level developers, to design and implement distributed systems that are simpler and more efficient. They are located at two different levels in a distributed system: OAM is in the deployment state, which is used to describe the overall topology of the application and the deployment and delivery process; while DAPR is working in the development state, allowing developers to discover and Call external resources. In this article, Baihaishi made a richer explanation of the two, and also looked forward to the future distributed application form.
The article introduces three tips for reducing the size of Docker images with detailed use cases.
Alluxio is a distributed storage system centered on memory. In this article, the author summarizes the challenges of Alluxio landing in high-performance distributed deep learning model training scenarios and the practice of optimizing model training performance on the cloud. Further, the author introduced how to improve AlluxioFUSE's performance optimization experience in high concurrent reading scenarios from multiple levels. Finally, the author implemented the distributed model training scheme based on Alluxio optimization, and performed performance verification in the ResNet50 scenario with 4 machines and 8 cards, and achieved good results.
" Alibaba Cloud Native focuses on microservices, serverless, containers, service mesh and other technical fields, focuses on cloud native popular technology trends, cloud native large-scale landing practices, and is the public number that understands cloud native developers best."