Both Apache Mesos and Kubernetes are excellent open source frameworks, and both support large-scale cluster management (of course, open source Kubernetes is currently limited to thousands, and tens of thousands of nodes still need to be customized, while Apache Mesos can lightly schedule tens of thousands of nodes), There are many mature applications in China. For example, NetEase Cloud and Huawei have deployed large-scale Kubernetes clusters, and iQiyi, Qunar, Ctrip, Dangdang, etc. have all chosen Mesos.
Generally speaking, if it is only used for container cluster management, Kubernetes is more suitable. If there are more customization requirements, or to build a big data platform, Mesos, which has a relatively loosely coupled architecture, is obviously more suitable. Of course, using Mesos + Kubernetes for container orchestration is also a feasible technical solution. It should be noted that both Mesos and Kubernetes require the team to have strong technical strength.
From the original intention of software design, Kubernetes hopes to become a leader in the field of container management, and AWS and Azure joined CNCF, and Docker officially stated that it supports Kubernetes natively, indicating that Kubernetes has an unshakable position in the container field with its excellent design from Google. The community and ecology are becoming more and more prosperous.
The goal of Mesos is resource sharing, allowing enterprises to put existing business loads, such as Hadoop and Spark, into a common management environment. As for whether to containerize, it depends on the demand for microservices and DevOps.
How to choose a containerized technology stack, NetEase Cloud Architects made a systematic review.
Thousand-node cluster, less customization: use open source Kubernetes (fine-grained design, in line with microservice thinking)
10,000-node cluster, multi-customization: use Mesos + Marathon (two-tier scheduling is sharp)
10,000-node cluster, strong IT capabilities: deep customization of Kubernetes (such as NetEase Cloud)
10,000-node cluster, strong IT capabilities: deep understanding of the use of DC/OS (DC/OS adds a lot of components on top of the most basic Marathon and Mesos)
Big data cluster: Spark on Mesos (it is recommended to deploy the computing part only based on the container, and deploy the data part separately)