beyond-the-basics-towards-making-thanos-production-ready-yun-sakaki-27dya-daepqu-thanosjun-chang-lao-dan-zha-mao-benjamin-huo-junhao-zhang-qingcloud-technologies.pdf

《beyond-the-basics-towards-making-thanos-production-ready-yun-sakaki-27dya-daepqu-thanosjun-chang-lao-dan-zha-mao-benjamin-huo-junhao-zhang-qingcloud-technologies.pdf》由会员分享，可在线阅读，更多相关《beyond-the-basics-towards-making-thanos-production-ready-yun-sakaki-27dya-daepqu-thanosjun-chang-lao-dan-zha-mao-benjamin-huo-junhao-zhang-qingcloud-technologies.pdf（36页珍藏版）》请在三个皮匠报告上搜索。

1、Beyond the Basics:Towards making Thanos production-readyBenjamin Huo,Manager of the Observability Team Junhao Zhang,Sr.SE of the Observability Team KubeSphere of QingCloud Technologies Agenda-Introducing Thanos-Introducing Whizard-The production adoption of Whizard in KubeSphere -RoadmapIntroducing

2、ThanosHow do you customize&deploy Thanos?Can I use kube-thanos?-Maintained by Thanos maintainers-Customize Thanos setup with jsonnet-Deploy with yaml manifests Cons:-Jsonnet is more developer friendly,not user friendly-The sample deployment is not for production:Only one instance of query,router,ing

3、ester,and store with one replica each.No compactor,queryFrontend The hashring contains only the default soft tenant -Too many efforts needed to make it production-readyHow do you customize&deploy Thanos?What about the Thanos Helm Chart?-Maintained by bitnami-Customize Thanos setup with helm values-D

4、eploy with helm Cons:-Too many values to customize-Thanos Stateful Components are scaled by CPU/Memory(HPA)-No Tenant related setting-Hashring is configured manually What else do you need for production?Whats still missing?-Create and maintain Thanos components and configs with ease like CRDs-Tenant

5、 configs can be simpler without configuring the hashring manually-Support deploying Thanos to mutiple K8s cluster-Evaluating recording rules for each tenant and remote write the metrics back on tenant basis-What else do you need for production?What if youve tens or even hundreds of clusters?-Using a

6、 single ingester to receive all the data from all the clusters is not feasible:The Ingester isnt more powerful than a single instance of Prometheus The Ingester has to be scalable to handle 10+or 100+clusters-The Compactor,Store,Ruler has to be scalable to handle 10+or 100+clusters too-Automatically