That’s an excellent feature for k8s. Ruijing and Jack can help do it, and Sofia can help do some documentation for it .
Thanks Xuan Jia Project Manager Big Data & IT Technology Research Center China Mobile Research Institute 32 Xuanwumen West Street, Xicheng Distirct, Beijing 100032, China Mobile: (+86) 13811000575 E-mail: <mailto:jiax...@chinamobile.com> jiax...@chinamobile.com 发件人: Gaoliang (kubi) [mailto:jean.gaoli...@huawei.com] 发送时间: 2017年12月18日 15:39 收件人: 龙雨 <yl...@biigroup.cn>; opnfv-tech-discuss <opnfv-tech-discuss@lists.opnfv.org> 抄送: 付乔 <fuq...@chinamobile.com>; jiaxuan <jiax...@chinamobile.com>; Guo, Ruijing <ruijing....@intel.com>; Brattain, Ross B <ross.b.bratt...@intel.com> 主题: RE: [opnfv-tech-discuss] [yardstick]k8s HA test Hi Louie， Thanks for pointing this out. According to Yardstick Fraser planning, we are trying to implement more test cases for container testing, especially for K8S. In Euphrates, Yardstick already supports HA test cases for LXC scenarios which are deployed by OSA and ping test case for K8S. Now, it is the time to talk about HA test cases for K8S. >From my perspective , Killing container and Killing progress are different >fault injection type. One is progress fault, the other is container fault. >although I believe It make sense to test K8S with Killing container, there are >still some topics to be discussed, including “how to kill the container”,” >which containers we should kill” and “what metrics we should monitor” . Based on our previous experience of LXC testing, I’m not sure if it make sense to kill the container instead of progress. In some containers, there are more than one openstack progress (such as cinder, it includes cinder-api and cinder-scheduler), you would extend the fault range with the killing container. I would like to invite HA team and Container4NFV team to join the thread. I guess they could give some valuable comments from their perspectives. Just my two cents. Regards, Kubi From: <mailto:opnfv-tech-discuss-boun...@lists.opnfv.org> opnfv-tech-discuss-boun...@lists.opnfv.org [ <mailto:opnfv-tech-discuss-boun...@lists.opnfv.org> mailto:opnfv-tech-discuss-boun...@lists.opnfv.org] On Behalf Of 龙雨 Sent: Thursday, December 14, 2017 4:16 PM To: opnfv-tech-discuss < <mailto:opnfv-tech-discuss@lists.opnfv.org> opnfv-tech-discuss@lists.opnfv.org> Subject: [opnfv-tech-discuss] [yardstick]k8s HA test Hi，all， This is louie from BII-CFIEC，CFIEC is a third party neutral SDN/NFV technology evaluation and innovation center. Recently I have participate in OPNFV CVP test， as we know CVP certification use dovatil program and dovetail integrate other OPNFV test programs like yardstick，functest, etc. When we do CVP test for our custom we meet a problem in HA test, the current CVP test tool doesn't support k8s, but our custom use k8s to realize HA. Yardstick use ssh to login the controller node and kill such as nova-api progress to test HA, but due to the k8s feature the nova service exist in one container behind the k8s ， and yardstick tes tool cann't login the container directly, this lead to the HA testcase failed. I have feedback this problem to dovetail and CVP，according to the HA testcase I put forward one method to solve this problem，use kill container to instead of kill nova-api progress，as for haproxy， due to the k8s use nginx to realize LB we can use kill nginx to instead. Meanwhile, I ask dovetail whether they can offer a patch in dovetail to match k8s HA test, the reply me this patch is all for yardstick and may do many changes, so they suggest me go to yardstick progrom seek for help. I know yardstick maybe support k8s in F version，but it seems to long for my current condition. By the way, in my test SUT every controller nodes have a container run opensatck service, such as nova-api, glance-api. We can access one controller node to kill the container by specify name id or use k8s command to kill the container, When kill the container k8s will auto recovey the container on this node, and the recovery time is decided by k8s configuration. This email first I want to know whether this kill container methond feasible； second，whether yardstick could offer such a patch for k8s HA test. Thanks, louie ------------------ 龙雨 Louie Long 邮箱:ylong <mailto:xdzh...@biigroup.cn> @biigroup.cn E-mail: ylong <mailto:xdzh...@biigroup.cn> @biigroup.cn 手机：+86 13261979365 Mobile: +86 13261979365 传真: 86-10-5638-1682-162 Fax: 86-10-5867-8466 邮编：101111 Postcode：101111 地址：北京市经济技术开发区经海五路58号院数字工场5号楼2层 Add: 2nd Floor, Building 5, No.58 Jinghai Road, BDA, Beijing, China 天地互连信息技术有限公司 下一代互联网国家工程中心 Website: <http://www.biigroup.com/> www.biigroup.com <http://www.cfiec.net/> www.cfiec.net
_______________________________________________ opnfv-tech-discuss mailing list opnfv-tech-discuss@lists.opnfv.org https://lists.opnfv.org/mailman/listinfo/opnfv-tech-discuss
