Cloud Native Computing Foundation KubeCon + CloudNativeCon China 2018 (Shanghai), 21 Apr 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes CI/CD Practices at Meetup - Rui Chen, Meetup

Description

Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io

Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects

Kubernetes CI/CD Practices at Meetup - Rui Chen, Meetup

The presentation would be like case-study style to introduce the Kubernetes practices @ Meetup around CI/CD. The agenda of the presentation would be divided into: * Current infrastructure status @ Meetup * Configuration stack * Deployment flow and process * Challenges with managing things at scale * Road ahead

For more info click here: https://sched.co/FuJq

A

Hi everyone I'm reaiiy, walking at a meet-up. I will go to the structure engineer today, I'm going to share some stories that we do around coven, NEC SED practice at the middle. This is like a nun user perspective to deploy application so before I jump into my story, I want to ask a simple question, which is actually a quote us SED. So when you push you a single line of code change, either configuration change or coach edge or just a typo to production.

A

How long does a tail so for us, we used to take seven days includes a bug fix were feature release. Now we reduce that into one hour, so it is pretty impressive. So here's our journey, the practice we have down.

A

We used to have a moralistic application, just like every China avocation has mixed up with the front end and back end and the operational code JavaScript so much so force so which ones we refactor our architecture into a macro service based solution and it works pretty good so and we use do have a silo teams by seller team. It's like different functional groups, infrastructure team, product engineering, team platform, team, different teams. They have a different purpose, different prioritization, which slows down stands. That's why we have seven days to release a single I, encourage we.

A

We actually do actually a breakdown infrastructure, team and embedded each engine infrastructure engineer into product engineering team, empower the product engineer aside to do more frequent release. So now we actually do continuous deployment to production. Every day every hour we will reduce release cycle by testing directory in production, which is pretty awesome.

A

We used to do reactive, we have our users could help us. We do good what we do bad now. We have a proactive mortar into a bunch of a loitering stack, stack and and a bead has them. So, let's first attach silo teams.

A

We have seen we used to have a single cluster runs on a cocoa cloud. Kubernetes cluster with different were closed, like they did haja briefly on machine learning inside, and we actually have machine learning. We have long-term teaming teaming jobs, we have shot him short ham, jobs running on a single cluster, with eco Picabo death based on the functional functionality and the product engineering leads, and now we we used to have one infrastructure team which actually has so many different kind of requirements. They get burned out. They get alcohol.

A

You have like 10 people on core theory middle of the night and nobody's have a good solution to troubleshoot in a problem. The capital cost will make us move faster and make sense more efficiently by deaf dialogues, culture which keeps us we actually, the industry has coined a term like DevOps engineer, full stack, engineer, Amitabh used to we have a developer, we have QA, we have Aboriginal engineer. Now we actually transform each of these responsibilities into just a win. One single step is called teacher of engineering. Some some companies called a generalist engineer.

A

Like I said we do. We do continuously deployed deployments in the production right now. So, let's check out the OCS ad pipeline. So, like some of loudspeakers has talked about this kilowatts flow. We adopted a kilo of the flow as well like will. We have a macro service, repo, our configuration, the the kubernetes configuration llamo first side by side without stellar Escala, micro-service repo. We have. We have a temporary repo which has many best practice the conventions. So basically you have a new service you're.

A

Speaking of a new repo, make some necessary changes to make this service more to the needs of the requirements, and basically we will use a kidnapper our version control system. We have a PR, we have lending. We have this code review process once the wescam urge we trigger a master Butte I'm, pushing through the container registry. We have we're running our code, lettuce, cluster right now and we'd obvious. We push a registry to UCR and then the kinetics cluster put in an image from the UCR and deploy your application.

A

So basically you can say cute steps. One is built up, the other one is deployment, I'm testing production, so maintenance is very hard. You have many of the people they have to deal with. The akima latest version upgrade right now, the community, a version upgrade, is really fast. Six six months you have a new release and the kinetics is actually assisted. By said, with the cloud provider, was a cloud provider has its own virtual?

A

For example, EWS has its own. You see two instance upgrades as well, so we have to constantly deal with this. The maintenance and is it is pretty hard to do. Keep up keep up. Velocity I mean he maintained. Different environment is very hard to like you have qmm, you have staging RAM, you have pre pro environment.

A

You have to constantly worry about the data refresh to your pre station before have a surprise in in production, I believe everyone, many of you have a similar disaster experience before as well, and sometimes it's not reliable, it could be a do to learn. Network negative issue is just so flakiness and you couldn't reproduce and for some time you just give it out and testing production is a truly continuously delivery.

A

It allows us to move faster than ever, but in a mean way we do care about our users, our members, so we do have a certain certain sense to car safeguard our release, which is continuous delivery. We have a feature flag. Our feature flag is our feature flag is your user century. We can determine a user who can access this feature or not? And what so? We can refer back pretty quickly as well.

A

We have been released so for each of so for you to have a set of features within a feature flag and with Pina release in batch, and we have a be testing to ensure the product is actually has a positive user impact on the user side.

A

The last thing I want to talk about is a proactive or moderate. So right now we're talking about macro service architecture, usually in practice the world helps send this pipeline clcd pipeline replicating to us ReWalk boland, each of the macro services. You have the constant need to monitoring the buter, successful failure and deployment success deploy video, so we actually find out Wally's messages in queue.

A

Slab I also have this across teams monitoring as well, so that every team's can cross-check the says the deployment status so that the one yes actually has is actually a transparency across the team and also monitoring external dependencies. We using the SAP solutions will use a kid hub. We use the Travis will use either base, so we need to care, we need to care about their status to so also. We we also advocated layout status information to our a snap channel as well. So that's our full story. Thank you.