Nebula Graph NebulaGraph for Beginners, 30 Jan 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: NebulaGraph Deployment Options for Beginners

Description

In this video, Wey, the advocate, will introduce you different options to deploy NebulaGraph and guide you to choose the proper deployment way that suits you. Any questions about this, you can leave a message below or join our slack to discuss: https://nebula-graph.io/

A

Hello, everyone, I'm lisa. This is my partner. Wait hi.

B

Hi lisa hi, everyone.

A

uh We hope you have a nice day today. We will have a brief talk about deployment of a network rob as many people know. Metal props have a variety of deployment matters, and today we will have an overall introduction to them. Hope the video can help you find the most proper method.

A

Then let's go through it to a point highway. Can you have a brief introduction to the deployment matters.

B

I'm sure lisa, okay, we will start so we will start with this decision tree and before then. I will dive into the results on how many options we already have on the nebula community that you can use to deploy a network cluster. So all those green circles are actually our options like in docker. You can use dock compose or dock swarm in kubernetes you, you can use the nebula helm, which is a hem chart for nebula, and we also have a nebula operators which is kubernetes operators to spawn you, a network cluster.

B

You can build from the source code, of course, and after both uh the debian and rpm binary packages are provided by nebula community and there is a nebulous ansible project to help you use the ansible to spawn a nebula clusters. So how do we choose so many options of them?

B

So we can refer to this decision tree like if you're running them, containerized or not running on kubernetes or not. You want to use the crd or operator pattern or not like you prefer to use a banner package. Is this running on the intel architecture or not, or your random arm, 64 or mips machines?

B

So basically, for example, if your answer is, I would like to use kubernetes as the infrastructure, and I prefer to use crd. Then you will go for level operators.

B

If you prefer use the hem chart, then you can go for the hem chart, but it's actually not recommended, because the hem chart is not good enough for maintaining the stateful workloads on kubernetes and you you would like to running on a container but not on kubernetes. You can still use docker and if you're only running them on a single server, you can use the docker compose it's the best fit for your single machine, containerized away, and it's also the best fit.

B

If you want to try it for your first time and you already have a linux machine, yeah nabla, docker swarm is a project. You can run your container way other than kubernetes on multiple machine, but it's actually not recommended. But still we have this option and it's like if you're running your workload on your arm based machine and not on your uh container in a containerized way, you can build it from the source code.

B

So, basically, that's all of the options and with all this question to be answered on this decision tree- and I think that is off- of mindset. Yeah.

A

Okay: okay, it's quite clear for me, even though.

B

A

A beginner too uh so about this. I have a question. uh All of these methods have the same hardware specification requirements when deploying the developer.

B

um Yes, it's a good question because actually um there's not too much uh of the fingerprints on different, uh you know types of workloads, so basically your um hardware requirements uh only is is only really really to your purposes or your business logic or your data volumes. But one thing to mention is that actually, for example, in enable operators there are some control plan. uh Workloads uh introduced, that's uh not too much, but it will consume you some actual resources. That's one thing.

B

Another thing is that if you are running only uh for your test purposes on your small data volume, you can even uh bootstream on your laptop that is like uh full cpu cores or even vcpus and like eight or four gigabyte rams.

B

But when it comes to the production level, uh runtime, you you better run you better use the faster storage like ssd or nvme ssd for your volume, yeah, okay,.

A

Got it uh so you mean uh only ssd is supporting.

B

A good question: lisa, the answer is yes, and no yes, because in your production scenarios, it's highly recommended to use the ssd or even nvme ssd.

B

But if you only want to test it to try some basic queries uh to to play with nabla graph, you can run it on your very weak machines, like your laptop or even very uh old servers on your lap. That is only provide the hdd disk. You can boot up the network and you can run the queries, but please expect some unexpected behaviors, but please note that the nebula community has started some work on making the hsd behaviors um more stable, but um it's never recommended to use hdd yeah.

A

Okay, okay, I understand so. I also have another question: what operating system do we support now and how about running network graph on windows or mac os.

B

Good question, lisa technically synapograph, was only implemented based on assumption of the operating system of linux on various of digitals.

B

But ideally, if you want to run a nebula on your, you know on your laptop, which is running on linux windows or even mac os, so it can be done with the help of the visualization tooling, like a docker, desktop or hypervisors of virtualbox or vmware player. That's that's still possible.

B

um One thing to mention is that it's a good news that last month in our daily build in our master, which is after the 2.6.1 after that we have a nicely built for the arm based linux image. So in fact now I can run uh nebula on top of the on my and one macbook on top of the container, on top of the docker desktop with the arm-based docker image, yeah.

A

Okay, okay, quite clear for me and, as I know, many users will want to use the nanograph in the production environment and many people of them asked me about the deployment of the three workloads like graffiti melody storage team in a production environment. So above them is there anything special to uh attention to.

B

um Actually, if you're familiar with the architecture of naval graph, you will know the answer uh correctly, that you will know the answer. That for grafti is stateless, so you can have any of them and for three and for storage the, which is uh the storage uh backhand instance.

B

At least you will have three of them if your deployment in a distributed way, but you can scale out to any scale of in his instances of uh stories, d and for metadata, is recommended to have three of them: yeah, lisa, okay,.

A

Okay, uh I know uh so. I also have a lot of questions. I must be a talented after so when deploying a network in a production environment, is it better to deploy kubernetes or on the local server.

B

How do they differ a good question? Lisa there is, I think the answer is. It depends um the general argument between the containerized or not or even uh kubernetes are not uh applies to nebula. Here, I'm not going to uh talk about that, but uh it did. It generally depends on your infer or ops teams.

B

For example, if your ops team prefers kubernetes and like all of your observation info logging tracking monitoring, infra kubernetes base, then it's better for you to run among kubernetes, like you are using nav operators, but, for example, if in your infra of kubernetes the the persistent volume class the pvc provided is only it's not fast enough. Like you, don't you only have a slowness or hdd-based backhand storage.

B

In that case, you will have to run nebula uh in a non-kubernetes way, because in that case you can leverage your fast on your bare mental operating systems, but it basically depends on your situation and there is no big change, a big difference and basically they are the same very same core battery in both the container images and your binary, packets or the source code that you would like to build.

B

One thing I would like to highlight here is that our nebula operators is actually rapidly involved. These days like last month, we just released the support on the rolling upgrade and, for example, other features like auto scaling is still not yet implemented.

B

So it's more than welcome that you guys to keep an eye on our network operators repository in github and try all the new features and feed up feed feedback to us. This will be huge contribution to the community and that's it lisa. Okay,.

A

Okay, quite an amazing explanation, so about today's discussion. It will be over now and we hope you can have some understanding of all the natural graph deployment options. So you can choose the right method that gtu.