GitLab GitLab and AWS, 3 Aug 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Overview GitLab AWS Implementation Patterns

Description

Table of Contents:
00:00 - Intro
00:59 - How Implementation Patterns Complimentin
05:18 - Pattern for Cloud Native Hybrid AWS EKS
12:55 - Resources

Resources:
- Explainer Video: Gitlab Reference Architectures, Cloud Native Hybrid and Gitaly: https://bit.ly/37e3c44
- GitLab AWS Implementation Patterns for GitLab Cloud Native Hybrid: https://bit.ly/3xhyJww
- AWS Quick Start for GitLab Cloud Native Hybrid on EKS: https://bit.ly/3lej9j4

Play Lists:
- Learning to Provision the AWS Quick Start for GitLab on EKS (https://youtube.com/playlist?list=PL05JrBw4t0Koi8VBnoVhmj_MstnbJjGw5)
- GitLab and AWS (https://youtube.com/playlist?list=PL05JrBw4t0Ko30Bkf8bAvR-8E441Fy2G9)
- GitLab Alliances Tech (https://youtube.com/playlist?list=PL05JrBw4t0KqaWCrU4avIY6TQveKTDMso)
- GitLab Reference Architecture, Cloud Native, Performance Testing, Scaling and HA (https://youtube.com/playlist?list=PL05JrBw4t0KqgVugqCld2fUvrBulanchq)
- GitLab and Kubernetes (https://youtube.com/playlist?list=PL05JrBw4t0Kqp16k3nHTMdrWdATrcDZ5C)
- DarwinTheorizes at GitLab (https://youtube.com/playlist?list=PL05JrBw4t0KoH74CcAtP0Ze7JXjX7zT6g)

A

Hey everyone, darwin sonoy here, I'm a staff solutions architect with git lab alliances. I want to take you through a few things today about get lab, aws implementation patterns before we dive in. I want to point out that this other video is really helpful background. So what I won't be talking about is the concept of gitlab reference, architectures cloud native, hybrid reference architecture and gitly. All of this is explained in this other video. That's about 11 minutes long. So I'd encourage you to take up that video as it'll help.

A

You understand a lot of the concepts. Also, I just want to quickly explain the term cloud native hybrid. We call it cloud native hybrid because we put as much as possible of gitlab into kubernetes, and then there is some of it that must run on instance, compute, essentially the gitly cluster and then the rest of it. We run on cloud platform as a service, so it's not technically all running in kubernetes, which would kind of be pure cloud native.

A

If you will so that's why it's called cloud native hybrid, I want to go down kind of a path of analyzing. uh The reference architectures that gitlab produces and implementation patterns and how they complement one another reference architectures are the baseline bedrock of how to deploy gitlab. They tell you a lot of information about gitlab running at scale, with guaranteed performance, that's tested with a performance tool and all of that information is covered in that previous video I mentioned implementation patterns, complement it by making it more opinionated towards a given platform.

A

Today, we'll be talking about the aws cloud platform. So, let's take a look at how this kind of breaks down in the area of compliance. The reference architectures are really the gold standard, so they are what we need to comply with. So the aws implementation pattern also then complies with reference. Architectures.

A

The compute requirements of reference architectures tend to be stated in vcpu and ram, so they're stated in general machine requirements. This allows them to be adapted to a lot of different deployment or implementation environments.

A

In the case of aws implementation patterns, we will be being very specific about what instance, sizes or sizes of things in the aws of terminology.

A

Our resource specifications in the reference architecture tend to be general, so when they're referencing a service, they might mention hey.

A

You need a load balancer instance, if you're doing this on hardware or certain virtual infrastructures, or you could use a reputable platform as a service, of course, in the implementation patterns, we'll be referencing very specific aws specifications for the same kind of thing as far as clusters, the reference architectures will be thinking in terms of kubernetes in general, for any cloud native, hybrid components and, of course, in aws we will be choosing their managed kubernetes, which is eks.

A

As far as internal encryption. The reference architectures tell you that you can use ssl certs throughout your instance. In order to have internal encryption and in the aws implementation patterns, we direct you to specifically use acm and kms.

A

Now, of course, you could even when you're implementing with the implementation patterns you, if you already have certificate authority, set up or some sort of encryption keys, then you could use those instead, when it comes to high availability uh services such as postgres and redis, the reference architectures will tell you how many nodes you need for implementations that are not on platforms as a service. They tell you about other cluster coordination roles.

A

You'll need with amazon, we'll be directing you directly towards rds and elasticash, because these services already have a lot of those ancillary services built in for coordinating the cluster. So you don't actually have to deploy separate resources for those when it comes to giddily giddily is the get repository storage subsystem of git lab with the reference architectures they'll tell you that hey you need three nodes of italy. uh You need to configure them with prefect and postgres sql database in order for it to coordinate the cluster activities in the implementation pattern.

A

We tell you specifically to use an asg elbs and then we also talk about potentially ebs being provisioned with iops. So we'll talk about some of the options of the actual local storage of those nodes in the case of reference, architectures there's some things that aren't highly opinionated, because there's not really a lot of bottlenecking created for the entire instance for them, so email smtp is an example of something that the reference architecture might mention need to point to one in amazon.

A

We will point you to a scalable platform as a service wherever amazon has one, and then there are some implementations that really aren't reference architecture oriented. So if you think about setting up a small per user training setup, each user has their own gitlab instance. This is something that's really out of scope for reference architectures.

A

However, with the amazon implementation patterns, we might help you specify interesting or novel structures or automation in order to facilitate that kind of pattern on the platform.

A

So that's kind of an overview of some of the differences between what a reference architecture is and how implementation patterns take and build on that to create value within a certain platform. Let's take a look at one of these, so this is the get lab implementation pattern for provisioning cloud native hybrid on aws eks.

A

It allows you to understand what can be done in the alliance's team. We have taken and tested this on aws, so this table is kind of showing you how this breaks down over here on the very left column is the get lab cloud native, hybrid reference architecture.

A

So we'll just take a look at the 10k one here within the regular reference architectures over the last six months, there has been a special section added called cloud native, hybrid, and so our cloud-native hybrid reference architectures are really just a special section within the regular ones uh they specify cluster topology and then, after the cluster topology, they specify all the other services you need and notice. These little numbers here indicate what can be run on platform as a service, so they try to tell you that hey.

A

You know these three roles here console postgres, sql and pg bouncer could be run on platform as a service. If you have that available in your platform, so we have the link to the actual reference architecture.

A

We then have the baseline testing, so all of our reference architectures are baseline tested and if we take a look here, we can see the actual testing numbers for 10k, and so this is the baseline anyone else, who's trying to create something similar would be trying to meet. So we want to meet this same baseline and so I'll show you in a minute that we are testing the implementation patterns the same way.

A

So you can compare them to what the reference architecture testing does and then, if you do any tweaks, you could run the gpt gitlab performance tool on your own instance and see if you too, can meet whatever it is. The reference architecture is doing as far as baseline performance uh we'll take a look at bill of materials in just a minute. That's just a link to this same page that we're already on- and here you can see that we have the gpt testing results for the actual cloud native as configured here on aws.

A

So, basically, the same test results. Drill down on 10k- and we can see right here- that we have the same test results so you're able to compare to the uh reference arc baselines. We have the aws calculator available with the particular instance, recommendations that we're making below now everywhere else that we do these recommendations.

A

We only use the on-demand price. So when we, when we look at the tables below you'll, see that we use on-demand prices when we say how much you know, your rds would cost you per hour, but with these cost estimators we are using a one year, compute savings plan on ec2 and one year on rds in elasticache.

A

You should note that once you load one of those, you can make changes and then save it for yourself and grab the url at the end. When you save it, and so you can customize it so you can use those cost estimators to to customize for your own purposes.

A

I'm going to proceed down the page here, a bit and we'll look at aws platform as a service qualified for all get lab implementations.

A

So what this is is all the platform as a service pieces that we're allowed to use and be cloud native uh reference architecture compliant are outlined here, some of them. The reference architecture talks to directly so like some of these load balancers, but some of them it doesn't so it doesn't. The reference architecture doesn't really care what you use for dns.

A

It doesn't necessarily speak to log aggregation other than you probably should do it and then, of course, gitlab does prometheus for a lot of its internal metrics. But you can see here that there are supplemental services as well that are handled. So these are other things that you can use in concert with the regular infrastructures code.

A

Let's go down and take a look at the 10k.

A

We come down to 10k the test results are repeated as well and then, when we come down to the bill of materials, so it tells you here that this is a bill of materials and what we've done is stuck to the vcpus and memory like the reference architecture does. As far as estimates.

A

Whenever there is something cited here, you can see underneath there's a link. So when we say hey, we want 20 pods. If you click this link, you can find out where we found the information about 20 pause and then 5, vcpus and 6.25 gigabytes of memory per pod, and once again you can find the reference information there. We get the total that we calculate here. We do the same for sidekick. The reference architecture architectures also have some general overhead built in, and so that is cited right here.

A

We then come down and have a reference arc raw total. So um that is the raw total that the reference architecture would dictate for the implementation pattern. We build in some overhead I've chosen to build in one node of overhead, and this is because we have things like the eks cluster, auto scaler in there and there's other miscellaneous things running in there. That eks will deploy to make it be more of a managed cluster for you.

A

We then get to that grand total, cpu and gigabytes of memory. We then want to choose an instance type that kind of blends this fairly smoothly. If you think about having to scale you don't want too big of an instance type or else every time you scale you'll only be using a small portion of the next new instance that comes online, so I've chosen the c5 4x large, which divides us into nine nodes. You can see that we can't make it perfectly we're nearly perfect on vcpus we're.

A

Only two over allocated on memory of substantial over allocation gitlab is very upfront about what the pods actually require and that doesn't always fit nicely into node sizes. So you end up with some over provisioning.

A

We also outline a possible idle configuration, so the reference architectures allow for a 75 scale down of pods, but they don't really speak to scale down of nodes. This is a scaled down of nodes.

A

The pricing here is based on on-demand pricing, so that is the kubernetes portion for the non-kubernetes portion. We detailed the platform as a service that is instance type dependent. So you can see here. We have postgresql we're also been testing the graviton instances, since you should get better bang for your buck and because it's a platform as a service gitlab doesn't even know that it's graviton or arm. In the background you can see here, we've tested a graviton and arm instances.

A

You can see the raw totals from the reference arc and then how the instance sizes divide that out and then you can see node pricing, then. Finally, we get down to the gitly cluster, as explained in the previous video gita cluster is the way that we store our git repository data. All other storage in gitlab can go to s3, but the git repository has to be on local storage in this cluster, and so it tells you what what to pick for that.

A

That's kind of the overview on how reference architecture is an implementation patterns relate together and what is in the implementation pattern. The implementation pattern is meant to complement the reference architecture so that anyone who wants to do this on aws has kind of a bill of materials and some tested places to start so just quickly. Some resources that will also be posted to the youtube description, so you can click them directly.

A

These are also just pages that we've been going through already. If you need any help with gitlab feel free to reach out to our sales department, they can get you wired up to get lab to try to make sure that you're doing gitlab in a way that will serve you best at your organization.

A

So, thanks for your time and attention.