Red Hat OpenShift Databases - OpenShift Commons Gathering 2022, 23 Feb 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Deploying Databases using OpenShift Virtualization - Cem Omurtak (Sahibinden) OpenShift Commons 2022

Description

Sahibinden's Cloud Transformation: Deploying MySQL, Cassandra, MongoDB using OpenShift Virtualization

Erkan Ercan (Red Hat)
Cem Omurtak (Sahibinden)
Jehlum Vitasta Pandit (Red Hat)
OpenShift Commons Gathering on Databases held on 02/23/2022

Slides: https://bit.ly/3HnII8P

Join OpenShift Commons: https://commons.openshift.org/index.html#join

Full Agenda here:
https://commons.openshift.org/gatherings/OpenShift_Commons_Gathering_on_Databases.html

A

So we have another very interesting customer session uh with sahih bindan next up and sahih bin, denis turkey's biggest online marketplace. They have a very unique use case and they're deploying several workloads, including databases such as mysql, mongodb, cassandra, click house and apache kafka android at openshift, and this supports their whole website and handles all the traffic um that they have received on their website.

A

We have jim omartak, who is a software and data architecture manager at sahih, bindan and erkin, who is a senior solutions architect at red hat and he helps our customers in their app development journeys uh and they're here to shed more light on the sahibundan story. So hi welcome to the comment section and thank you so much for being here.

B

Hello, everyone.

C

Hello, everyone.

A

And so I'm going to so and yes, so you can see their profiles up here and we can just dive right in. I won't waste any more time. um So can you can you tell me a bit about sahih bin and what does sahib indian actually do.

B

Five names a classified platform, this fights platforming from turkey, a diverse spectrum of new and second, and which is sold on the truck platform like real estate, cars cranes, clothing, food with close to 60 million active users, 30 million pages per month. Fiona is the fourth largest online online classifies platform in the world, trailing on the craigslist in usa, between russia and the bomb in france.

B

It's also one of the topics the size in turkey, according to alexa current. This is close. Two thousand employees, 250 of which are working in uh randy department side, is currently running on more than four four thousand pms, more more than uh 150 distinguished distributed between two on-prem sites on a public cloud site on gcp.

B

uh That's a rob! So thank you.

A

Okay, so, as I understand you decided to undergo a massive cloud transformation project, what were the challenges that sahib indian was facing, and why did you decide to go for such a transformation project.

B

Inside vietnam, we first wanted to modernize our underlying technology stack of the services we provide enabling product with scalable can raise reusability improvements as well as we want to reduce time to market optimize return of investment and stop establish a more secure and valuable technology infrastructure to our understanding, every company needs to be a technology company and should provide its business good needs to deliver new services to the customers in a reliable, safer scalable and a fast way.

B

We also want to employ top talent which driving companies with up-to-date data, take the up-to-date technology stack.

B

Although public providers uh check all these boxes, we are subject to stick governance regulations in the form of a local version of gpu gdpr.

B

That means in some going point in the future. We may have to suddenly abandon abandon our public cloud as extensions, so we are kinda stuck with on-prem uh data centers uh before before 2020. We have built up upon our our virtualization of an automation platform around debian linux, and then we come with some homegrown python based infrastructures code tooling. We named simon and cloud systems uh when we need that that kind of infrastructure, uh the tools, the modeling tools, are not very major actually and this.

B

This tooling is, although very customizable, because it's written by us uh has a very steep learning curve and do not really attract people who already learned and used open source tools for that purpose.

B

So, in fact, we wish to minimize the footprint of self-provision, operated security and government technology components and focus our efforts, engineering, careful effort to our application around data layer, but we also envisioned that the actual containerization of of all our applications is will be a multi-year project and we decided to uh instead approach the we decided not to make a big bank actually, so we decided to go with the lifter shift approach, where we keep running our workload on virtual machines and focus on replacing the underlying platform.

B

First, with these ideas and restrictions uh in 202- and we started a two-phase project in the first phase, we moved a part of our workload to google public plot, where we diverted 20 percent of of our traffic to keep it alive, and we actually actually liked our experience there, because they are, we are not bound by any kind of hardware restrictions.

A

Yeah so, like you mentioned so for the second phase of the project, you launched a request for proposal. What were your requirements and your selection criteria for that.

B

In the second phase of the project, we tried to extend the benefits of clarification of our application services to on-premise data centers by employing a private cloud solution that mimics a public cloud experience as much as possible. I mean the limits of technological and financial feasibility and we therefore released nrfp to private cloud vendors. Many private plot providers, including, of course retired, responded to the rfp, putting cybernet specific details aside.

B

I think the rfp, the I guess in the rfp can be summarized as follows. Firstly, we we needed a major platform which regard both uh virtualization and continuization has heard first-class citizens. By that I mean we should be able to work our workloads flawlessly on uh on both virtualization uh part and quantization part, and their connectivity should be also seamless.

B

This kind of technology will always allow us to shift our applications from traditional vms to containers in our own pace, because there's a big transformation project- and we can't you know close the shop- and we are doing this- probably can't do that. So we need time for that and without a big bank, such a platform will will enable us this transformation be as flow as graceful as possible.

B

We also are looking for a platform where we can apply their second principles, strengthen our security governance, as well as improve our operational efficiency.

B

At the same time, we are trying to change the culture in the company. Like development teams has stronger autonomy, the teams are going to be not only responsible from development of their products, but also its architecture, scaling, monitoring, etc.

B

They should be able to control uh production concerns like number and composition of pots, auto scaling policies, selection of databases and other subsystems and test them with these by selecting them from application catalog, and then push those developers and changes to production with minimal infrastructure, team involvement.

B

We are also trying to avoid van der lucking, while we can keep the symmetry between our own from uh data centers and cloud data centers. uh This is also very important for us uh also. uh Lastly, uh to our experience, uh operation of large, open source. Kubernetes installations are not really three will work, especially the upgrade yeah. You can get it work, but the upgraded really upgrade part is really painful uh by upgrading different parts, you can easily end up with the configuration that has never been tested anywhere as before you, so you may be.

B

You may be in dark, so we need them vendors, which we can trust for support and also we. We are sure that they have released a very well tested bundle to us. This is the essence of directly.

A

Okay, so I think now we have quite an in-depth understanding of the challenges and what you were looking for. So can we get a little bit technical and see what the proposed solution architecture was and how it essentially mapped to what sahih bindan was looking for.

C

Yeah, actually, when we got to the rfp and analyzed the requirements, we thought uh openshift on openstack based solution, supported with sf storage, can be a good fit because uh most of their workloads were running on vms, and so uh we built a solution based on that. But uh after we presented this to the sahibundan uh saibindan cto stated that they were looking actually more hyper-converged and the unified solution to run containers and virtual machines together seamlessly.

C

So they don't want to invest on two separate stacks for integration, migration and maintenance. So this became an additional effort for them. So that's why we changed our solution to open shift virtualization based, so we offered them a bare metal open shift, uh supported with openshift data foundation, uh to run both virtual machines and containers, together with the support of openshift uh virtualization.

C

So uh when we propose that they actually the cybernet team liked this one, even openshift virtualization was a fairly new technology. At that time uh they we decided to go on this solution.

A

Okay, and so why did cybenin choose openshift and openshift virtualization uh to get this massive goal over the finish line.

C

um uh As jim mentioned, actually they were looking for a stable and the security focused uh comprehensive platform uh to run containers and virtualized applications together.

C

So I think openshift was a very good fit for that, and so we were a leader in these enterprise kubernetes, and so it was very popular in also in turkey as well, and jim also mentioned, as you remember, they were trying to improve their developer productivity change, the culture in so they can able to try and test new applications very easily like in a cloud environment. So I believe these were the you know, pillars that uh you know. Cybinding shows us.

A

Okay, that's very interesting, and so what's the what's the current architecture, can you talk us through that? How is the traffic being distributed? um How are you delivering this project because it's so nice.

C

Yeah, okay, let me start with how we deliver this project very quickly, so we started the project. On february 2021 uh we made the bare metal openshift deployment in cybind and ankara data center, uh so uh cybin dance cloud system. You know they were using the provision, the virtual machines and the configuration other infrastructure components, so they integrated their cybin and cloud system with using openshift uh virtualization apis.

C

uh They did it very quickly actually, after that they migrated more than 1 500 virtual machines. So these workloads include uh this some java applications, saving them api gateways, even active directory and many stateful applications. I think that was the topic of our today's session, so um I think jen will mention much more, but uh we onboarded mysql servers, mongodb servers, cassandra and kafka clusters so and as so on april, this went live and after that these clusters started to get data synchronization traffic from the other data centers.

C

After getting that uh cybin done, team uh actually rooted the start to root some image and content traffic, which is the majority of the traffic happening inside bindan. You know when the people upload their uh pic photos of their houses cars. So these are kept in a kind of a cdn in private cdn network. So uh these these kind of things were started to serve on openshift but and on july 2025 uh cybinder rooted 80 percent of their whole site traffic to openshift.

C

Now so now we are working on uh the new other data center in istanbul, so openshift is set up there and cybernine team is migrating that one as well as currently we, as you can see uh side is. Eighty percent of cybernet traffic is uh uh route to uh openshift. The clusters in ankara and 20 percent is going to do um google cloud. uh They are currently istanbul data center uh legacy, istanbul data center is decommissioned, and so we are now modernizing the the other data center. Now.

A

Okay, and so where do we see this going in the near future,.

C

Yeah, I think in in a short time uh this uh the cybin team will complete the migration of other applications in uh to to the istanbul data center as well, and after that, each openshift cluster will be handling around 40 of the whole site traffic. So, in the case of an emergency, they can able to switch traffic very easily between sites and after that we will be starting on actually modernizing the applications and onboarding containerized applications on openshift.

A

So now can we talk a bit about the workloads as well, that are being deployed on openshift? What are each of the components being used for and what applications are being supported.

B

Well, psy been done, has.

B

Monolith, a very large one like millions of code, which is that back-end business services and two two more applications and they are a bit more lightweight the rallies resource access, our api gateway and bitmain is our web application.

B

These applications by themselves span like close to 600 machines on each data center and majority of the application lies there. There are many small micro services all orbiting this, this environment, but majority of tumor work is there and what we did was.

B

As I said, uh we, we don't want to tackle two problems at the same time uh and we just did a different shift operation where we moved. They were already running on.

B

Virtualization for middleware, we use mainly memcache monkey. We have to use because we haven't used cash, uh it's our application, cash and as our manchester merchant, we have an internal stevia which mainly runs varnish and it serves.

B

It contains in sketch, like close to one megabyte image. We also have to use elasticsearch because we have to search stuff in the classified in our classified site. So at least 100 vms are different. Elasticsearch clusters we heavily, we again have to use kafka for even streaming and inter process messaging.

B

As far as databases, we we we use mysql, it's the one with absolutely we treat as absolute source of truth, because uh we also have because it's transactional, we all say longer by the time this decision was made, doesn't have any transaction support so uh more like uh ephemeral data like login information or the the kind of data.

B

If we lose, we don't cry that kind of data lives in, but it's reliable. Actually in click house, we contain user activities, it's kind of analytical uh database, uh but it can handle very big queries on really fast and you can actually connect your front-end application to uh and it's much faster that we got much faster than when sql in those type of situations, and we also have used cassandra.

B

It's the impression count when somebody clicks on on unclassified.

B

People can track how many clicks made on them, etc.

B

B

Applications run on their own vms, on uh let it open shift visualization at the moment, and we are in the process of uh modernizing the now. It's it's time to modernize the applications of application themselves. uh We started uh recently microservices project where we attacked all monoliths to uh models, and you know separate to smaller parts. So so we can run them on virtualization on uh contain containers instead of virtualization.

A

Okay, that's very interesting, and so uh what? What was your experience like of deploying these workloads on openshift using openshift virtualization? How did you migrate them on um how? How was it like.

B

uh Our infrastructures code tooling actually allowed us easy to allow that easily. It's.

B

Virtualization as a visualization provider- and we made this change- I mean we already were able to set up a data data scientist really fast. I mean the migration of data etc. Those are different, different stuff, but I mean the actual provisioning of the data center was really automatic.

B

Virtualization at another provider, and it was as easy as as our old way. So um this is this. This was our approach. As far as running databases on official virtualization is concerned, we didn't feel any difference. I mean it's not getting any worse, it's better! Actually, it's very they it's more performance compared to that other than that, we haven't used the database, the existing databases from the application catalog, but we don't have to which we don't have to operate ourselves, but with microservices transformation.

B

We expect to uh the developers include their data, their own databases from the application catalog, and then they can develop whatever they want and push them to the to the production where they themselves decide on the size size of the database. How many choose how many replicas it's a topology? So we are. We are hopeful about the future.

A

Okay, okay, I think those are most of my questions. uh Thank you so much nemanja. This has been really excellent. Thank you for going over the things in such great detail. Sorry.

C

Thank you very much.

A

Thank you for spending this time with us.

A