Leading banking and infrastructure development specialists shared cases and practices that ensure the stable 24/7 operation of the bank's digital services.
"In a bank, your work immediately faces high loads—you simply cannot do a poor job"
Instant payments, convenient purchases, and authorization in a couple of seconds—any modern bank must operate at high speeds. However, processing tens of thousands of transactions and operations non-stop is only possible for high-load systems.
To ensure the uninterrupted operation of 185 digital services, Halyk uses various technologies and approaches. These include microservice architecture, load balancing between data centers, database optimization, and more.
"What I like about development in a bank is that all your work immediately faces high loads, so you simply cannot do a poor job. We have 11.1 million active users in the Halyk SuperApp, and the availability of all digital services is one of the main performance indicators for the entire IT team. A few years ago, we set a KPI for downtime across all our products, and we also track the speed of operation processing. At the same time, withstanding high loads and meeting business requirements is impossible without the continuous improvement of IT systems and infrastructure," said Ruslan Serazitdinov, Managing Director of the bank, welcoming the meetup participants.
At first glance, authorization is a basic service present in any bank, but behind its apparent simplicity lie high loads on the entire system. This is especially true when hundreds of thousands of clients want to use the services simultaneously. In addition to core banking services, authorization covers all Halyk SuperApp products, including insurance, market, travel, kino.kz, and others. This does not even account for integrations with partner ecosystems through the BaaS service Halyk ID.
According to Nikolay Sulimov, Deputy Director of the Halyk SuperApp Department, numerous metrics help the IT team record load increases. Specialists also monitor request queues, the occurrence of errors, and more.
"Why does authorization always break first? Whether you are making a payment or buying a concert ticket, the entry point remains the same. All this happens on the same service, which by definition has the highest traffic. You need to be prepared for the storm in advance. Using markers, we identify bottlenecks, expand infrastructure for peak loads, and train the team. We also regularly conduct load tests to evaluate the performance of services at their limits, and we optimize resources," said Nikolay Sulimov.
Backstreet Boys: Managing It All
The next presentation was dedicated to the July launch of ticket sales for the Backstreet Boys concert in Almaty via the Kino.kz service. As Kirill Frolov, Acting Director of the Innovative Technologies Department, noted, several preventive measures helped avoid a service failure due to a sudden influx of 50,000–60,000 people:
- Expanding processing power (CPU) to speed up request processing, and random access memory (RAM) to handle the increased volume of data.
- Optimizing code and queries by the Kino.kz IT team.
- Launching a digital queue on a separate landing page and grouping concert ticket buyers into a separate category so as not to block other sales on the platform.
- Anti-fraud measures preventing users from bypassing the queue.
- Introducing a time limit for each user in the system to remove those from the queue who changed their minds but would otherwise delay others.
- Limiting the number of users simultaneously on the platform during ticket purchases.
- A well-designed UX to brighten up the waiting time for clients. Developers added the band's songs and interesting facts about the group to the landing page. At the same time, clients could see their place in the queue and its movement speed.
"In total, we increased the resources of the kino.kz platform by 100% compared to normal operations. We chose a simple technology stack—the Go language for applications, because we have a lot of expertise in it. On the frontend, we used NuxtJS to quickly launch applications, and the Redis database for state storage. Our system automatically redirected buyers to a separate landing page with a queue and checked the ticket purchase status every 5 seconds. As a result, the system held up, and we successfully sold 80% of the tickets in the first two days," shared Kirill Frolov.
On the Art of Balancing Toward Multi-Cloud
Presentations in the second part of the meetup covered physical and cloud infrastructure. According to Oleg Ivakin, Deputy Director of the IT Infrastructure Department, high-load systems are characterized by situations where existing resources are no longer sufficient to process all requests. However, buying one large server is not an option either.
The expert highlighted two approaches to scaling IT infrastructure: vertical and horizontal. The first involves expanding performance and capacity using hardware. The second accelerates request processing through caches—temporary storage for data copies.
"For effective scalability, balancing is needed; otherwise, it will be useless. To achieve this, we use various request management technologies. For example, L4, which quickly and efficiently distributes connection flows across servers, as well as L7, for a more detailed analysis of requests and routing them to the necessary services. We also apply the VxLAN fabric approach to manage a network of several data centers in different locations as a single system. We also conduct mandatory real-time monitoring," noted Oleg Ivakin.
Modern banking systems rely not only on physical but also on cloud infrastructure. The final presentation by Alexey Kalinnikov, Deputy Chairman of the Management Board at Halyk Kazteleport, was dedicated to when it is worth transitioning to multi-cloud and the advantages of distributing the load across different clouds.
