No room for old tech: from Sybase to Kafka at AccorHotels

Hotel group is overhailing its tech stack to meet demand for more tailored services

There's never a dull moment in the AccorHotels technology team.

"Over last three years we've brought in a lot of new tech, almost every month we introduce something new," said VP distribution and integration systems, Julien Ramakichenin.

These new technologies include hyperconverged infrastructure in the shape of Nutanix, microservices plus containers and Kubernetes, and Apache Kafka for stream processing. Another newcomer was the NoSQL database Apache Cassandra which enabled the group to streamline its operations from 50 servers to a single Cassandra node.

The drive to bring in the new is not just about efficiencies or keeping the team entertained. Rather, it is gearing up for a significantly more connected and real-time future, as the hotel group shifts to the world of 'augmented hospitality', with offers hinging on loyalty programmes and services tailored to clients' individual needs, a change that brings far more complex data requirements.

Ramakichenin's team works on what he calls the reservation lifecycle - everything that happens after an online customer clicks ‘Book'.

"It used to be quite simple, we'd process the reservation and send it to the hotel," he told Computing. "But it became more and more complex, and it's going to become more and more complex still".

AccorHotels runs 4,200 properties and has 250,000 employees worldwide. It operates 38 separate brands (up from 12 just a few years back) and has 36 major partnerships with businesses ranging from car hire operators to concert venues. So now when the customer clicks ‘Book', Accor's systems need to complete a whole raft of new tasks before sending the reservation to the hotel, including processing the payment, fraud detection, adding dietary requirements and room preferences from previous bookings, factoring in loyalty points, sharing information with partners and selecting suitable travel assistance options.

Up until two years ago, bookings were handled comfortably by a system constructed around Sybase, but as demands grew more complex the only way to scale up was to add more batch processes, said Ramakichenin, and this was problematic. Eventually, the number of sequential processes exceeded the 20 mark, which is when the technology team - which is surprisingly small considering the size of the company - started to look around.

The root of the problem, they decided, was their dependence on the Sybase database. It was not easy to scale and it could not be easily moved into the cloud. What was needed was a more modern streaming solution that could pull data from anywhere and allow for real-time operations. They looked at Tibco since they already used that firm's Enterprise Service Bus (ESB) but decided that Tibco's streaming capabilities would still require a specialist unit, whereas their plans required some radical democratisation. So, they looked at open source and landed on Kafka and the company set up by its founders, Confluent.

"Kafka is simple for devs to get their hands on," said Ramakichenin. "For this type of processing and starting small it's easier to do with Kafka, it was easier to add more options and to scale. With ESB, you'd end up with one team managing it, and that was not the best way for us."

New tech means new skills

Initially, the team built an additional service layer built on Kafka, and keep everything else much as before, but it became apparent that this halfway-house approach would not work.

"We thought there'd be very few people needed with the other devs sticking to Java. That was a mistake," said Ramakichenin. "We now see Kafka as the backbone of the whole information system. We are trying to remove that additional layer, and we now insist that everyone is aware of Kafka."

The AI and Machine Learning Awards are coming! In July this year, Computing will be recognising the best work in AI and machine learning across the UK. Do you have research or a project that you think deserves wider recognition? Enter the awards today - entry is free.

This new focus on Kafka will mean bringing back home many of the services currently provided by outsourcing firms, with a target of eventually having it all in-house, and there is also a desire to make much more use of cloud services (currently almost everything is on-premises). This will mean developing and hiring new skills. The roles they are looking to recruit include cloud managers and developers with expertise in automation and Terraform. It's important to find the right people, Ramakichenin explained.

"We invest in the skills, we invest in the people, and we found out also that these developers get passionate about the industry so they want to stay with us, so it's a good thing for us to invest in them," he said, adding that it's vital for developers to keep their skills fresh in an environment of constant change.

We used to hire very good people, but after two years we'd discover they became bad - Julien Ramakichenin

"We used to hire very good people, but after two years we'd discover they became bad. They got stuck on one tech or application, so we're really trying to get out of that, keep them skilled, send them to events and move them to other projects and teams so they stay on the market," he said.

"The first thing is they become bad developers and then they won't leave you because they can't go anywhere else! It's a good thing devs come and go."

'If it's going into production employ a partner'

Open source tools like Cassandra and Kafka can be taken up by anyone of course, but, Ramakichenin said, AccorHotels has a rule: if it's going into production employ a partner. So, Cassandra is deployed through DataStax, and with Kafka they turned to Confluent for training, support, for help on the architectural side and to learn lessons from the experience of the company's other customers.

"We always buy support from suppliers when we go to production, but it's not just that. If you don't get good service before that then you never go to production. When you look at Cassandra, Kafka, Kubernetes it's useful to have that support to fast-track you to production," Ramakichenin said.

"In the next two years, we are going to have a complete shift to a Kafka-based architecture, which is beginning now. So, we are using more services from Confluent."

In that time they hope to have built a new project called Welcome Connect to integrate the brands and partners into the back office and to have moved development to micrososervice and containers.

"I want to be out of Sybase completely so each system is decomposed from the others and so we can move easily to new things," Ramakichenin said.