Migrating Carousell from DigitalOcean to Google Cloud

In 2016, Carousell, the online classifieds marketplace, ran on DigitalOcean in Singapore, in a three-tier architecture of web, API gateway, and data layers. As the platform grew, Carousell wanted to make it more robust and reliable, and chose to move to Google Cloud.

At the time, the platform handled about 2,000 queries a second, with about 700 GB in PostgreSQL, 100 GB in Elasticsearch, and about 10 TB of static assets. Sakura Sky worked with Carousell’s engineers and with Google to plan the move to Google Cloud. Today this kind of work is part of our Cloud practice.

What we did

The project started on 22 June 2016. Sakura Sky provided the architect and an analyst, alongside Carousell’s engineering team and Google’s account and engineering team.

• Architecture. We designed the target architecture on Google Cloud.
• A phased plan. Discovery and assessment, architecture, proof of concept, data migration, application migration, deployment, testing, tuning, retirement of the old platform, and support after go-live.
• Changes kept to a minimum. We advised making as few changes as possible during the move. Carousell also wanted to move from Ubuntu to Debian and upgrade some core systems, and the plan allowed for this.
• A migration method for each data store. We planned how each one would move over a VPN: PostgreSQL by promoting a replica, Elasticsearch by joining new nodes to the cluster, Cassandra by rotating nodes, Redis by replication and failover, and RabbitMQ with the shovel plugin. VPN replication was tested and approved in the proof of concept.

Outcomes

Carousell went live on Google Cloud on time and on budget, about five weeks after kick-off. The move delivered cloud savings and increased uptime.

If you want a more robust and reliable platform, contact us to plan a move.