
ColorSing scales real-time scores and rankings for live singing with Redis
- Under 1 msp99 latency at peak
- ~20,000requests per second
- Zerofailed requests since moving to Redis Cluster
ColorSing is a live-streaming platform centered on singing. Streamers, called "L Singers," perform live while listeners join through comments, virtual gifts, and real-time interactions. The platform helps people find an audience, build a community, and develop a career around singing.
As ColorSing grew, the team needed a data foundation that could keep scores and rankings accurate and instant during peak activity. ColorSing has run these workloads on Redis Cloud, a fully managed service, from the start. As traffic grew, the team moved from a single-node database to a clustered database on Redis Cloud, gaining a horizontally scalable, low-latency architecture that a small team can run.
Challenge: Keeping live scores and rankings instant as the audience grows
On ColorSing, every gift and comment can change a score or ranking within seconds. A single score record may be updated many times in a short window, and update volume rises with every new stream and listener. In a live experience, the value comes from everyone seeing and reacting to the same moment together.
The risk was that growth itself would degrade the experience. Delayed, dropped, or inconsistent updates could make users feel their support was not counted fairly. It would also pull engineers into repeated rounds of reactive performance work instead of building new features.
Limits of a single node
The team wanted to scale beyond a single-node setup before it became a user-facing bottleneck. The workload was not simple store-and-read. It required rapidly changing state that had to be available immediately. The team also needed a platform a small group could operate without heavy cluster management. Their criteria were low latency, horizontal scale, and low operational burden.
Solution: Redis for real-time scores, rankings, and locking
Why Redis
ColorSing compared Redis with Memcached and Google Cloud Memorystore. Memcached offered fast key-value access but lacked the richer data structures, such as Sorted Sets, that suit scores and rankings. Memorystore was a natural fit because ColorSing already runs on Google Cloud, but at the time of evaluation it did not support the cluster architecture the team needed. Redis combined the right data model and performance with a cluster architecture designed for future growth.
"Redis matched our workload, and it gives us a platform we can rely on as ColorSing scales.
A safe migration, one step at a time
Because ColorSing is a live service, stability came first. The team ran the single-node and clustered Redis Cloud databases in parallel and moved the workload gradually, verifying behavior at each step. The work included unwinding EVALSHA-based paths and resolving hash-tag key naming. Running two environments temporarily raised infrastructure cost, which the team treated as an intentional investment in protecting the user experience.
"The quality of the rollout strategy matters as much as the technology being introduced."
One endpoint, simple resources
The Redis Proxy presents the cluster to the application as a single node behind one endpoint. Engineers can issue variadic commands such as MGET without thinking about the cluster or needing special client logic. Sizing comes down to two parameters: storage size and operations per second.
Redis in the wider architecture
Redis powers latency-sensitive real-time processing, including live scores and rankings, and per-user distributed locks that serialize concurrent writes. It sits alongside Cloud Spanner, which remains the primary store for much persistent application data. ColorSing uses Redis selectively, where high update frequency and strict latency make it the right tool. Redis metrics flow through an OpenTelemetry Collector on GKE into Google Cloud Monitoring, with custom dashboards built by the team. Redis support helped the team interpret those metrics and put them to use.
Impact: Confidence to grow without reworking the foundation
Smooth experiences at scale
Scores and rankings continue to respond naturally as activity increases. Since moving from a standalone setup to Redis Cluster, ColorSing's deployment has handled about 20,000 requests per second at peak, with p99 latency under 1 ms across all Redis operations, even during the daily peak hours of 9 pm to midnight JST (September–October 2026). From late March to late October 2026, the cluster served about 112 billion requests with zero failed requests. These figures come from Redis Cloud metrics, which ColorSing collects in its own monitoring.
"Most users will never know that Redis is involved, and that is a good outcome."
Less reactive work for a small team
The engineering team now has a clearer scaling path and spends less time responding to predictable scaling concerns. Redis handles much of the cluster-level work, so engineers can focus on the product while maintaining live-service reliability.
A foundation for what's next
ColorSing wants to create more ways for singing to help people find an audience, build community, and develop a career. As users and experiences grow, the platform will need higher scalability and reliability. The team plans to meet that with an architecture that is scalable and as simple as possible, using Redis where its strengths are essential.
Get started
Learn how Redis can power real-time leaderboards and rankings for your own application.


