How Uken Games reduces observability costs by 87% with ClickHouse

Uken Games replaced Datadog with an open-source observability stack centered on ClickHouse, achieving an 87% cost reduction. The new architecture consolidates all traces onto a single ClickHouse node, storing approximately 170 GB of data. This contrasts with Datadog’s distributed model, where pricing scales with infrastructure complexity, such as containers, EC2 instances, and disks, leading to unpredictable costs. The shift eliminates vendor-specific agents and proprietary pricing tiers, which previously influenced architectural decisions like choosing between serverless and EC2 for cost reasons.

The data pipeline relies on OpenTelemetry collectors deployed in two layers: lightweight agents on ECS instances and a gateway layer for batching and buffering. This design minimizes overhead on running applications, as traces are handed off locally before being forwarded to ClickHouse. The separation allows independent tuning of the gateway layer, which handles scaling and write optimization. OpenTelemetry’s vendor-agnostic standard ensures flexibility, as the choice of storage backend (ClickHouse) is decoupled from instrumentation, avoiding lock-in.

For application performance monitoring (APM), Uken uses SigNoz, an open-source tool built on ClickHouse. SigNoz replicates Datadog’s functionality, automatically inferring service relationships and metrics from telemetry data. This maintains the user experience Uken’s team relied on for debugging and player behavior analysis. The stack also integrates Prometheus for metrics and Grafana for visualization, covering all three pillars of observability (traces, metrics, logs) without proprietary dependencies.

The cost savings stem from eliminating Datadog’s per-container pricing and reducing infrastructure overhead. Datadog’s model charged additional fees for monitoring containers, disks, and services, which compounded as Uken scaled. By contrast, ClickHouse’s columnar storage and compression allow a single node to handle the same workload, reducing hardware and operational costs. The open-source stack also avoids the architectural constraints imposed by Datadog’s pricing, such as avoiding serverless components due to higher monitoring costs.

While the new stack achieves cost efficiency and flexibility, it requires in-house expertise to deploy and maintain. Uken’s team had to configure OpenTelemetry collectors, tune ClickHouse for performance, and integrate SigNoz, Prometheus, and Grafana. This contrasts with Datadog’s managed service, which abstracts much of the operational complexity. The trade-off is control over the stack and long-term cost predictability, but it demands engineering effort to replicate features like automated service discovery and alerting.