The video compares Datadog, Grafana, and ClickHouse in the observability space, highlighting the trade-offs between costly, convenient commercial solutions and more affordable but complex self-hosted options. It advises teams to choose based on their size, budget, and expertise, emphasizing that effective data management is key to controlling observability costs rather than merely switching tools.
The video discusses the rising costs of observability tools, highlighting an extreme example where Coinbase received a $65 million bill from Datadog just for monitoring their software. While such cases are rare, typical monthly bills for observability and logging can still be substantial, often costing more to watch systems than to run them. This has led to criticism of these services as “bankruptcy as a service.” The video aims to clarify the differences between popular tools—Datadog, Grafana, and ClickHouse—and guide viewers on when to build their own observability stack versus buying a commercial solution.
Datadog, Grafana, and ClickHouse serve different roles in the observability ecosystem. Datadog is an all-in-one, hosted product that collects, stores, and visualizes data. Grafana is a DIY approach where users run separate components like Loki for logs, Mimir for metrics, and Tempo for traces, with Grafana providing the visualization layer. ClickHouse, on the other hand, is solely a columnar database designed for efficient storage and querying of large volumes of telemetry data, requiring additional tools for data collection and visualization. Comparing these tools directly is like comparing a car, its engine, and a pile of parts—they serve different purposes.
The high costs of Datadog stem from its pricing model, which charges per host and scales with auto-scaling environments, custom metrics, and log ingestion—even for logs that are filtered out and never searched. This can lead to unexpectedly large bills, especially when debug logging is left on inadvertently. ClickHouse offers a cost-effective alternative by using columnar storage that compresses repetitive telemetry data significantly and allows fast querying using standard SQL. This approach can flatten the cost curve, making self-hosted solutions cheaper than vendor services once data volumes reach a certain threshold.
However, self-hosting observability infrastructure is not without challenges. While it may reduce direct costs, it requires dedicated engineering resources to manage upgrades, storage tuning, and incident response. The operational overhead can be significant, and users do not get the convenience of a ready-made dashboard out of the box. Moreover, many high bills are due to poor data hygiene—logging excessively, retaining unnecessary metrics, and running multiple overlapping tools—rather than the tools themselves. The real solution lies in better data management rather than simply switching vendors or building new systems.
The video concludes by advising teams to choose observability solutions based on their size, budget, and willingness to manage complexity. Small teams with limited resources may benefit from paying for Datadog to save time and avoid operational headaches. Teams already invested in Prometheus and open-source tools might prefer the Grafana stack. Organizations with massive data volumes and SQL expertise could leverage ClickHouse or packaged builds like ClickStack or Signalz. Ultimately, the decision is about balancing cost, control, and convenience, ensuring teams understand what monitoring their software truly costs before the invoice arrives.