Case Study: Modal achieves real-time observability at scale with ClickHouse

How Modal uses ClickHouse to power real-time observability for AI workloads

Modal, an infrastructure platform for running large-scale AI and ML workloads, needed a way to capture and analyze massive volumes of observability data in real time. As usage grew across thousands of GPUs and containers, its existing logging system began to struggle with write and read scaling, making it harder for customers to debug and monitor their workloads. Modal turned to ClickHouse Cloud to support its real-time observability needs.

With ClickHouse, Modal built a pipeline that streams events through Kafka and ClickPipes into a single ClickHouse table, powering logs search and three real-time dashboards for function scaling, call timelines, and performance metrics. The system now ingests 1–2 million events per minute, stores around 500 billion logs, and still returns sub-second queries; Modal also reports scanning over 100 million rows in under four seconds.

Modal

Ro Arepally

Engineer