How we scaled raw GROUP BY to 100 B+ rows in under a second

ClickHouse Cloud introduces a new feature called parallel replicas, enabling unprecedented horizontal query scaling by distributing a single query across thousands of cores. This feature allows complex queries, such as GROUP BY operations, to be executed on datasets as large as 100 billion rows in under half a second without pre-aggregation or data reshuffling. The technology achieves this by treating multiple nodes as virtual replicas, allowing for elastic scaling: as more nodes are added, query speed increases without data movement. The parallel replicas, currently in beta, enhance ClickHouse's capability to handle massive data throughput efficiently and are poised to become a benchmark for analytics performance, offering interactive speed across extensive datasets. This advancement allows ClickHouse to maintain its foundational goal of rapid data aggregation while extending its reach from single-node efficiency to cloud-scale operations, providing users with the flexibility to scale workloads seamlessly from individual nodes to large, distributed clusters with minimal configuration changes.