Real-time OLAP

Real-time OLAP refers to analytical databases built to do two things traditional OLAP systems generally cannot do together: ingest new data continuously, often streamed in from a source like Apache Kafka, and answer aggregation queries — sums, counts, group-bys over millions of rows — with latency low enough for a live dashboard, typically well under a second, on data that may be only seconds old.

Conventional data warehouses handle either fast, ad hoc analytical queries on relatively static, batch-loaded data, or fast streaming ingestion feeding simple lookups, but not both together at low latency; real-time OLAP engines close that gap with architectures combining columnar storage, aggressive indexing, and incremental pre-aggregation, in some cases resembling a continuously updating materialized view rather than raw scan-on-query.

These systems are used for user-facing analytics dashboards, operational monitoring, and applications where staleness of minutes or hours is unacceptable, such as ad delivery metrics or fraud monitoring. The trade-off is usually reduced flexibility compared to a general-purpose warehouse — real-time OLAP engines are tuned for a narrower set of known query patterns and struggle more with arbitrary, unplanned ad hoc analysis than a batch warehouse does.