Apache Pinot

Use it when

• Analytics face end users: dashboards embedded in SaaS, feeds, and panels needing millisecond answers at very high query volume.
• Freshness is seconds from Kafka, with consuming segments answering queries as rows arrive.
• The indexing toolbox is the draw: inverted, sorted, range, JSON, geospatial, star-tree, and vector indexes hold latency stable under concurrency.
• You need upserts on a primary key over streaming data.

Think twice when

• Queries need rich ad-hoc joins; the multi-stage engine adds them opt-in, but StarRocks or ClickHouse handle exploratory SQL more naturally.
• The real workload is internal BI at modest concurrency; the operational complexity (controllers, brokers, servers, ZooKeeper) buys headroom you would not use.
• The team wants managed; StarTree is the dedicated option, otherwise the cluster is yours to run.

How it runs

Self-hosted: controllers, brokers, and servers over a deep store, coordinated by ZooKeeper and Helix; scale servers for data volume and query rate. Real-time tables consume Kafka directly while batch segments load beside them. StarTree Cloud is the managed offering built by its creators.