Apache Druid is often selected for fast analytics over event data. To compare it fairly, replay a representative ingestion stream with late arrivals and corrections, then query recent and historical windows. Measure latency, storage use, and how aggregations change when delayed events arrive.
Check retention, compaction, schema evolution, and the operational work needed to restore a failed segment or node. Benchmark on similar hardware with the same query shapes; a single count query is too easy. Make sure downstream dashboards agree on totals before any cutover. A candidate should simplify the workload without silently changing time semantics or cardinality.
Akousa