Hi all StarRocks for a mixed point-lookup + segmentation workload — isolation question Marketing automation platform, currently splitting user data across Vertica (segmentation, reachability) and MongoDB (contact lookups on the campaign trigger path). Considering collapsing to one engine. Scale, multi-tenant: - ~2B events/day ingest (~23k/sec), ~100 TB/year - Point lookups: 5k req/s peak, p99 <= 100 ms - Analytical: only 5–10 concurrent, but segment/reachability counts < 10s - Sparse per-tenant custom attributes Main question: can one PK-table cluster hold p99 <= 100 ms on 5k req/s point lookups while 10 concurrent full-population scans run? Published benchmarks test these separately. Looking for real workload-isolation / resource-group experience at this shape. Secondary: best modelling for sparse per-tenant custom attributes — JSON columns or something else?