Scaling

When to scale

API: horizontally (stateless). The LB balances — sessions are stored in Postgres, not in memory.
Evaluator: vertically for now. Beyond >10k rules you need leader election + partitioning by rule_id hash.
TSDB: vertically up to ~20 vCPU / 64 GB. Beyond that — cluster mode (select/storage/insert).
Postgres: read replicas for UI-readonly. Writes stay on the primary.

See Multi-region. Each region has its own TSDB. The cabinet selects a region via ?region=.