32 lines (22 loc) · 889 Bytes

Capacity Model

Hardware Assumptions

PostgreSQL 16+
16 vCPU / 64 GB RAM baseline node
NVMe SSD with high write IOPS
WAL on dedicated fast storage class

Throughput and Replay Targets

Sustained ingest/decision target: >= 5k TPS
p95 decision latency target: < 200ms
p99 fraud decision target: < 300ms
Replay benchmark: 30 days < 4h

Partition Strategy

Monthly partitioning for large fact tables (transactions, decision_audit_log)
Time-range pruning for replay and stress workloads

Resource Isolation

Tenant quotas from tenant_limits
Breach monitoring via tenant_limit_breaches
Queue backpressure monitoring via queue_backpressure

I/O Profile

Write-heavy from ingestion, queue and ledgers
Read-heavy from replay, stress and analytics windows
Burst-heavy from backtesting, counterfactual and forecast workloads