Gate 1
Cluster eligibility
Record engine, engine version, Region, platform version, minimum and maximum ACUs, scale-to-zero eligibility, maintenance state, and current quotas.
Replayable database burst test
Aurora Serverless Agent Workload Readiness
Describe the exact cluster, platform version, agent burst, queries, connections, latency objective, failure budget, upgrade plan, and cost boundary. Generate an acceptance test without treating AWS's capacity claim as an application guarantee.
Four readiness gates
Gate 2
Burst trace
Capture quiet baseline, burst arrival, concurrency, connections, query mix, locks, retries, tool calls, and end-to-end latency for the actual agent path.
Gate 3
Change safety
Define backup, version upgrade, maintenance window, alarms, failover, rollback, restore, connection-pool, and dependency tests before changing production.
Gate 4
Economics
Measure ACU history, I/O, storage, backup, transfer, logs, retries, model calls, and downstream service costs per accepted workload.
Minimum acceptance run
Freeze engine, application, query data, agent trace, and acceptance thresholds. Replay quiet-to-burst, sustained load, scale-down, resume, failure, and rollback. Record capacity, connections, latency percentiles, errors, retries, query plans, I/O, downstream limits, and total cost. Change one causal surface at a time.
Ready
Eligibility is verified and the burst, recovery, failure, cost, and rollback tests pass.
Upgrade then retest
Platform version is 1 or 2 and a controlled upgrade is required before measuring the new behavior.
Keep current design
Connections, queries, locks, downstream services, or cost—not capacity scale-up—remain the binding limit.
Read the source-backed guide
Aurora Serverless scaling for agentic AISeparate the 12-ACU announcement from application latency, upgrade, scale-to-zero, and cost evidence.Official facts checked August 15, 2026. Recheck current engine, Region, platform version, quota, pricing, and operational documentation.
Frequently Asked Questions
It turns the cluster, version, workload, latency, safety, and cost evidence you provide into a burst and upgrade test plan. It does not inspect or modify an AWS cluster.
AWS says platform versions 3 and 4 receive the enhancement by default; versions 1 and 2 can upgrade to the latest platform version 4.
No. Capacity scale-up is only one part of end-to-end latency; connections, queries, locks, storage, application code, models, tools, and downstream systems still matter.
Do not assume that. Compare cold, warm, scheduled, and minimum-capacity scenarios using the same trace, service objective, and full cost.