# Amazon Neptune Cluster

> Amazon Neptune bills per instance-hour for every cluster member plus per GB-month of storage and per million I/O requests. ZopNight stops and starts non-production graph clusters on schedule through the Neptune management API and watches 90 days of Neptune namespace metrics for clusters serving no requests.

Source: https://zop.dev/integrations/aws/neptune
Updated: 2026-08-19

---

Amazon Neptune is a managed graph database service billed per instance-hour plus storage and I/O. Development and analytics graph clusters frequently sit idle outside working hours while still billing full compute.

## Graph queries stop; the meters do not

A Neptune cluster bills per instance-hour for every member instance, writer and replicas alike, at the rate of its instance class, plus per GB-month for cluster storage and per million I/O requests against it. The compute meters run whether the graph answers a million traversals an hour or none, which matters because graph workloads skew exploratory: proof-of-concept knowledge graphs, fraud-analysis experiments, and analytics clusters that work business hours at most.

## Off-hours behavior

ZopNight stops and starts clusters on schedule through the Neptune management API. Stopping halts every member's compute meter while storage keeps billing, and AWS imposes the same 7-day cap it applies to Aurora: a stopped Neptune cluster restarts itself after 7 days, so a recurring schedule is the only durable way to keep a dev graph parked.

## The forgotten-experiment pattern

Neptune's dominant waste shape is the experiment that outlived its question. A team loads a graph, runs its analysis, and moves on. The cluster keeps billing every instance-hour for months, because graph databases rarely sit in anyone's operational review. Replicas provisioned for read scale that never came add multipliers to the same forgotten bill.

## Watching for silence

Clusters are discovered on the 6-hour cycle. Hourly CloudWatch metrics from the Neptune namespace run with a 90-day lookback, per-cluster cost comes from Cost Explorer or CUR 2.0, and idle recommendations key on sustained request silence rather than a single quiet day.

## Cluster and instance rows in the console

Neptune console, then Databases (shared with the RDS-style layout): the cluster row expands into member instances, each an instance-hour meter. The Monitoring tab's request charts are where an abandoned graph shows its flat line.