Neo4j Aura provides a Database-as-a-Service (DBaaS) through public cloud providers. It runs on containerized infrastructure, allowing AuraDB instances to allocate dedicated CPU resources. In this context, a "CPU" refers to a vCPU or CPU thread.

AuraDB instances are designed to deliver highly available, 24/7 database services. For AuraDB Business Critical, and AuraDB Virtual Dedicated Cloud tiers, each database runs on a three-node Causal Cluster, with each node deployed in its own container. Other tiers use a single-node architecture with built-in high availability at the infrastructure level.

On the Aura Metrics page, the Average CPU Usage (%) metric represents the average usage across all nodes in the cluster (for three-node clusters). In typical read-heavy workloads, CPU usage is generally balanced across nodes.

However, under write-heavy workloads, such as during bulk imports or ingestion jobs, the leader node (which handles all write transactions) often experiences significantly higher CPU usage. It may surpass 100%, even though the average across all nodes appears lower (e.g., around 60%). Using over 100% CPU during short periods is only possible because Aura uses vCPU. This may mask the fact that the leader is becoming CPU-bound, potentially leading to degraded performance.

In single-node AuraDB tiers, the reported CPU usage reflects the actual usage of the only node available, so monitoring spikes becomes more straightforward but equally critical.

Viewing statistics for individual instances

Leveraging the Detail View toggle, you can see which member is responsible for the heavy CPU utilization.

Statistics for the Leader

To find if the CPU spike comes from the leader member, navigate to the Database view and enable the Leader only filter toggle.

Recommendation

If you observe consistent performance degradation under write workloads, even when average CPU usage appears low you should consider scaling up your AuraDB instance to increase allocated CPU resources.

For more information on metrics and monitoring, refer to the official documentation:
Neo4j Aura – Viewing Metrics