Services or capabilities described in AWS documentation might vary by Region. To see the differences applicable to the AWS European Sovereign Cloud Region, see the AWS European Sovereign Cloud User Guide.

Instance-specific performance and resource monitoring

Monitoring on instance level is key to understand connection skew, workload skew and data skew, as well as when to add routers or split shards to scale up for higher throughput with retained latency.

Overview

When your application issues a query against , that request traverses a sophisticated distributed system before returning results. A seemingly simple SELECT statement might touch multiple database instances, each playing a distinct role in processing your request. Understanding this journey—and the instances that power it—transforms how you design applications, interpret monitoring data, and diagnose performance issues.

This guide provides deep technical insight into instance architecture:

Instance architecture fundamentals

achieves horizontal scalability through functional separation across two specialized instance types:

This separation allows to scale connection handling, query coordination, and data storage independently based on your workload characteristics.

Router and shard comparison

Monitoring instance performance

Understanding instance-level performance is critical for operating effectively. Instance-specific monitoring reveals the distribution patterns that impact performance: connection skew, workload skew, and data skew.

Detecting skew

In an ideal deployment, workload and resources distribute evenly across instances. In practice, applications frequently experience skew—uneven distribution that concentrates load on specific instances.

Three types of skew to monitor:

Database Insights load distribution

The fastest way to assess instance-level health is Database Insights' Load Distribution view, which provides immediate visibility into how Active Sessions distribute across instances.

To access Load Distribution:

Healthy pattern: Load distributed relatively evenly across instances

Concerning pattern: Significant concentration on specific instances

CloudWatch metrics

For deeper analysis beyond Database Insights, CloudWatch provides instance-specific metrics that reveal resource utilization patterns.

The ServerlessDatabaseCapacity metric with dimension DBShardGroupInstance shows ACU consumption per instance, providing the most direct view of resource utilization.

When to investigate:

Router monitoring and troubleshooting

Routers can experience performance issues from two primary causes: uneven connection distribution and cross-shard workload concentration.

Unevenly distributed sessions

Symptom: One router handles disproportionate share of connections

Root cause: DNS caching causes multiple connection requests to resolve to the same router endpoint.

Most common during:

Remedies:

Shard monitoring and troubleshooting

Shards experience performance issues from three primary causes: resource constraints, data skew, and workload skew.

Shard resource utilization

A shard with popular shard keys will have more data and higher workloads. This manifests as resource utilization, i.e. the instance will consume more ACUs.

Remediation strategies:

When to split shards:

Shard data volumes

Use SQL functions to query data volumes:

SELECT subcluster_id, subcluster_type, pg_size_pretty(db_size) 
FROM rds_aurora.limitless_stat_database_size('postgres_limitless') 
ORDER BY 1;
            

To view per-table and per-shard data:

SELECT * FROM rds_aurora.limitless_stat_relation_sizes('public', 'table_name');
            

Resolving uneven utilization

When workload or data skew concentrates on specific shards, splitting shards redistributes load across more instances.

Important considerations:

Splitting shards allow further scaling, and consecutive shard splits is the path to higher throughput and further scaling, while retaining low latency.

Limitations

Be aware of these operational constraints:

Router limitations:

Shard limitations:

Mitigation strategies: