Amazon Aurora PostgreSQL CloudWatch Dashboard - Track Health

Use this dashboard to monitor an Aurora PostgreSQL cluster across serverless capacity, replica lag, memory and cache, transactions, instance storage I/O, cluster volume growth, and WAL activity.

Amazon Aurora publishes cluster and instance metrics to the AWS/RDS CloudWatch namespace. The Prometheus CloudWatch Exporter reads them, and the OpenTelemetry Collector prometheus receiver scrapes the exporter.

Dashboard Coverage

Use these panels to:

• Separate the writer from the readers: Instance panels group by DBInstanceIdentifier, so each instance is its own series, and Commit Throughput splits by the Role dimension into WRITER and READER.
• Size a Serverless v2 cluster: Track provisioned Aurora Capacity Units against the utilization of the configured maximum.
• Measure replica lag: Watch how far each reader trails the writer over the shared storage volume.
• Spot memory pressure: Read freeable memory, buffer cache hit ratio, and swap together.
• Watch storage growth and billed I/O: Cluster volume size and billed read and write operations are cluster-scoped. They have no per-instance series.
• Stay ahead of transaction ID wraparound: Track the age of the oldest unfrozen transaction ID.

Metrics Included

Cluster & Capacity

• Serverless Database Capacity: Aurora Capacity Units provisioned for the cluster (ServerlessDatabaseCapacity). Serverless v2 only.
• ACU Utilization: Share of the instance maximum ACU setting in use (ACUUtilization).
• CPU Utilization: CPU used by the database engine on each instance (CPUUtilization).
• Database Connections: Client connections open on each instance (DatabaseConnections).
• Aurora Replica Lag: Lag between a reader and the writer (AuroraReplicaLag). Reported by reader instances only.
• Engine Uptime: Time since the engine last started (EngineUptime). A drop marks a restart or failover.

Memory & Cache

• Freeable Memory: RAM available on each instance (FreeableMemory).
• Buffer Cache Hit Ratio: Share of requests served from the buffer cache (BufferCacheHitRatio).
• Swap Usage: Swap space used on the instance (SwapUsage).

Transactions

• Commit Latency: Average time to commit a transaction (CommitLatency).
• Commit Throughput: Commits per second split by writer and reader role (CommitThroughput).
• Deadlocks: Deadlocks per second detected by the engine (Deadlocks).
• Maximum Used Transaction IDs: Age of the oldest unfrozen transaction ID (MaximumUsedTransactionIDs).

Instance Storage I/O

• Instance IOPS: Read and write operations per second on local storage (ReadIOPS, WriteIOPS).
• Instance I/O Latency: Average time per read and write operation (ReadLatency, WriteLatency).
• Instance I/O Throughput: Bytes read and written per second (ReadThroughput, WriteThroughput).
• Disk Queue Depth: Outstanding I/O requests waiting on the disk (DiskQueueDepth).
• Network Throughput: Client network traffic to and from the instance (NetworkReceiveThroughput, NetworkTransmitThroughput). Storage traffic to the cluster volume is not counted here.

Cluster Storage Volume

• Volume Bytes Used: Storage consumed by the cluster volume (VolumeBytesUsed).
• Volume Read I/O Operations (5 min) and Volume Write I/O Operations (5 min): Billed operations against the cluster volume (VolumeReadIOPs, VolumeWriteIOPs). CloudWatch reports these as a count per 5-minute interval rather than a rate, so divide by 300 for operations per second.

WAL & Replication Slots

• Transaction Logs Disk Usage: Disk consumed by write-ahead logs (TransactionLogsDiskUsage).
• Replication Slot Disk Usage: Disk held by replication slots (ReplicationSlotDiskUsage).
• Oldest Replication Slot Lag: WAL retained for the furthest-behind replica or logical slot (OldestReplicationSlotLag).

Dashboard Variables

Use these filter variables:

• dbcluster_identifier: Filter the cluster-scoped panels, meaning storage volume and serverless capacity.
• dbinstance_identifier: Filter the per-instance panels. AWS/RDS carries no engine dimension, so this lists every Aurora instance in the account. Select a cluster first to narrow it.

Panels That Read Empty

Some panels stay empty until the matching condition exists, which is expected rather than a misconfiguration:

• Cluster Storage Volume panels take about an hour to appear on a newly created cluster, and CloudWatch emits them at 5-minute intervals.
• Aurora Replica Lag has no series for the writer, because Aurora reports it on reader instances only.
• Transaction Logs Disk Usage reads -1 while Aurora PostgreSQL uses log records rather than transaction logs, which is the case unless logical replication or AWS DMS is active.
• Replication Slot Disk Usage and Oldest Replication Slot Lag read 0 on a cluster with no logical replication slots.

Next Steps

• Collect Aurora engine metrics to see query-level activity alongside this dashboard.
• Set up alerts on replica lag, connection count, and volume growth.
• Set up a notification channel for Slack, PagerDuty, or email.

Related Dashboards

• Amazon Aurora PostgreSQL Engine Metrics
• Amazon Aurora MySQL CloudWatch
• PostgreSQL