
Trying Out VPC Support for Aurora PostgreSQL Express Configuration and Changing a Cluster's VPC
This page has been translated by machine translation. View original
Introduction
On November 19, 2026, two updates were announced: VPC support for the Aurora PostgreSQL express configuration, and support for moving clusters from outside a VPC into a VPC and between VPCs. Moving between VPCs also applies to Aurora MySQL.
Existing clusters can be moved with ModifyDBCluster, without creating a new cluster or migrating data.
In this article, I created Aurora PostgreSQL clusters with the express configuration both inside and outside a VPC, and tried moving the cluster outside a VPC into a VPC, as well as moving between VPCs.
Test Environment
| Item | Details |
|---|---|
| Region | ap-northeast-1 |
| AWS CLI | 2.41.2 |
| Cluster | Aurora PostgreSQL 17.9 express configuration, Serverless v2, Data API enabled |
| Setup | 2 clusters, 1 instance per cluster (ap-northeast-1a) |
| VPC A | 10.20.0.0/16 |
| VPC B | 10.30.0.0/16 |
| Resources created in each VPC | Subnets in ap-northeast-1a, 1c, and 1d, a security group, and a DB subnet group |
| vpcmove-x | Cluster created without association to a VPC |
| vpcmove-y | Cluster created in VPC A |
| Test data | Created an items table in each cluster and inserted 100 rows (id from 1 to 100) |
Prerequisites
In the VPCs where the clusters are created and moved to, prepare a DB subnet group with subnets that have plenty of free IP addresses, and a security group that clients can connect through.
Creating the Clusters
I created Aurora PostgreSQL with the express configuration, which became supported in March 2026.
On November 13, 2026, CloudFormation support was also added, but at that point creating clusters inside a VPC was not supported.
Creating Outside a VPC
I created vpcmove-x without specifying a VPC.
aws rds create-db-cluster \
--db-cluster-identifier vpcmove-x \
--engine aurora-postgresql \
--with-express-configuration
Creating in VPC A
Using the VPC support added in this update, I created vpcmove-y in VPC A.
aws rds create-db-cluster \
--db-cluster-identifier vpcmove-y \
--engine aurora-postgresql \
--with-express-configuration \
--enable-vpc-networking \
--no-enable-internet-access-gateway \
--db-subnet-group-name vpcmove-a \
--vpc-security-group-ids <VPC A の SG の ID>
Moving a Cluster Not Associated with a VPC into a VPC
I move vpcmove-x to VPC A.
Command and Required Permissions
Use --enable-vpc-networking to enable VPC networking, and --no-enable-internet-access-gateway to disable the internet access gateway.
aws rds modify-db-cluster \
--db-cluster-identifier vpcmove-x \
--enable-vpc-networking \
--no-enable-internet-access-gateway \
--db-subnet-group-name vpcmove-a \
--vpc-security-group-ids <VPC A の SG の ID> \
--apply-immediately
With --apply-immediately, the move happens immediately. Without it, or with --no-apply-immediately, the move happens in the next maintenance window.
Running the command requires the following three permissions on the DB cluster.
- rds:ModifyDBCluster
- rds:EnableVpcNetworking
- rds:DisableInternetAccessGateway
Because a DB subnet group is specified, the rds:ModifyDBCluster permission on that subnet group is also required.
Values Before and After the Move
| Item | Before | After |
|---|---|---|
| Network | Internet access gateway | VPC A |
| DB subnet group | None | vpcmove-a |
| Security group | None | The one in VPC A |
| Cluster endpoint | vpcmove-x.cluster-xxxxxxxxxxxx.ap-northeast-1.rds.amazonaws.com | Same |
| Resolved IP addresses | Public IP addresses of the gateway (3) | Within the range of VPC A (1) |
| Row count and min/max id of the items table | 100 rows, id 1 to 100 | 100 rows, id 1 to 100 |
Duration and Name Resolution
Times in the log excerpts are UTC, and times in the text are JST (UTC+9). All values are measurements from a single move.
The move took about 29 seconds, from the CloudTrail ModifyDBCluster record (09:47:04) to the RDS completion event (09:47:33).
2026-11-20T00:47:12.306000+00:00 None The VPC configuration modification for DB cluster vpcmove-x has started.
2026-11-20T00:47:33.432000+00:00 None The VPC configuration modification for DB cluster vpcmove-x has completed.
I resolved the cluster endpoint with dig about every 10 seconds. The resolver was the system default, and the TTL was 5 seconds.
| Time (JST) | Resolved IP addresses |
|---|---|
| 09:47:14 | Public IP addresses of the gateway (3) |
| 09:47:24 | Public IP addresses of the gateway (3) |
| 09:47:34 | Range of VPC A (1) |
| 09:47:44 | Range of VPC A (1) |
The observation 9 seconds before the completion event (09:47:33) returned the public IP addresses of the gateway, and the observation 1 second after returned an IP address in the range of VPC A.
Data API Calls
I ran SELECT 1 continuously through the Data API about every 0.5 seconds, without retries. To connect to vpcmove-x, I used the secret of a separately created DB user, rather than the IAM-authenticated master user that is the default for the express configuration.
| Interval | Calls | Failures | Max response time |
|---|---|---|---|
| Entire measurement (09:46:24 to 09:48:16) | 211 | 0 | 341 ms |
| Around the move operation (09:47:03 to 09:47:50) | 90 | 0 | 200 ms |
The median response time over the entire measurement was 27 ms.
Moving Between VPCs
I move vpcmove-y from VPC A to VPC B.
Command
Pass the destination DB subnet group and security group to modify-db-cluster.
aws rds modify-db-cluster \
--db-cluster-identifier vpcmove-y \
--db-subnet-group-name vpcmove-b \
--vpc-security-group-ids <VPC B の SG の ID> \
--apply-immediately
Values Before and After the Move
I obtained the values before and after the move using the following methods.
- DB subnet group, VPC, security group, endpoint:
describe-db-clustersanddescribe-db-instances - IP addresses: name resolution of the cluster endpoint
- items table:
SELECTvia the Data API
| Item | Before (VPC A) | After (VPC B) |
|---|---|---|
| Cluster endpoint | vpcmove-y.cluster-xxxxxxxxxxxx.ap-northeast-1.rds.amazonaws.com | Same |
| Cluster resource ID (DbClusterResourceId) | cluster-XXXXXXXXXXXXXXXXXXXXXXXXXX | Same |
| DB subnet group | vpcmove-a | vpcmove-b |
| VPC of the DB instance | VPC A | VPC B |
| Security group | The one in VPC A | The one in VPC B |
| Resolved IP address | Within 10.20.0.0/16 | Within 10.30.0.0/16 |
| Row count and min/max id of the items table | 100 rows, id 1 to 100 | 100 rows, id 1 to 100 |
Duration and Events
The move took about 35 seconds, from the CloudTrail ModifyDBCluster record (09:44:09) to the RDS completion event (09:44:44).
2026-11-20T00:44:20.407000+00:00 None The VPC configuration modification for DB cluster vpcmove-y has started.
2026-11-20T00:44:44.684000+00:00 None The VPC configuration modification for DB cluster vpcmove-y has completed.
Name Resolution Switchover
I resolved the cluster endpoint with dig about every 10 seconds.
| Time (JST) | Resolved IP address |
|---|---|
| 09:44:38 | Range of VPC A |
| 09:44:48 | Range of VPC A |
| 09:44:58 | Range of VPC B |
| 09:45:08 | Range of VPC B |
The observation 4 seconds after the completion event (09:44:44) returned the pre-move IP address, and the observation 14 seconds after returned the post-move IP address.
Data API Calls
I ran SELECT 1 continuously through the Data API about every 0.5 seconds, without retries. For vpcmove-y, I connected using a Secrets Manager secret with password authentication for the master user (changed from IAM authentication after creation).
| Interval | Calls | Failures | Max response time |
|---|---|---|---|
| Entire measurement (09:42:38 to 09:45:31) | 324 | 0 | 320 ms |
| Around the move operation (09:44:07 to 09:45:00) | 101 | 0 | 245 ms |
The median response time over the entire measurement was 32 ms. The User Guide says that connections are interrupted during the move, but there were 0 failed Data API calls.
Summary
I confirmed that Aurora PostgreSQL clusters can be created inside a VPC with the express configuration, and that both moving from outside a VPC into a VPC and moving between VPCs complete in about 30 seconds.
Previously, moving a cluster to a different VPC required restoring from a snapshot or using Blue/Green Deployments with replication, and the preparation and migration time tended to be a problem.
Because this update makes it possible to move clusters in a short time, one possible use is, in an environment where VPCs are separated for testing and production, to move a DB that has been evaluated in the test environment directly into the production environment.
After the move, the DB endpoint (DNS name) is retained, but the IP address changes. I have not verified connections from inside a VPC, so for workloads that use connection pools, I recommend evaluating beforehand whether they can keep up with the IP address change. Also, the User Guide states that some configurations, such as Blue/Green Deployments, do not support moving VPCs. Please read it and evaluate thoroughly in a test environment before use.
