Migration guide

Moving PostgreSQL off Amazon RDS: a zero-downtime plan

Move RDS for PostgreSQL to a cluster in our cloud, your cloud account or your data center, with logical replication keeping both sides live until the switch.

  • Operated by AceMQ
  • 160+ customers
The migration in phases

To migrate PostgreSQL off Amazon RDS with minimal downtime, copy the schema with pg_dump, turn on logical replication on the RDS instance, subscribe the new cluster, wait until it has caught up, then move the application. Logical replication does not copy sequences or schema changes, so those are handled before the switch.

Which migration method fits

Logical replication is the method that keeps RDS live while the new cluster catches up. A dump and restore is simpler, but it copies a single point in time, so it fits databases that can stop writing for the length of the copy.

Illustrative: each phase is planned and rehearsed with your team before traffic moves.
MethodWhat it doesBoth sides liveSource
Logical replicationA publication on RDS and a subscription on the new cluster. Existing table data is copied first, then row changes stream until the cut.YesPostgreSQL docs
pg_dump and pg_restoreA logical dump of one database, restored into the new cluster. It is a consistent snapshot as of the dump start; later changes are not in it.No: writes must stop for the final copyPostgreSQL docs
pg_dump for the schema onlyUsed together with logical replication, because the schema and DDL commands are not replicated and the initial schema is copied by hand.YesPostgreSQL docs

Sources: PostgreSQL and AWS documentation, linked in each row; checked 30 Oct 2026.

How the methods were chosen: only methods that PostgreSQL itself documents are listed. Tools that wrap them are decided in the design review.

Check these before you start

Three things catch teams out: turning on logical replication on RDS needs a restart, logical replication skips some objects, and a slow subscriber can fill the source disk. Each is documented.

A restart on RDS.
Logical decoding on RDS needs the rds_superuser and rds_replication roles and the static parameter rds.logical_replication set to 1, which also sets wal_level, max_wal_senders, max_replication_slots and max_connections. The instance must be rebooted for it to take effect (AWS docs). Plan that reboot as a maintenance step.
What replication does not copy.
The schema and DDL commands, sequence data and large objects are not replicated, and only tables are supported, not views, materialized views or foreign tables (PostgreSQL docs). Schema changes are frozen or applied to both sides, and sequences are set to the latest values before the switch.
Replica identity.
UPDATE and DELETE need a primary key or a replica identity on each published table (PostgreSQL docs).
WAL growth.
A replication slot that is not consumed reserves WAL and can eventually fill the disk, so the subscription is monitored, and an abandoned slot is dropped (PostgreSQL docs).
Versions and extensions.
Logical replication can run between different PostgreSQL major versions (PostgreSQL docs). The extensions in use on RDS are listed and installed on the new cluster before the schema is restored.

Sources: AWS and PostgreSQL documentation, linked above; checked 30 Oct 2026.

The migration in phases

The new cluster is built and loaded while RDS keeps serving the application. The application only moves once the subscription has caught up and the checks pass.

  1. Review

    Database size, change rate, extensions, large objects, tables without keys, and the RDS parameter group.

  2. Build the new cluster

    Version, extensions, sizing and configuration come from the design. The schema is copied with pg_dump.

  3. Enable and subscribe

    Logical replication is switched on at RDS, a publication is created, and the new cluster subscribes. Existing rows copy first, then changes stream.

  4. Verify

    Row counts and checksums are compared, lag is watched, and the application is tested against the new cluster.

  5. Cut over

    Writes to RDS stop, the new side confirms it has applied everything, sequences are set, and the application connects to the new cluster.

The timeline and runbook for Amazon RDS migrations are written in the design step, from your data volume, traffic and change windows, and you approve them before the cutover is scheduled.

Cutover and rollback

The last step is a controlled switch, not a long outage. Writes stop on RDS only until the new cluster has applied every change and the sequences are set, and RDS stays untouched until you are sure.

  1. Subscription lag is at or near zero and the verification queries match.
  2. Application writes to RDS stop, briefly, and the last changes replicate.
  3. Sequences are advanced to the latest values, as the PostgreSQL docs describe for a switchover.
  4. The application connects to the new cluster. RDS stays available and unchanged as the fallback until the new side has run normally.

The sequence step comes from the PostgreSQL logical replication restrictions (PostgreSQL docs); checked 30 Oct 2026. The length of the write pause is measured in rehearsal, not promised in advance.

How OSStend runs the move

We run the migration and own the cutover, with both sides live until you switch. Zero downtime is the plan.

Managed PostgreSQL from OSStend is sized, tuned and operated by AceMQ engineers, with the extensions and versions the workload needs. It runs in our cloud, your cloud account or your data center. The reasons teams leave RDS are set out on the Amazon RDS alternatives page, and platform database users can read the Heroku Postgres alternatives page. The migration hub lists the other moves.

Operated by AceMQCompany figures, as published on acemq.com
160+customers served by AceMQ, which operates OSStend
10+years running open source data and messaging
24/7on-call engineers, not a ticket queue

AceMQ names a customer only with that customer's written permission, so most of this record stays anonymous. Ask for a similar workload in the 45-minute review.

When this move fits, and when it does not

The move fits when RDS limits the extensions, versions, upgrade timing or places the database can run. It does not when a standard RDS setup already meets the need.

Contact us if

  • You need extensions, versions or upgrade timing that RDS does not offer, or a database outside AWS.
  • The database is too busy for a long write pause, so a dump and restore will not do.
  • You want engineers to run the move and the database afterwards.
Talk to an engineer

Not a fit if

  • A standard RDS setup meets the need and the workload stays on AWS: staying avoids a migration.
  • The database is small enough to dump, restore and switch in a quiet window.
  • You want to run the move yourselves: the PostgreSQL methods above work without OSStend.

Questions about moving off RDS

Can PostgreSQL be migrated off RDS with no downtime?

Logical replication keeps RDS live while the new cluster catches up, so the application runs throughout. The final switch needs writes to pause briefly while the last changes apply. RDS needs a reboot to turn logical replication on (AWS docs, checked 30 Oct 2026).

Does logical replication copy sequences?

No. Sequence data is not replicated, so serial and identity sequences on the new cluster must be advanced to the latest values before the switch (PostgreSQL docs, checked 30 Oct 2026).

Is pg_dump enough for a large database?

pg_dump makes a consistent snapshot, but changes made after it starts are not in it, and the copy takes longer as the database grows. For a live database, pg_dump is used for the schema and logical replication for the data (PostgreSQL docs, checked 30 Oct 2026).

Talk to an engineer

Plan your RDS move with an engineer

45 minutes with an engineer: the current setup is reviewed against the documented method and the cutover is sketched. No ticket queue.

  • A design sketch for your workload, not a plan menu
  • A zero-downtime cutover path for what you run today
  • A price for that design, after the review
See pricing →

Your request goes to the OSStend team. No newsletter sign-up.

Get a quoteTalk to an engineer
2Engines
3Where should it run?
4What do you need?

Your request goes to the OSStend team. No newsletter sign-up.