Heroku Postgres 15 Is Being Retired: Why It's a Good Moment to Move Your Database to Your Own Cloud Account
Heroku stopped provisioning new Postgres 15 databases on November 11, 2026, starts forced upgrades on January 28, 2027, and reaches end of life on February 28, 2027. Here is what that forces you to do, how to upgrade on Heroku or move to RDS, Cloud SQL, or Azure Database for PostgreSQL with minimal downtime, and what it costs.
The dates, from Heroku's own changelog: Heroku sent the Postgres 15 deprecation notice on August 4, 2026. Since November 11, 2026 "You can't create Postgres 15 databases", although forks and followers of existing databases still work. Forced upgrades start January 28, 2027, and from February 28, 2027 Heroku starts "restricting database access and deleting databases" still on Postgres 15 (Heroku changelog).
Your existing Postgres 15 database keeps running for now. Nothing breaks today, but every new environment, review app, or script that pins version 15 fails, and you have roughly ten weeks before Heroku schedules the upgrade for you.
Heroku retires versions faster than PostgreSQL does. Heroku supports each major version "for three years on the platform" (Heroku version support). The PostgreSQL community supports Postgres 15 until November 11, 2027 (PostgreSQL versioning policy), and Amazon RDS keeps it in standard support until February 29, 2028 (RDS release calendar).
You have two real options: a pg:upgrade on Heroku (around 10 minutes of downtime on most databases), or moving to Amazon RDS, Aurora, Google Cloud SQL, or Azure Database for PostgreSQL in your own account. A move done with pg_dump and pg_restore into Postgres 17 or 18 is your major version upgrade too, so you test once.
Plan the move around a write freeze. Heroku does not support logical replication to external instances and does not grant the superuser or replication roles that AWS DMS change data capture needs. Rehearse on a Heroku fork, time the dump and restore, and size your maintenance window from that number.
Somebody on your team got an email from Heroku on August 4, 2026. Since November 11, new Postgres 15 databases cannot be provisioned. On January 28, Heroku starts upgrading the ones you have left.
Few teams planned to touch their database this quarter. The deprecation email forces a question most of them had been postponing, about who decides when the database changes version.
This guide covers what the change forces you to do, the upgrade path if you stay, the migration path if you leave, and an honest cost comparison from public price pages.
What exactly did Heroku announce for Postgres 15, and when?
Heroku is deprecating every database running Postgres 15, with an end-of-life date of February 28, 2027. New Postgres 15 databases cannot be created after November 11, 2026, forced upgrades start January 28, 2027, and Heroku starts restricting access and deleting non-compliant databases after end of life (Heroku changelog: Heroku Postgres 15 deprecation).
Here is the schedule exactly as Heroku published it:
Date
What Heroku says happens
August 4, 2026
"We send a deprecation notification to Heroku Postgres 15 database users."
November 11, 2026
"You can't create Postgres 15 databases after this date. You can still create forks and followers of existing databases."
January 28, 2027
"Forced upgrades on deprecated Postgres 15 databases start."
February 28, 2027
"We start restricting database access and deleting databases for Postgres 15 databases after this end-of-life date."
The same day, Heroku made Postgres 18 generally available and switched the default: "New databases default to Postgres 18" (Heroku changelog: Postgres 18 GA).
None of this is a surprise move. It follows Heroku's published policy: "When we release a new major version, we deprecate the third major version released before it." Provisioning stops three months after the deprecation date, forced upgrades are scheduled five months after it, and end of life lands six months after it (Heroku Postgres version support).
The policy also tells you what comes next. Postgres 16 on Heroku deprecates the day Heroku ships Postgres 19. If you upgrade to 16 today to get through this, you will do this again.
What does the deprecation force you to do right now?
If you run Postgres 15 on Heroku, you must upgrade or migrate that database before January 28, 2027, or Heroku schedules the upgrade for you. Anything that creates new Postgres 15 databases is already broken since November 11, 2026.
Here is what I'd check this week:
Inventory. Run heroku pg:info on every app and list every database still on 15, including staging and review-app databases.
Pinned versions. Search app.json, CI scripts, Terraform, and internal tooling for anything that provisions Heroku Postgres with version 15. Those calls now fail or get a different version than you expect.
Environment parity. New review apps and staging databases default to Postgres 18 while production stays on 15. Tests pass on 18 and production runs 15, so you are no longer testing what you ship.
Extensions. List what is installed with SELECT extname, extversion FROM pg_extension;. Extensions decide whether an upgrade or a move goes smoothly.
Owner and date. Pick a person and a target date before January 28. A forced upgrade at a time you did not choose is the one outcome to avoid.
Forks and followers of existing Postgres 15 databases still work. That matters, because a fork is the cheapest way to rehearse either path below.
Why does Heroku retire Postgres versions earlier than AWS, Google Cloud, or Azure?
Heroku supports each major version for three years on its platform, while the PostgreSQL community supports each major version for five years, and the three hyperscalers keep standard support at least until community end of life, then sell up to three more years of extended support. On Heroku, Postgres 15 reaches end of life about eight and a half months before the community stops patching it.
The PostgreSQL Global Development Group "supports a major version for 5 years after its initial release" and lists November 11, 2027 as the final release of Postgres 15 (PostgreSQL versioning policy). Heroku's own policy is "three years on the platform" (Heroku version support).
Platform
Postgres 15 standard support ends
What happens after
Source
Heroku Postgres
February 28, 2027 (end of life)
Forced upgrades from January 28, 2027, then access restricted and databases deleted
I don't think Heroku is wrong to do this. Running a shared fleet on fewer versions is how a managed platform stays safe and affordable. The consequence for you is that Heroku owns your upgrade calendar, and its calendar runs faster than everyone else's.
On RDS, the trade is explicit. AWS says RDS Extended Support "allows you to continue running a database on a major engine version past the RDS end of standard support date for an additional cost", for "up to 3 years" (Amazon RDS Extended Support). You still have to upgrade eventually. You just decide whether a late upgrade costs you money or a forced maintenance window.
Should you upgrade on Heroku or move Postgres to your own cloud account?
Upgrade on Heroku if the database is small, your app is staying on Heroku for the next year, and nobody is asking for data residency, VPC access, or cost control. Move to your own cloud account if you were already planning a Heroku exit, need the database inside your own VPC, or want version upgrades on the community's five-year clock instead of Heroku's three.
Path
Typical downtime
Who sets the next upgrade date
Effort
Best for
pg:upgrade on Heroku (direct upgrade)
"around 10 minutes for most use cases"
Heroku, three years per version
Low
Small teams staying on Heroku
Follower failover on Heroku
"20-30 minutes"
Heroku
Medium
Standard tier and above, extra control over the switch
Data copy on Heroku
"about 3 minutes per GB", databases under 10 GB only
Heroku
Low
Essential tier and small databases
pg_dump / pg_restore to RDS, Cloud SQL, or Azure
Write freeze sized by your rehearsal
You, with five years of community support plus paid extended support
Medium
Databases up to roughly 100 GB, teams leaving Heroku
Application-level dual write to the new database
Close to zero
You
High
Large, write-heavy databases that cannot freeze writes
Downtime figures for the Heroku methods come from Heroku's upgrade guide. The 100 GB guideline for dump and restore comes from Heroku's PostgreSQL migration planning post, which cites AWS guidance and warns to "plan for a generous amount of downtime" for large databases.
If you were going to leave Heroku in 2027 anyway, I would not upgrade on Heroku first. Moving with pg_dump into Postgres 17 or 18 is the major version upgrade. You test your application against the new version once, on the infrastructure you will actually run, instead of twice.
If you are staying, that's a fine answer too. Do the upgrade now, on your own schedule, and skip to the FAQ.
How do you upgrade Postgres 15 if you stay on Heroku?
Run heroku pg:upgrade on a fork first to test, then on production during a maintenance window; by default it targets "the latest supported version", which is Postgres 18, and you can pass --version to choose 16, 17, or 18 (Heroku upgrade guide).
The tier decides your method:
Essential tier. A direct upgrade "immediately upgrade[s] your database version" by moving the data to a new database. Test with the copy method, because forking only works on Advanced-tier and classic Standard-tier or higher plans (Heroku forking docs).
Standard tier and higher. Heroku runs the direct upgrade like a maintenance you can schedule. Test with a dry run or a fork first (Heroku: testing version upgrades).
Followers. "If your database has any followers, the followers are also automatically upgraded to the target version after your leader database is upgraded."
Pick 18 if your extensions and drivers support it. Postgres 16 on Heroku will be deprecated when Heroku ships Postgres 19, so jumping to 16 buys the least time.
How do you move a Heroku Postgres database to RDS, Cloud SQL, or Azure with minimal downtime?
The reliable way out of Heroku Postgres is pg_dump and pg_restore during a short write freeze, rehearsed on a Heroku fork so you know the exact duration before the real cutover. Near-zero-downtime replication tools are mostly off the table, because Heroku does not support logical replication to external instances and does not grant the roles those tools need.
Why can't you just replicate Heroku Postgres to RDS?
Heroku's help center is direct: "We don't support logical replication from external instances for Heroku Postgres instances" (Heroku help). On third-party tools, Heroku adds that tools "such as Amazon DMS, may require SUPERUSER or REPLICATION roles", and "Heroku Postgres does not give customers access to the superuser role, or allow customers to grant other users the replication role" (Heroku help).
AWS confirms the other half. For DMS "full-load plus CDC tasks or CDC-only tasks, grant superuser permissions for the user account specified for the PostgreSQL source database" (AWS DMS PostgreSQL source). Trigger-based tools such as Bucardo usually expect superuser access too, so treat any change data capture approach as something to prove on a fork, never something to plan a date around.
Heroku followers do not help either. They are physical replicas that live inside Heroku, so a follower can't be an external RDS instance.
What does a dump-and-restore cutover look like?
Here is the sequence I recommend, with the target database already provisioned in your account:
Provision the target. Create RDS, Aurora, Cloud SQL, or Azure Database for PostgreSQL on Postgres 17 or 18, in the same region as your new app, in a private subnet.
Fork and rehearse. Create a fork of production on Heroku (still allowed after November 11), dump it, restore it into the target, run your test suite, and write down the total time.
Check extensions. Compare pg_extension on Heroku with the RDS supported extensions list or your provider's equivalent. If the restore complains about heroku_ext, Heroku's docs say to run CREATE SCHEMA heroku_ext; on the target first (Heroku import and export).
Freeze writes. Put the app in maintenance mode or read-only mode and stop workers and schedulers.
Dump and restore. Use the newest pg_dump client, in directory format with parallel jobs, from a machine close to both databases.
Validate. Compare row counts on the largest tables, check sequence values, and run smoke tests against the new database.
Switch. Point DATABASE_URL at the new database, start the app on its new home, and keep Heroku read-only for a few days as your rollback.
The dump and restore commands look like this:
BASH|Dump from Heroku with a pg_dump client at least as new as the source (directory format allows parallel jobs)
pg_dump --format=directory --jobs=4 --no-acl --no-owner \
--dbname="$HEROKU_DATABASE_URL" --file=./heroku_dump
# Restore into the new Postgres 17 or 18 instance in your own account
pg_restore --verbose --no-acl --no-owner --jobs=4 \
--dbname="$TARGET_DATABASE_URL" ./heroku_dump
The PostgreSQL docs say "the output of pg_dump can be expected to load into PostgreSQL server versions newer than pg_dump's version" and that "pg_dump cannot dump from PostgreSQL servers newer than its own major version" (pg_dump docs). Use an 18 client, dump 15, restore into 18. The --jobs option only works with the directory format, and the --no-acl --no-owner flags match what Heroku recommends.
Skip heroku pg:backups for anything large. Heroku says "PGBackups are intended for moderately loaded databases up to 20 GB" (Heroku import and export).
How do you shrink the freeze on a bigger database?
You shrink the freeze by moving less data inside it. Three techniques work without replication privileges:
Pre-copy cold data. Append-only tables (events, audit logs, old invoices) can be copied days ahead by primary key range. During the freeze you copy only rows above the last copied ID.
Run the dump inside the target network. A restore that runs from a pod or VM in the same VPC as the new database avoids a round trip through your laptop. On Qovery, a one-off job in the target environment does this.
Dual write in the application. For a database you cannot freeze, write to both databases for a period, backfill, compare, then switch reads. It's the most work and the only route to close-to-zero downtime from Heroku.
Size the window from the rehearsal. If the fork takes 40 minutes end to end, announce a one-hour window. Don't promise "a few minutes" because a blog post said so.
Own your Postgres upgrade calendar.
Qovery deploys your app and a managed PostgreSQL (Amazon RDS on AWS) into your own cloud account, with git-push deployments and a preview environment per pull request. Start deploying in under 10 minutes.
Which managed Postgres should you move to: RDS, Aurora, Cloud SQL, or Azure?
Pick the managed Postgres on the cloud where your application will run, because cross-cloud latency between app and database costs more than any feature difference. On AWS, RDS for PostgreSQL is the default and Aurora fits when you need its storage model; on GCP, Cloud SQL; on Azure, Azure Database for PostgreSQL flexible server.
Option
Version currency
Extended support after standard support
Notes
Amazon RDS for PostgreSQL
New major versions "within 30 days of the community release"
Paid, up to 3 years
Closest match to Heroku Standard and Premium
Amazon Aurora PostgreSQL
New major versions "within 8 months" of the community's first minor
Paid, up to 3 years
Different storage engine, long-term support minor versions
One Azure detail to plan for: "Servers running unsupported PostgreSQL versions are automatically enrolled in extended support", and the only way to stop the charges is to upgrade (Azure extended support FAQ). On RDS, enrollment depends on how you created the instance, so check the EngineLifecycleSupport setting. Every hyperscaler gives you more runway than Heroku, and on AWS and Azure that extra runway is billed.
If you want a broader list, including hosted Postgres vendors that are not hyperscalers, see our Heroku Postgres competitors roundup.
Does moving Heroku Postgres to RDS save money?
Moving a single Heroku Postgres database to RDS saves little at the entry tier and more on high-availability plans. A 4 GB Standard-0 at $50 a month compares to about $55 on a single-AZ RDS instance, while a 4 GB Premium-0 with high availability at $200 compares to about $109 on Multi-AZ RDS.
I priced this from public pages only: Heroku's add-on pricing and the Amazon RDS for PostgreSQL pricing page (on-demand, US East N. Virginia, 730 hours a month, gp3 storage at $0.115 per GB-month single-AZ and $0.23 Multi-AZ).
Heroku plan (price, RAM, storage)
Closest RDS setup
RDS math
RDS total per month
Standard-0: $50, 4 GB, 64 GB
db.t4g.medium (4 GiB), single-AZ, 64 GB gp3
$0.065/h x 730 + 64 x $0.115
about $55
Premium-0: $200, 4 GB, 64 GB, HA
db.t4g.medium, Multi-AZ, 64 GB gp3
$0.129/h x 730 + 64 x $0.23
about $109
Standard-2: $200, 8 GB, 256 GB
db.m7g.large (8 GiB), single-AZ, 256 GB gp3
$0.168/h x 730 + 256 x $0.115
about $152
Premium-2: $350, 8 GB, 256 GB, HA
db.m7g.large, Multi-AZ, 256 GB gp3
$0.337/h x 730 + 256 x $0.23
about $305
Caveats, so this table stays honest:
db.t4g is burstable. A steady, CPU-heavy database belongs on db.m7g or similar, which costs more than the t4g rows above.
Reserved Instances and Savings Plans lower the RDS numbers for a one- or three-year commitment. Heroku plans are flat.
Not included: data transfer, extra backup retention, Performance Insights retention, and the engineering time to run the migration.
Heroku Advanced-tier plans exist alongside the classic plans priced here. I did not compare them.
The bigger financial lever is usually the application tier around the database and, if you are moving to AWS, migration funding. See how to fund a Heroku to AWS migration with credits before you sign anything.
Cost alone rarely justifies moving Postgres. Control over the upgrade calendar, data in your own account and VPC, and one bill with your cloud discounts are what make the move worth it.
Where does Qovery fit when you move your database out of Heroku?
Qovery deploys your application and a managed PostgreSQL into your own cloud account, so the database you migrate into lives next to the app, in your VPC, on your cloud bill. On AWS, Qovery provisions Amazon RDS for PostgreSQL through a Blueprint, and you pick the major version (for example PostgreSQL 17) when you create it (Qovery database docs).
Qovery is an Agentic DevOps platform. The Qovery control plane is separate from where your workloads run. Your apps run on Kubernetes in your own AWS, GCP, Azure, or Scaleway account, on a cluster Qovery manages for you, or on an existing Kubernetes cluster you already operate. You get the Heroku habits back (git-push deployments, a preview environment per pull request, environment cloning) without hiring a platform team first.
For the database specifically:
AWS. The RDS Blueprint is "a versioned, ready-to-use template" that materializes as a Terraform or OpenTofu service in your account (Qovery Blueprints). Your platform lead decides which templates exist and RBAC decides who can deploy them.
Scaleway. Managed databases use Scaleway Managed PostgreSQL.
GCP and Azure. Provision Cloud SQL or Azure Database for PostgreSQL with your own Terraform or OpenTofu module, run as a Qovery Terraform service, and inject the connection string into your app.
Rehearsal. Clone the production environment, point the clone at a restored copy of the Heroku fork, and run your tests before cutover day.
What Qovery does not do is pick your upgrade date for you. RDS major version upgrades stay your decision, on AWS's timeline, which is the point of moving.
How do I do this with Qovery from Claude Code or Cursor?
Install the Qovery Agent Skills, then ask your coding agent in plain language to deploy your app and a managed PostgreSQL to your own cloud account. The agent asks you to approve each action, and the Qovery MCP Server is read-only unless you explicitly enable write access.
Install the skills globally for Claude Code, Cursor, Codex, OpenCode, and Gemini CLI. This adds the skill files to your agent and changes nothing in your cloud yet:
Add the Qovery MCP Server to Claude Code. It opens an OAuth sign-in and stays read-only by default:
BASH
claude mcp add --transport http qovery https://mcp.qovery.com/mcp --callback-port 4242
Only if you want the agent to change resources, add the write-enabled server. Write access must also be turned on in the Qovery Console:
BASH
claude mcp add --transport http qovery "https://mcp.qovery.com/mcp?read_write=true" --callback-port 4242
The skills include qovery-onboard, qovery-deploy, qovery-troubleshoot, qovery-optimize, qovery-speedup, qovery-preview, and qovery-terraform. Then ask something like: "deploy this Rails app and a managed PostgreSQL 18 to my AWS account with Qovery". The database migration itself (fork, dump, restore, validate) stays a step you run and approve.
When does Heroku stop creating Postgres 15 databases?
Heroku stopped allowing new Postgres 15 databases on November 11, 2026. Forks and followers of existing Postgres 15 databases can still be created. Forced upgrades start January 28, 2027, and after the February 28, 2027 end-of-life date Heroku starts restricting access to and deleting remaining Postgres 15 databases.
What happens to my existing Heroku Postgres 15 database?
It keeps running until Heroku's forced upgrades start on January 28, 2027. If you have not upgraded or migrated by then, Heroku schedules the upgrade for you. After February 28, 2027, Heroku restricts access to and deletes databases still on Postgres 15, including followers.
Is the Postgres 15 deprecation a reason to move my database to my own cloud account?
It is a good trigger if you were already considering leaving Heroku, need the database in your own VPC, or want upgrades on the community's five-year support window instead of Heroku's three. A pg_dump and pg_restore move into Postgres 17 or 18 on RDS, Cloud SQL, or Azure doubles as your major version upgrade. If your app is staying on Heroku, a pg:upgrade to Postgres 18 is the simpler answer.
Can I migrate Heroku Postgres to Amazon RDS with zero downtime?
Not with standard replication tools. Heroku does not support logical replication to external instances and does not grant the superuser or replication roles that AWS DMS change data capture requires. Plan a dump-and-restore cutover with a write freeze sized by a rehearsal on a Heroku fork, or build application-level dual writes if you cannot freeze writes at all.
How long does Amazon RDS support PostgreSQL 15?
RDS keeps PostgreSQL 15 in standard support until February 29, 2028. Paid RDS Extended Support then runs until February 28, 2031. Heroku's end of life for Postgres 15 is February 28, 2027, a year earlier than RDS standard support ends.
Which Postgres version should I upgrade to on Heroku?
Postgres 18, if your extensions and drivers support it. It is Heroku's default since August 4, 2026, and under Heroku's policy Postgres 16 deprecates as soon as Heroku releases Postgres 19, so picking 16 or 17 brings the next deadline closer. Test on a fork or with the copy method first.
Melanie leads content at Qovery. She covers platform engineering trends, Kubernetes operations, FinOps, and the tools that help engineering teams ship faster.
Next step
Own your Postgres upgrade calendar.
Qovery deploys your app and a managed PostgreSQL (Amazon RDS on AWS) into your own cloud account, with git-push deployments and a preview environment per pull request. Start deploying in under 10 minutes.