Key takeaways
- PostgreSQL 14 end of life is 12 November 2026. The final release, 14.25, shipped the same day, and the project will not patch that branch again.
- On RDS, Aurora, Azure, and Cloud SQL, the cost of PostgreSQL 14 end of life appears first on the invoice, with paid extended support starting between December 2026 and March 2027.
- Choose the method by downtime budget: in-place for minutes, logical replication for seconds, and dump and restore when simplicity matters more than speed.
- Rehearse once on a restored copy of production. The nasty surprises are almost always an extension or a query plan, rarely the engine itself.
Your application is running. The dashboards are green. The only sign that anything changed on 12 November is one line in a release announcement that most of your team will never read.
A few weeks later, the problems begin. Your security scanner flags the database as running unsupported software. An auditor asks how you are patching it. Finance asks why the cloud bill has gone up. At that point, the upgrade is no longer a planned project but an emergency. That is a bad time to find out that an extension doesn’t support the new version, or that your 2 TB database won’t finish upgrading within your maintenance window.
PostgreSQL 14 reached end of life on Thursday, 12 November 2026. The PostgreSQL project released 14.25 as the final update for version 14. It came out on the same day as updates 18.7, 17.12, 16.16, and 15.20. There will be no more fixes for version 14, including security fixes.
PostgreSQL end-of-life dates for versions 12 to 19
The PostgreSQL version support policy has been the same for years. A major version gets five years from its first release, then one last minor release, then nothing. The final releases land in November, which is why every end date below sits in that month.
| Version | First release | Last community release | Status today |
|---|---|---|---|
| 12 | 3 Oct 2019 | 21 Nov 2024 | Unsupported |
| 13 | 24 Sep 2020 | 13 Nov 2025 | Unsupported |
| 14 | 30 Sep 2021 | 12 Nov 2026 | Unsupported |
| 15 | 13 Oct 2022 | 11 Nov 2027 | Supported |
| 16 | 14 Sep 2023 | 9 Nov 2028 | Supported |
| 17 | 26 Sep 2024 | 8 Nov 2029 | Supported |
| 18 | 25 Sep 2025 | 14 Nov 2030 | Supported (current) |
| 19 | 19 Nov 2026 | Not published; roughly Nov 2031 | Supported (newest) |
Version 19 went GA on 19 November 2026, a week after 14 reached end of life. Its end date isn’t published yet, so read 2031 as an estimate. If you are still on 13 or 12, you are already past the line, and everything below applies to you with less time to spare. Keep these PostgreSQL end-of-life dates where your team will actually see them, because the same question comes back every November.
PostgreSQL 14 support ended on 12 November 2026. Version 18 has the longest runway, to November 2030.
What PostgreSQL 14 end of life means for a running system
PostgreSQL 14 reached its official community end-of-life (EOL) on November 12, 2026, meaning the database will no longer receive security patches, bug fixes, or maintenance updates from the PostgreSQL Global Development Group.
- It will not crash or shut down: The database service will continue to start up, accept connections, process SQL queries, and run scheduled backups normally on and after November 13, 2026.
- Zero security patches: If a critical security vulnerability or data-corruption bug is discovered in PostgreSQL 14 post-EOL, the community will not release a fix for version 14 (though updates may land in supported versions like 15, 16, 17, or 18).
- Compliance and audit risks: Running unsupported software often violates enterprise security frameworks, SOC 2, HIPAA, PCI-DSS, or ISO compliance standards.
- Ecosystem and driver drift: Over time, third-party drivers, cloud platforms, ORMs, and database extensions will drop PostgreSQL 14 from their official test matrices and support scopes.
PostgreSQL 14 end of life on RDS, Aurora, Azure, and Cloud SQL
Managed services buy you time, and they charge for it. All three clouds enrol PostgreSQL 14 instances in paid extended support automatically once their own standard support ends. None of them waits for the team to opt in.
Amazon RDS and Aurora: Standard support for version 14 ends on 28 February 2027. Starting 1 March, extended support costs $0.10 per vCPU-hour in US East for two years, rising to $0.20 on 1 March 2029 until coverage ends on 28 February 2030. Aurora shares the same timeline and rates. Charges apply to all instances, including Multi-AZ standbys. An eight-vCPU deployment (primary plus standby) costs about $584/month extra initially, increasing to ~$1,168/month in year three. AWS continues releasing security fixes during extended support. Charges stop only when instances are upgraded or deleted, not when stopped.
Azure Database for PostgreSQL: Microsoft’s published pages don’t agree with each other. Some list 11 December 2026 as the last day of standard support, others 31 January 2027, with paid cover running to 11 November 2029. Plan around the earlier date and confirm in your portal. One more catch: extensions such as pgvector and TimescaleDB are not upgraded for you during a major version change, so you update them yourself afterwards.
Google Cloud SQL: Instances on 14 move into extended support on 1 February 2027, and the version is deprecated on 1 February 2030. We couldn’t find a published price for 14 on Google’s or Microsoft’s pages, so check your region’s pricing page before you put a number in a budget.
All three clouds enrol PostgreSQL 14 in paid extended support automatically. Azure bills first, then Cloud SQL, then RDS and Aurora.
Which version should you upgrade to after PostgreSQL 14 reaches end of life? (17, 18, or 19)
Move to PostgreSQL 18 if you can. It is supported until November 2030, and an in-place upgrade to it keeps your planner statistics. Choose 17, supported until November 2029, if an extension or vendor isn’t certified on 18 yet. Skip 15 and 16, which end in November 2027 and November 2028. Wait on 19 until it has received its first minor release.
Move to PostgreSQL 18 if you can. Choose 17 only if an extension or vendor is not certified on 18 yet.
18 is supported until November 2030, the longest runway of anything you can run in production today. It is also the first release where the in-place tool carries your planner statistics across, so the slow first hour after cutover mostly disappears. It also adds a faster directory-swap mode and asynchronous I/O. One catch, covered below: New 18 clusters enable data checksums by default.
17 runs to November 2029 and has had longer to collect extension certifications. If PostGIS, TimescaleDB, or a vendor-packaged app isn’t certified on 18 yet, upgrade PostgreSQL 14 to 17 now and make the 17-to-18 hop later. That second step is much gentler.
Versions 15 and 16 only delay the problem. Under the PostgreSQL version support policy, 15 ends in November 2027 and 16 in November 2028, and you would repeat this whole exercise while 18 sits there.
Delay adopting 19. Postgres 19 is out, but wait for its first minor release (expected around 11 Feb 2027) before running it in production.
Which is the best way to upgrade after PostgreSQL 14 reaches end of life? Methods Compared
Use an in-place upgrade if you can accept a few minutes of downtime. Use logical replication if you can only accept seconds. Use dump and restore if your database is small or you have a long maintenance window. Your downtime budget decides it, not the size of your team.
| Method | Typical downtime | Rollback | Best for | Main catch |
|---|---|---|---|---|
| In-place upgrade (copy, clone, link, or swap) | Minutes | Easy in copy mode; backup restore in link or swap mode | Large databases with a short maintenance window | Link and swap leave no clean way back |
| Logical replication | Seconds | Easy, because the old cluster stays untouched until cutover. | Busy systems that cannot go down | Needs primary keys, manual sequence sync, and two clusters running at once |
| Dump and restore | Hours, growing with data size | Easy, because the old cluster stays untouched | Small databases, or when you want a clean rebuild | Longest downtime of the three |
How does an in-place upgrade work?
An in-place upgrade with pg_upgrade reuses your existing data files and rewrites only the system catalogs. That is why even a multi-terabyte cluster can finish in minutes. You choose how the files are transferred:
- Copy: safest and slowest, because it duplicates every file.
- Clone: fast on file systems that support it.
- Link: very fast, but the old and new clusters share files. Once the new server starts, going back means restoring from backup.
- Swap (new in PostgreSQL 18): moves directories into place and is fastest on clusters with a huge number of objects. The old cluster cannot be used afterwards.
--copyDuplicates every file. Safest and slowest.--cloneFast on file systems that support it.--linkVery fast. Old and new clusters share files.--swapMoves directories into place. Fastest with a huge number of objects.Copy and clone keep a way back. Link and swap trade it for speed, so take a tested backup first.
How does logical replication work for a major version upgrade?
You build a new cluster on the target version, stream changes into it from the old one, and switch your application over once it has caught up. Downtime shrinks to the few seconds of the switch, and you can rehearse for weeks beforehand.
The preparation is heavier:
- Every table needs a primary key or a replica identity.
- Sequences are not replicated, so sync them by hand before cutover.
- Schema changes are not replicated, so freeze them during the migration.
- You pay for two clusters while they overlap.
When is dumping and restoring the right choice?
Dump and restore is the slowest method and the hardest to get wrong. Your downtime equals dump time plus restore time, and it grows with your data. In return, you get a freshly written cluster with no bloat, and you can enable data checksums along the way. It suits databases up to a few hundred GB or any case where a weekend window is acceptable.
How do I choose between them?
Ask one question: how long can the application be unavailable?
- Seconds: logical replication.
- Minutes: in-place upgrade.
- Hours: dump and restore.
Whichever you pick, rehearse it once on a restored copy of production before the real cutover.
Seconds means logical replication, minutes means an in-place upgrade, hours means dump and restore.
A practical upgrade checklist
Here is the order we follow for a PostgreSQL 14 end-of-life project, self-managed or in the cloud.
- Find every instance of 14: Primaries, replicas, standbys, staging, and the analytics box somebody built in 2022. In the cloud, query your inventory by engine version.
- Check extensions against the target: Compare what you use with pg_available_extensions on a target-version server. PostGIS, pgvector, and TimescaleDB are the usual blockers.
- Read the release notes for 15, 16, 17 and 18: Note removals, renamed functions, and changed defaults.
- Rehearse on a restored copy of production: Time each phase. Capture your slowest queries with pg_stat_statements first, so you can compare afterwards.
- Settle the checksum question: New 18 clusters enable checksums, and an in-place upgrade requires old and new to match. Either create the new cluster without them or enable them on the old one while it’s offline.
- Diff your configuration: The new cluster starts on default settings, not your tuned postgresql.conf. Carry over pg_hba.conf too.
- Take a backup and restore it: A backup you haven’t restored is a hope.
- Run the dry-run check against real data and fix what it reports:
pg_upgrade --check \
--old-bindir=/usr/lib/postgresql/14/bin --new-bindir=/usr/lib/postgresql/18/bin \
--old-datadir=/var/lib/postgresql/14/main --new-datadir=/var/lib/postgresql/18/main
- Cut over, then clean up: Run the update_extensions.sql the tool generates, then vacuumdb --all --analyze-in-stages --missing-stats-only to rebuild statistics that weren’t carried over.
- Watch for 48 hours: Keep the old cluster or snapshot until you trust the new one. On RDS, the equivalent is a snapshot, a pre-check, and a trial on a restored copy.
- Find every instance of 14Primaries, replicas, standbys, staging, the forgotten analytics box
- Check extensions against the targetPostGIS, pgvector and TimescaleDB are the usual blockers
- Read release notes for 15 to 18Removals, renamed functions, changed defaults
- Rehearse on a restored copy of productionTime each phase. Capture slow queries with pg_stat_statements first
- Settle the checksum questionNew 18 clusters enable checksums. Old and new must match
- Diff your configurationCarry over postgresql.conf tuning and pg_hba.conf
- Take a backup and restore itA backup you haven't restored is a hope
- Run the dry-run check on real data
pg_upgrade --check, then fix what it reports - Cut over, then clean upRun update_extensions.sql, then the missing-stats vacuumdb pass
- Watch for 48 hoursKeep the old cluster or snapshot until you trust the new one
The order we follow for a PostgreSQL 14 end-of-life project, self-managed or in the cloud.
What Goes Wrong After a PostgreSQL 14 End of Life Upgrade? 8 Common Problems and Fixes
Most trouble after a PostgreSQL 14 end of life upgrade isn’t dramatic. It’s small things that stack up on day one.
- Slow queries at first: Before 18, planner statistics didn’t survive an upgrade. 18 keeps most of them, but statistics created with CREATE STATISTICS are left behind, so run the missing-stats pass.
- Old extension versions: Extensions aren’t updated automatically. Run ALTER EXTENSION ... UPDATE for each one.
- Settings back at defaults: Memory, autovacuum, and connection settings quietly revert if you didn’t copy them.
- Permission errors on the public schema: Since 15, ordinary users can’t create objects there by default. Apps that relied on it start failing with “permission denied.”
- Broken backup scripts: 15 removed the old exclusive backup functions, so scripts calling pg_start_backup stop working.
- Search path surprises: From 17, maintenance operations use a safe search path. Functions behind expression indexes or materialized views that relied on the old behavior can break, so schema-qualify names.
- Lost replication slots: Logical slots aren’t carried over from 14. Recreate subscriptions and change-capture connectors such as Debezium.
- Password warnings: 18 deprecates MD5 passwords. Move roles to SCRAM before the warnings turn into a policy.
Most trouble after a PostgreSQL 14 end-of-life upgrade isn't dramatic. It's small things that stack up on day one.
How Geopits Helps With Your PostgreSQL 14 Upgrade
Most teams aren’t stuck on the commands. They’re stuck on the unknowns: which of the forty databases matter, whether an extension will build on the new version, and how long the downtime window really is.
Geopits starts with an inventory and risk map of everything still on 14. The upgrade is then rehearsed on a copy of your production data, so the downtime figure is measured, not guessed. The method is chosen against your downtime budget, and the runbook includes rollback. Geopits can run the cutover alongside your team and stay on through the tuning week afterwards. If PostgreSQL 14 end of life is already a line on your risk register, now is the right time to bring us in.
Frequently asked questions
1. When is PostgreSQL 14's end of life?
12 November 2026. The final release, 14.25, shipped that day. The community will not issue any more fixes for version 14, including security fixes.
2. Is it safe to keep running PostgreSQL 14 after that date?
It will keep running, but it won’t be patched. If the database is reachable from the internet or holds regulated data, don’t stay on it. If it sits on a private network with tight roles, segmentation, and monitoring, you can buy a few months. Plan in weeks, not years.
3. How much does RDS PostgreSQL extended support cost for version 14?
It starts on 1 March 2027 at $0.10 per vCPU-hour in US East, then rises to $0.20 from 1 March 2029. Multi-AZ standbys are billed too, and other regions differ. At list price, 8 vCPUs across primary and standby come to about $584 a month. The charge stops once you upgrade.
4. Can I upgrade PostgreSQL 14 to 17 or 18 in one step?
Yes. The in-place upgrade tool supports jumping from any release since 9.2, and logical replication works across versions. Testing is the real limit, not the tool, because you cross several major versions of behavior changes. On RDS, check the console for the list of valid targets from your current minor version.
5. What does PostgreSQL 14 end of life mean for managed databases?
Your provider continues to run the engine and charges you for extended support. Billing starts between December 2026 and March 2027 depending on the cloud: Azure first, then Cloud SQL on 1 February, then RDS and Aurora on 1 March. When extended support ends in 2029 or 2030, the provider forces the upgrade on its own schedule.


