On this page · 10 sections
- What happened on 12 November 2026
- Why 14.25 still matters: install it first
- What end of life means on AWS, Azure and Google Cloud
- Choose the target: 17, 18 or 19
- What changes between 14 and 18
- How to upgrade: three methods
- A 10-week plan to be off 14 before February
- For teams in India, the UK and the US
- How eCorpIT can help
- References
Summary. PostgreSQL 14 is now end of life. On 12 November 2026 the PostgreSQL Global Development Group released 18.7, 17.12, 16.16, 15.20 and 14.25, and said 14.25 is the final release of the 14 branch, which will get no further security or bug fixes. The same update closed 52 vulnerabilities; by our count of the project's security page, 40 of them list 14.25 as the fixed version, and 27 of those score 8.8 on CVSS 3.1. PostgreSQL 19 is at Release Candidate 2, with general availability planned for 19 November 2026. Managed services give a short grace period, then charge. Azure ends standard support on 31 January 2027, Cloud SQL starts extended support on 1 February 2027, and Amazon RDS starts Extended Support billing on 1 March 2027. AWS has not yet listed a PostgreSQL 14 rate; in US East (Ohio) it charges $0.10 per vCPU-hour for versions 10 to 13. Our verdict: install 14.25 this week and run its reindex steps, then upgrade to PostgreSQL 18 (supported to 14 November 2030) or 17 (to 8 November 2029) before February. Treat 19 as the next upgrade, not this one.
About this article. This is for database administrators, platform engineers and CTOs in the US, UK and India who still run PostgreSQL 14, self-managed or on AWS, Azure or Google Cloud. We checked every version number, date and price on 16 November 2026 against postgresql.org (the versioning policy, the 12 November release announcement, the 14.25 release notes and the security page) and against the AWS, Microsoft and Google documentation and price lists for their managed services. Cloud prices and dates change, so confirm them on the provider pages before you budget. This is general technical guidance, not advice for a specific system, and no vendor paid for inclusion. Researched with AI assistance and checked against the primary sources listed in the references.
What happened on 12 November 2026
The PostgreSQL project supports each major version for five years after its first release, then ships one last minor release and stops. PostgreSQL 14 first shipped on 30 September 2021, and its final release, 14.25, shipped on 12 November 2026. The release announcement carries an end-of-life notice: 14 will no longer receive security and bug fixes, and production users should plan an upgrade. The 14.25 release notes say this is expected to be the last 14.x release.
The versioning page lists each branch and its final date. We have put them beside the next release.
| Version | Current minor | First release | Final release (end of life) |
|---|---|---|---|
| 14 | 14.25 (final) | 30 September 2021 | 12 November 2026 |
| 15 | 15.20 | 13 October 2022 | 11 November 2027 |
| 16 | 16.16 | 14 September 2023 | 9 November 2028 |
| 17 | 17.12 | 26 September 2024 | 8 November 2029 |
| 18 | 18.7 | 25 September 2025 | 14 November 2030 |
| 19 | RC 2 | Planned 19 November 2026 | Not yet listed |
The dates for 14 to 18 come from the project's versioning policy page; the 19 date comes from the release announcement and the developer roadmap. The project schedules minor releases on the second Thursday of February, May, August and November, and the next ones are 11 February, 13 May, 12 August and 11 November 2027. None of them will include a 14.x build.
Why 14.25 still matters: install it first
The last release is also a very large security release. The announcement lists 52 CVEs and more than 140 bug fixes across the supported branches. The project's security page records the fixed versions for each CVE: 40 of the 52 list 14.25, and 27 of those 40 carry a CVSS 3.1 base score of 8.8. Most of the 27 are described as letting a user with an ordinary database login execute arbitrary code. That is our count from the page on 16 November; check the page for any CVE that matters to you.
Two consequences follow. First, a server still on 14.24 or earlier is exposed to dozens of published flaws with fixes available, so the minor update comes before any major upgrade project. Second, every flaw found from now on will be fixed in 15 to 19 only.
Applying 14.25 needs no dump or pg_upgrade: stop the server, install the new binaries and restart. But the release notes ask for follow-up work, because several fixed bugs could already have corrupted indexes.
- Rebuild C-language extensions. 14.25 adds checks against composite types being changed while queries use them. Full protection needs C-language extensions rebuilt, and some extensions need source changes. A new setting,
composite_type_mismatch, can turn the checks off, but the project warns this carries security risk.
- Reindex B-tree indexes on bpchar columns. Deduplication was unsafe for columns declared as
bpcharwith no length, and could make index-only scans return wrong results. The fix stops new damage but does not repair existing entries. Columns declared aschar(n)are not affected.
- Reindex pg_trgm GiST indexes. If you use the pg_trgm extension, reindex indexes built with the
gist_trgm_opsoperator class. The fix does not repair existing indexes.
- Reindex GIN indexes you suspect. Two GIN bugs could leave entries pointing at unrelated rows or make searches miss rows.
- Reindex tables that report missing chunk number errors. A bug affecting
REINDEX CONCURRENTLYon TOAST indexes could lose index entries; the table data itself is intact.
A query like this lists the pg_trgm GiST indexes in a database:
SELECT i.indexrelid::regclass AS index_name,
i.indrelid::regclass AS table_name
FROM pg_index i
JOIN pg_opclass o ON o.oid = ANY (i.indclass::oid[])
WHERE o.opcname = 'gist_trgm_ops';
On a managed service you do not install binaries yourself. Amazon RDS commits to offering new community minor versions within 7 days of release, and Azure applies minor versions during your maintenance window, region by region. On 16 November, Google's Cloud SQL version page still listed 14.24 as its PostgreSQL 14 minor version. Check the minor version each instance actually reports before you call it patched.
What end of life means on AWS, Azure and Google Cloud
Managed services do not switch PostgreSQL 14 off on 12 November. Each keeps it running for a short period, then enrols your instances in a paid extended support programme automatically.
| Service | Standard support ends | Extended support starts | Extended support ends |
|---|---|---|---|
| Amazon RDS for PostgreSQL | 28 February 2027 | 1 March 2027 (year 3 pricing from 1 March 2029) | 28 February 2030 |
| Azure Database for PostgreSQL flexible server | 31 January 2027 | 1 February 2027 | 11 November 2029 |
| Google Cloud SQL for PostgreSQL | Regular support to 31 January 2027 | 1 February 2027 | Deprecated 1 February 2030 |
These dates come from the AWS release calendar, Microsoft's version policy and extended support pages, and Google's version policy page.
Four details in the provider pages change how you plan:
- You cannot opt out by doing nothing. Azure enrols servers on unsupported versions automatically, and upgrading is the way to stop the charges. Google does the same for Cloud SQL, and after three years in extended support it upgrades a deprecated version to the default version during regular maintenance.
- Azure stops engine patches at community end of life. Its version policy says that once the community retires a version, Azure stops applying bug and security patches to the database engine, though it keeps patching the host. Its extended support programme then offers security patches and critical bug fixes for up to three extra years.
- AWS charges per vCPU, and the rate doubles in year 3. AWS's own worked example uses PostgreSQL 12 in US East (Ohio): $0.100 per vCPU-hour in years 1 and 2, then $0.200, on top of normal instance charges, and Reserved Instance discounts do not apply. Its price list for Ohio shows the same $0.10 and $0.20 for RDS for PostgreSQL 10, 11, 12 and 13, but no PostgreSQL 14 line yet.
- AWS bills standbys too. Extended Support charges apply to standby instances in Multi-AZ deployments, and start the day after the end of standard support.
A worked example, assuming AWS prices 14 like 10 to 13. One 8-vCPU instance at $0.10 per vCPU-hour for 730 hours is $584 a month (8 × 730 × 0.10), or $7,008 a year (8 × 8,760 × 0.10). At the year 3 rate it is $14,016 a year. A Multi-AZ pair doubles both figures. At the ECB reference rates of 13 November 2026, crossed and rounded to two decimals (₹94.24 per US dollar, ₹124.65 per pound, so $1 is about £0.76), $7,008 is about ₹6.60 lakh or £5,298 a year. Your rate depends on region and on the vCPUs of every billed instance, so price your own fleet once AWS lists 14.
The Azure and Cloud SQL deadlines arrive first. From today, 16 November, there are 76 days to 31 January 2027 and 104 days to 28 February 2027. Leave room for the December holidays, when many teams freeze changes.
Choose the target: 17, 18 or 19
Version 15 is not a target: it reaches end of life on 11 November 2027, a year from now. Version 16 buys two years. The real choice is between 17, 18 and the 19 release due on 19 November.
| Factor | PostgreSQL 17 | PostgreSQL 18 | PostgreSQL 19 |
|---|---|---|---|
| Status on 16 Nov 2026 | Stable, 17.12 | Stable, 18.7 | Release Candidate 2 |
| Community support ends | 8 November 2029 | 14 November 2030 | Not yet listed |
| Time in production | Over two years | Over a year | None yet |
| Planner statistics kept by pg_upgrade | No | Yes | Yes |
| Data checksums on by default in initdb | No | Yes | Yes |
| MD5 passwords | Supported | Deprecated, warnings on CREATE or ALTER ROLE | Warning after each MD5 login |
| Best fit | Conservative estates, extensions not yet certified on 18 | Most upgrades from 14 this winter | The next upgrade, after early minor releases |
Sources for the table: the versioning page, the PostgreSQL 18 announcement and release notes, and the PostgreSQL 19 release notes.
Why 18 is our default. It gives four years of support from today instead of three. Its pg_upgrade keeps planner statistics through the upgrade, so a busy system does not run on poor plans while ANALYZE catches up, and it adds parallel checks with --jobs and a new --swap mode. It also adds asynchronous I/O, for which the project reports gains of up to three times in certain benchmark scenarios. Our PostgreSQL 18 tuning guide covers the AIO settings.
When 17 is the better call. If you depend on extensions or vendor software that do not yet certify 18, take 17. It still brings incremental backups with pg_basebackup and pg_combinebackup, a vacuum memory structure that uses up to 20 times less memory, and pg_createsubscriber for turning a physical replica into a logical one.
Why not 19 now. The project says the RC will be mostly identical to the final release, though more fixes may land before general availability, and the RC 2 announcement already lists fixes made since RC 1. It also removes RADIUS authentication and forces standard_conforming_strings on, which can break old dumps and clients. Our guide to the PostgreSQL 19 changes lists what to test. Moving from 14 to 18 now and to 19 later in its life is less risky than betting a deadline-driven upgrade on a brand-new major. For a fuller comparison, see our PostgreSQL 14 upgrade target decision guide.
What changes between 14 and 18
You can upgrade from 14 straight to 18 without stopping at 15, 16 or 17, but the project recommends reading the release notes of every major version in between. These are the incompatibilities most likely to affect a typical application.
| Version | Change | What to check |
|---|---|---|
| 15 | New databases no longer let every user create objects in the public schema | Upgraded databases keep old permissions; scripts that create fresh databases may fail |
| 16 | Role inheritance is set at GRANT time; several old settings removed | Role grants, and config files using promote_trigger_file or vacuum_defer_cleanup_age |
| 17 | Maintenance commands run functions with a safe search_path | Functions used in expression indexes or materialised views must set their own search_path |
| 17 | old_snapshot_threshold removed; adminpack removed | Config files and monitoring scripts |
| 18 | initdb enables data checksums by default | pg_upgrade needs matching settings; use --no-data-checksums if 14 had none |
| 18 | MD5 passwords deprecated | Move roles to SCRAM before 19, where MD5 logins warn |
| 18 | AFTER triggers run as the role active when the event was queued | Code that switches roles inside a transaction |
| 18 | VACUUM and ANALYZE now process inheritance children | Maintenance jobs on inherited tables; use ONLY for the old behaviour |
| 18 | Full text search reads its files with the cluster's default collation provider | Reindex full text search and pg_trgm indexes after pg_upgrade on ICU or builtin clusters |
The table summarises the migration sections of the 15, 16, 17 and 18 release notes.
How to upgrade: three methods
The project documents three ways to move to a new major version: dump and restore, pg_upgrade, and logical replication.
| Method | Downtime | Rollback | Best for |
|---|---|---|---|
| pg_dump and restore | Long, grows with data size | Old cluster untouched | Small databases, or a move to new hardware |
| pg_upgrade with --link or --swap | Minutes on most clusters | Limited once the new cluster has started | Large self-managed clusters on one host |
| Logical replication to a new 18 server | Seconds at switchover | Keep the 14 primary until you are sure | Large databases that cannot take an outage |
The downtime figures are the project's own descriptions: pg_upgrade can run in minutes, particularly in link mode, and a logical replication switchover takes only several seconds.
pg_upgrade. It supports upgrades from 9.2 and later, so 14 to 18 is a single step. Always run it with --check first; that leaves the old cluster unmodified. Link mode is fast and uses little extra disk, but the old cluster cannot be used once the new one starts. Install the new version's builds of every extension's shared library, and do not run CREATE EXTENSION in the new cluster, because pg_upgrade brings the definitions across.
# Dry run: checks both clusters and changes nothing
/usr/lib/postgresql/18/bin/pg_upgrade \
--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 \
--jobs=4 --check
Paths vary by distribution. After the real run, rebuild the statistics pg_upgrade did not carry over; the 18 documentation suggests vacuumdb --all --analyze-in-stages --missing-stats-only, then vacuumdb --all --analyze-only.
Logical replication. Build an 18 server, subscribe it to the 14 primary, let it catch up, then switch the application over. One limit applies to everyone leaving 14: PostgreSQL 17 made logical replication slots survive a major upgrade, but only for upgrades from 17 onward. If you already use logical replication on 14, plan to recreate those slots and subscriptions.
Managed services. Google documents two routes off an end-of-life version in Cloud SQL: an in-place major version upgrade, or migrating the data to a new instance. Whichever route your provider offers, rehearse it on a restored snapshot first, because every extension version and parameter setting must be valid for the target.
A 10-week plan to be off 14 before February
| Week | Dates | Work |
|---|---|---|
| 1 | 16 to 22 November | Install 14.25 everywhere; run the reindex steps; list every 14 instance, size, extension and owner |
| 2 | 23 to 29 November | Pick 17 or 18 per system; check extensions and drivers support the target; find MD5 roles and move them to SCRAM |
| 3 to 4 | 30 November to 13 December | Restore a recent backup and run pg_upgrade --check or a managed upgrade on the copy; fix every blocker |
| 5 to 6 | 14 to 27 December | Run application tests and compare query plans and timings on the upgraded copy; agree change windows |
| 7 to 9 | 28 December to 17 January | Upgrade non-production, then lower-risk production systems; rebuild statistics; watch errors |
| 10 | 18 to 24 January | Upgrade the remaining production systems, leaving a week before Azure's 31 January date |
If you run on RDS only, you have four more weeks before billing starts on 1 March, but there is little reason to spend them: 14 has received its last community fix wherever it runs.
Three checks that help with week 1 and week 2 work, run as a superuser:
SHOW data_checksums;
SELECT extname, extversion FROM pg_extension ORDER BY extname;
SELECT rolname FROM pg_authid WHERE rolpassword LIKE 'md5%';
The first tells you whether to initialise the 18 cluster with --no-data-checksums for pg_upgrade. The second gives you the extension list to check against your target. The third lists roles that still use MD5 passwords, which 18 deprecates.
For teams in India, the UK and the US
The upgrade work is the same everywhere; the pressure points differ. Indian teams on AWS Mumbai or Hyderabad, or on Azure Central India, should price extended support for their own region rather than the US example, because AWS says rates vary by region. UK and EU teams with security audits coming up should be ready to explain any end-of-life database engine still in service, and the date it will be gone. US teams on Cloud SQL or Azure face the 1 February date first.
If PostgreSQL 15 is also in your estate, it ends on 11 November 2027, so plan it into the same programme; our PostgreSQL 15 end of life guide has the dates.
How eCorpIT can help
eCorpIT is a Gurugram-based technology organisation, founded in 2021, assessed at CMMI Level 5 and MSME certified, with senior-led engineering teams working across AWS, Microsoft and Google platforms. Our cloud computing services team can inventory your PostgreSQL 14 estate, rehearse the upgrade to 17 or 18 on a copy and run the production cutover on RDS, Azure, Cloud SQL or your own servers. Talk to us at /contact-us/.
Last updated: 16 November 2026.