PostgreSQL 14 reached end of life on November 12, 2026, with 14.25 as its final minor release, according to the official PostgreSQL versioning policy. It no longer receives security fixes from the project. Upgrade straight to PostgreSQL 18 (supported until November 14, 2030), or to 17 (supported until November 8, 2029) if an extension or platform you depend on is not ready for 18, using pg_upgrade --link when a short maintenance window is acceptable, or logical replication when it is not.
Dates from the official calendar, checked November 21, 2026. The project supports each major version for five years after its first release, then ships one final minor version. Moving from 14 to 15 or 16 would buy you one or two years before you repeat this exercise, which is why we suggest jumping further.
The PostgreSQL Global Development Group ships bug fixes and, when needed, security fixes in minor releases at least once every three months. Version 14 is now marked "Supported: No" on the versioning page. Fixes for problems found from here on will go to supported versions, not to 14. Your database keeps running exactly as before; what disappears is the stream of patches.
Three practical consequences follow.
If you run 14 on a managed service, your provider may have its own timeline or paid extended support. We have not verified any provider's current policy for this page, so check your provider's documentation directly rather than relying on a blog post.
Most teams know about their main production cluster. The clusters that surprise them are the reporting replica someone built two years ago, a Docker image pinned in a test suite, and the analytics database a vendor set up. Search everywhere a version string can hide.
For the file search, the EOL deadline radar helps: it is free, accepts a pasted Dockerfile, compose file or .tool-versions, and flags each pinned version against its official cutoff with dates.
You can upgrade from 14 directly to 18 without installing 15, 16 or 17 in between, but the official upgrade guide tells you to read the major release notes of every intervening version, especially their "Migration" sections. These are the items that most often bite a 14 cluster.
The PostgreSQL 15 release notes remove the default CREATE permission for PUBLIC on the public schema, and make the new pg_database_owner role its owner. Upgrades and dump restores keep your existing permissions, so an upgraded database behaves as before. New databases created afterward do not. Migration tooling or tenant provisioning scripts that create a fresh database and expect any role to create tables in public will fail until you grant the privilege explicitly.
Version 15 removed the long-deprecated exclusive backup mode. pg_start_backup() and pg_stop_backup() became pg_backup_start() and pg_backup_stop(), and pg_backup_start_time() and pg_is_in_backup() were removed. Homegrown backup scripts written years ago against 9.x or 10 often still call the old names. Grep your backup tooling before cutover, not after the first nightly job fails.
plpythonu and plpython2u no longer exist. If any function uses LANGUAGE plpythonu, pg_upgrade --check will flag it and the function must be rewritten for plpython3u before you can move.
The PostgreSQL 16 release notes restrict CREATEROLE: changing other roles, including adding members, now requires ADMIN OPTION on those roles. A role's inheritance setting is also fixed when a membership is granted, rather than applying to existing memberships. User-provisioning jobs, Vault-style dynamic credential engines and admin scripts that ran under a CREATEROLE account deserve a test against 18 in staging.
Per the PostgreSQL 17 release notes, ANALYZE, CREATE INDEX, REINDEX, VACUUM, CLUSTER and REFRESH MATERIALIZED VIEW now run with a safe search_path. Functions used in expression indexes or materialized views that refer to objects in other schemas must set their search_path explicitly, or these commands will fail to find those objects after the upgrade.
pg_stat_bgwriter lost buffers_backend and buffers_backend_fsync (pg_stat_io covers them). In pg_stat_statements, blk_read_time and blk_write_time became shared_blk_read_time and shared_blk_write_time. old_snapshot_threshold was removed. Dashboards and alert queries written for 14 will show blanks or errors. Update them in the same change as the upgrade so you are not blind on day one.
The PostgreSQL 18 release notes say initdb now enables data checksums by default, and pg_upgrade requires the old and new clusters to match. Many 14 clusters were created without checksums, so initialize the 18 cluster with initdb --no-data-checksums if you plan to use pg_upgrade (logical replication does not have this constraint). MD5 passwords are deprecated, with removal planned in a future major release; CREATE ROLE and ALTER ROLE now warn when setting one.
VACUUM and ANALYZE now process inheritance children unless you add ONLY. COPY FROM no longer treats a line containing \. as end of file in CSV data. RESTRICT foreign key violations report SQLSTATE 23001 instead of 23503, which matters if application code catches the old code. On clusters using ICU or the builtin collation provider, full-text search follows the default provider, and the notes say to reindex full-text search and pg_trgm indexes after pg_upgrade.
For a past-deadline version, the plan should be short and decisive. The key decision is the cutover method, and it follows from how much write downtime you can accept.
pg_upgrade --link upgrades the cluster in place. The pg_upgrade documentation says link mode uses hard links instead of copying files, which is much faster and uses less disk space, and the upgrade guide says upgrades can be performed in minutes with it. The trade-off is explicit: once you start the new cluster, you cannot go back to the old one without restoring from backup. The data directories must be on the same file system. Choose this when a planned window is acceptable and you have a tested backup.
Logical replication builds a new 18 server, replicates from 14, and switches over. The upgrade guide says this works because logical replication supports replication between different major versions, and that the switchover results in only several seconds of downtime. The costs are setup work and restrictions: schema and DDL are not replicated, sequence changes are not replicated, large objects are not replicated, and tables need a replica identity (the primary key by default) for UPDATE and DELETE to replicate. On 14, the publisher needs wal_level set to logical. Choose this for clusters where a long write freeze is unacceptable, and when you want the old 14 primary intact as a fallback.
A dump and restore (pg_dump -Fd -j with pg_restore -j, plus pg_dumpall --globals-only for roles) is the simplest option for small databases and the only clean way to change hosts and file systems at once.
List every 14 instance, its size, extensions (SELECT extname, extversion FROM pg_extension), replicas, backup method and owning team. Record p95 query latency for your top queries, replication lag, checkpoint behavior and nightly job durations. These numbers are your pass or fail line later.
Restore a recent production backup to a staging host, then run pg_upgrade --check --link against an 18 cluster. The docs say --check runs only the checks, can run while the old server is up, and outlines manual adjustments; passing the mode flag adds mode-specific checks. Install extension binaries that match the new server, but do not run CREATE EXTENSION, because pg_upgrade carries the definitions over. Run the application test suite and your monitoring queries against the result.
Upgrade the least critical clusters first: dev, CI, internal tools, then a read-mostly service. Switch backup hosts to pg_dump from 18, since the docs recommend the newer version's dump programs and an old pg_dump refuses to dump a newer server. For logical replication clusters, create the publication on 14 and the subscription on 18 now, copy the schema with pg_dump --schema-only, and let the subscriber catch up while traffic continues.
For pg_upgrade --link: stop the application, stop 14, run the upgrade, upgrade standbys with the documented rsync --archive --delete --hard-links --size-only --no-inc-recursive command (or rebuild them), start 18, and run smoke tests. If the new cluster has not been started yet, the docs describe how to reuse the old one by removing the .old suffix from global/pg_control. Once 18 has started, rollback means restoring from backup. For logical replication: pause writes, confirm the subscriber has caught up, set sequences on 18 to values above the 14 maximums, point the application at 18, and keep 14 running untouched as the fallback.
pg_upgrade on 18 carries most optimizer statistics across, but not everything, so run vacuumdb --all --analyze-in-stages --missing-stats-only and then vacuumdb --all --analyze-only. Reindex full-text search and pg_trgm indexes on ICU clusters. Replace MD5 password hashes before a future release drops them. Delete the old data directory and the 14 packages once your rollback period ends, and remove postgres:14 from every image and CI file.
A 14 upgrade needs a database engineer who has run pg_upgrade and logical replication in production, a backend engineer who owns application compatibility and test coverage, and a DevOps or platform engineer who handles images, CI, infrastructure as code and the cutover runbook. On a large estate, add a second database engineer so one person can rehearse while another prepares the next cluster. If you need help, look at our SQL developers and DevOps engineers, or DevOps consulting for the pipeline side.
Many teams begin with a pod of 3 to 5 senior engineers (a tech lead plus engineers), and a full pod is about seven. Ryz engineers on your team start from $7,000 per month (our published range is $7,000–$15,000 per engineer per month, mostly mid-level to senior), quoted per team.
Because 14 is already unsupported, speed matters here: once you say go, engineers can typically start in under two weeks. Ryz engineers work on your team, reporting to your leads, inside your change process. To get a proposal, scope your team.
Nothing broke on November 12, 2026, and your data did not change. What stopped is upstream patching for 14, so any vulnerability discovered from now on stays open on your servers. Treat it as a known, growing risk: move to 14.25 if you are behind, limit network exposure, and schedule the major upgrade now.
PostgreSQL 18, supported until November 14, 2030, for most teams. Choose 17, supported until November 8, 2029, if an extension, driver or hosting platform you rely on does not support 18 yet. Going to 15 or 16 means repeating the upgrade within one or two years.
Five things drive it: total data volume, how many clusters and standbys you run, any third-party extensions, the write freeze you can tolerate, and how much backup or monitoring code touches removed functions and columns. Link mode itself is fast; the preparation and rehearsal are where the time goes.
The PostgreSQL project's published policy is five years of support followed by a final minor release, and it describes no extended support option for 14. Some managed database providers and commercial PostgreSQL vendors have their own support programs. Check your platform vendor's current documentation and pricing directly.
Budget with simple arithmetic: people, times months, times a monthly rate that begins at $7,000 per engineer and reaches $15,000 for senior staff. Each quote covers a whole team and arrives with a plan and named engineers.
Questions we didn't answer? Email info@ryzlabs.com.
Thanks — your message has been sent. We’ll get back to you soon.
Something went wrong while sending your message. Please try again or email info@ryzlabs.com.