Ryz Labs/Migrations/PostgreSQL 15 end of life
Migrations

PostgreSQL 15 End of Life: A Calm 2027 Upgrade Plan

You have time, so use it well: one jump from 15 to 18, a rehearsed cutover, and a cleaner path for every PostgreSQL upgrade after this one.

PostgreSQL 15 reaches end of life on November 11, 2027, per the support calendar the PostgreSQL project publishes. From then on the community stops shipping patches for it. Because the deadline is still ahead, you can plan instead of react: skip 16 and 17, land on 18 (in support through November 14, 2030), and pick a cutover method that fits your downtime budget. A planned, rehearsed upgrade during 2027 costs far less attention than an emergency one after the date.

Key dates

VersionReleasedEnd of lifeStatus on November 21, 2026
PostgreSQL 15October 13, 2022November 11, 2027In support, latest patch 15.20
PostgreSQL 16September 14, 2023November 9, 2028In support, but only a year beyond 15
PostgreSQL 17September 26, 2024November 8, 2029In support
PostgreSQL 18September 25, 2025November 14, 2030In support, our suggested landing point
PostgreSQL 19November 19, 2026November 19, 2031In support, brand new

Source: the PostgreSQL project's own release calendar, checked November 21, 2026. Every major line is maintained for a five-year span from its debut. Version 19 shipped two days before this page was updated; most teams will want 18, which has a year of minor releases behind it, and can reassess 19 later in 2027 once their extensions and hosting platform support it.

What end of life means for your systems

Until November 11, 2027, nothing changes. The project keeps publishing 15.x minor releases on its regular cadence (at least every three months), and you should keep applying them: the project's advice is to stay on the newest patch level of whatever major you use. After the final 15 minor ships, the line becomes unsupported, and new security issues will not be patched for it.

Why care a year early?

Find your exposure

A calm upgrade starts with a complete list. Collect it now, while you still have the luxury of fixing the stragglers one at a time.

A shortcut for the repo search: drop a docker-compose.yml or Dockerfile into the EOL deadline radar (free) and it lists each version it detects next to the date that version loses support.

What changes in the upgrade

Skipping versions is allowed. The official upgrading chapter says you can go from one major to another without the intervening ones, provided you study the notes for each skipped release, especially the part titled "Migration". For a 15 cluster that means the notes for 16, 17 and 18. These items deserve a ticket each.

Role administration tightened (16)

Version 16's migration notes narrow what CREATEROLE allows. To alter a different role or grant membership in it, you now need ADMIN OPTION on it, and changing attributes such as CREATEDB, REPLICATION or BYPASSRLS requires holding them yourself. Inheritance is now set per membership at GRANT time. If an onboarding script, secrets manager or tenant service creates and grants roles, run it against 18 early.

High-availability settings removed (16)

promote_trigger_file is gone; failover tooling must call pg_ctl promote or pg_promote() instead. vacuum_defer_cleanup_age was removed, and force_parallel_mode is now debug_parallel_query. A configuration file carried over from 15 with these settings needs editing before the new server will start cleanly. REINDEX DATABASE and reindexdb also stopped touching system catalogs, so maintenance jobs that relied on that need REINDEX SYSTEM.

search_path is locked down during maintenance (17)

Version 17's notes say VACUUM, REINDEX, CREATE INDEX, ANALYZE, CLUSTER and materialized view refreshes now pin search_path to a safe value. If an expression index or materialized view calls a function that looks up objects outside the default schemas, that function must declare its search_path. Find these before the upgrade, because the failure shows up during routine maintenance, not at connect time.

Observability views reshaped (17)

Two pg_stat_bgwriter columns moved to pg_stat_io, pg_stat_statements renamed its block timing columns to shared_blk_read_time and shared_blk_write_time, pg_stat_progress_vacuum changed column names, and old_snapshot_threshold no longer exists. Export your Grafana or Datadog queries now and fix them against an 18 test server.

Logical replication survives future upgrades (17)

Starting with 17, pg_upgrade keeps logical replication slots on publishers and subscription state on subscribers. The logical replication upgrade page says this only applies when the old cluster is 17.0 or later; on older clusters, slots are silently ignored. So the upgrade away from 15 cannot carry your slots, but landing on 18 means the next one can. That is a strong reason to skip 16.

New cluster defaults and password warnings (18)

In version 18, initdb turns checksums on unless told otherwise, while pg_upgrade insists that both clusters agree on that setting; pass --no-data-checksums when initializing the 18 cluster if 15 was created without them. MD5 password support is deprecated and planned for removal in a later major, and the server now warns whenever one is set. Use the year you have to move clients off MD5.

Application-visible behavior (18)

RESTRICT foreign key violations now raise SQLSTATE 23001 rather than 23503, AFTER triggers run as the role that queued the event, unlogged partitioned tables are rejected, and plain VACUUM or ANALYZE on a parent now reaches its inheritance children too (add ONLY for the old scope). Error-handling code that switches on SQLSTATE and ORMs that map error codes are the usual places this shows up.

The upgrade plan

With a year of runway, you can choose the cutover method on merit rather than speed, and you can rehearse it more than once.

Prep work you can do on 15 today

Much of the risk can be retired before anyone touches an 18 binary. Each item below is safe on a running 15 cluster and makes the eventual cutover shorter.

Why 18 and not 19

Version 19 is in support through November 19, 2031, a year longer than 18. It is also two days old as of this update. Extensions, connection poolers, backup tools and managed platforms each need to publish support for a new major, and your rehearsals depend on all of them. Plan on 18, keep a short note in the plan to re-check 19 midway through 2027, and switch only if every dependency on your list has caught up.

Inventory and baseline

Build the list from the exposure section, then capture what normal looks like: p95 and p99 latency for your heaviest queries, replication lag, autovacuum behavior, connection counts and backup durations. Note every installed extension and its version, and confirm each one publishes a build for 18.

Compatibility testing

Stand up 18 next to 15; the upgrade guide recommends concurrent installations so client applications can be tested before switching. Restore a production snapshot, run pg_upgrade --check with the transfer mode you intend to use, and fix every item it reports. Point a copy of the application and its test suite at the result. Switch CI and developer environments to 18 here, so new code is written against the target.

Staged rollout

Order clusters by blast radius: internal and analytics databases first, customer-facing ones last. For each, pick a method:

Cutover and rollback

Write the rollback before the cutover. With --clone or logical replication, the 15 cluster stays intact, so rollback is a connection-string change, provided you have not taken writes on 18 that you cannot replay. With --link, rollback after start means a restore from backup. In a logical switchover, remember that sequence changes do not replicate: set each 18 sequence above the 15 value before reopening writes.

Cleanup

Generate statistics with vacuumdb --all --analyze-in-stages --missing-stats-only, then vacuumdb --all --analyze-only, as the pg_upgrade docs describe. Clusters on ICU (or the builtin provider) should get fresh text-search and trigram indexes. Retire the 15 packages, images and parameter groups, and add the 18 end-of-life date to your calendar so the next upgrade is just as planned.

PhaseGoalKey techniqueExit criteria
Inventory and baselineA complete, owned list of 15 databasesRepo grep, console walk, radar report, metric snapshotEvery cluster has an owner, extension list and baseline
Compatibility testingFind breaking changes from 16, 17 and 18Side-by-side 18 install, pg_upgrade --check, CI on 18No open check items, test suite passes on 18
Staged rolloutUpgrade by blast radius--clone or --link per cluster, logical replication for critical onesEvery low-risk database serving from 18 without incident
Cutover and rollbackMove critical clusters with a fallbackRehearsed switchover, sequence reset, written rollbackCritical databases live on 18; 15 kept until the rollback window closes
CleanupFinish and prepare for the next oneStatistics rebuild, index rebuilds, delete 15 artifactsNo 15 references left; next EOL date tracked

Risk matrix

RiskLikelihoodImpactMitigationEarly warning
Plan slips past November 11, 2027MediumHighSet an internal deadline well before the official one, upgrade low-risk clusters firstInventory still incomplete when 2027 begins
wal_level restart not scheduled in timeMediumMediumChange wal_level to logical during a routine maintenance restartLogical replication chosen but publisher still at replica
Role scripts fail under 16 CREATEROLE rulesMediumMediumRun provisioning jobs against 18 in staging, grant ADMIN OPTION where neededPermission errors from onboarding automation in test
Expression indexes break under safe search_pathLowHighAdd SET search_path to affected functions before upgradingREINDEX or ANALYZE errors on the 18 test server
Failover tooling still uses promote_trigger_fileMediumHighSwitch to pg_ctl promote or pg_promote() and test a failoverA failover drill on the 18 test server does not promote
Provider deprecates 15 earlier than the projectLowMediumRead the provider's lifecycle page now and follow its noticesDeprecation email or console banner from the provider
Error handling relies on SQLSTATE 23503 for RESTRICTLowMediumSearch code for the old code and handle 23001 tooUnhandled exceptions in integration tests on 18

Typical team shape

For a planned 15 upgrade, the core is one person who knows PostgreSQL internals and owns the method choice, rehearsals and runbook; an application developer to fix role scripts, search_path functions and error handling; and a platform or DevOps engineer to move images, CI and infrastructure code to 18. Because the work can run alongside the roadmap, you can add people for the critical cutovers and scale back once they are done. Ryz has database-focused engineers and platform people for exactly this, and cloud migration help if hosting changes too.

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. Ryz engineers work on your team, reporting to your leads. When you want dates on the plan, scope your team and we will propose people.

Related

FAQ

Is PostgreSQL 15 still safe to run after end of life?

Today, yes: 15 is maintained until November 11, 2027, so keep installing each 15.x patch. Past that date, newly found flaws get no upstream patch. Aim to have production on 18 well before the cutoff.

Which version should we upgrade to?

Most teams should target 18; the project maintains it through November 14, 2030, and you can skip 16 and 17 entirely. Landing on 17 or later also means future pg_upgrade runs can keep logical replication slots. Version 19 is newly released; consider it once your extensions and platform support it.

How long does the upgrade take?

That depends on your cluster count, their size, which extensions they use, how much write downtime each can tolerate, whether you choose pg_upgrade or logical replication, and how much of your role, backup and monitoring tooling touches changed features.

Can we pay for extended support?

Not from the community itself, which publishes a five-year window and nothing beyond it. Hosting companies and commercial distributions sometimes sell longer coverage. Ask your platform vendor, and read its official terms for 15.

What does it cost to bring in engineers?

Our rates run from $7,000 to $15,000 a month for each engineer, rising with seniority. A budget is headcount multiplied by months multiplied by rate; we quote each team with named people before work begins.

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.

Explore Ryz Labs

Software development servicesAI development companyIndustriesStaff augmentationAI pod teamsNearshore software developmentHire engineers by roleEngineer ratesGuidesBuyer guidesCase studiesHow we vet engineers

Ryz Labs

Senior engineers on your team. AI pod teams that ship.

Tell us what you're building. You get a scoped plan, a price and the names of the people who would do the work.

  • Only the top 1% of tens of thousands interviewed make it
  • On US business hours, including New York hours
  • Trusted by Fortune 500 engineering teams

Tell us who you need

A Ryz partner replies with a scoped team plan.

Start a conversation →

Prefer to talk? Book a 15-min call →