Launching something? List it on Upfynd for free.

Submit your product
Software Guides

PostgreSQL 19 upgrade: JIT is off, RADIUS is gone, and what else to check before you move

PostgreSQL 19 shipped on November 19, 2026. Its new features matter less on upgrade day than its changed defaults: JIT off, RADIUS removed, md5 warnings, io_workers replaced. Here is what to audit, what blocks pg_upgrade, and which hosts offer 19.

Nihar Rout · · Updated · 10 min read · 70 views

PostgreSQL 19 was released on November 19, 2026, and the changes most likely to hurt a postgresql 19 upgrade are not the headline features. They are changed defaults: JIT is off, RADIUS is gone, md5 logins warn and io_workers no longer exists. Below: each change, what blocks pg_upgrade, and which managed providers offer version 19 as of November 20, 2026.

Everything below comes from the official PostgreSQL 19 release notes, mainly the "Migration to Version 19" section, and the version 19 reference docs. Where we say a provider does or does not offer version 19, that is what its own docs said when we checked on November 20, 2026.

PostgreSQL 19 breaking changes that will actually bite

The migration section lists 22 incompatibilities. Most touch extension authors or rare setups. These are the ones that change behavior for an ordinary application database:

Change in PostgreSQL 19What happensWhat to check before you move
JIT disabled by defaultLarge analytical queries lose JIT unless you turn it onQueries that ran faster with JIT
RADIUS support removedpg_hba.conf entries using radius stop workingAny radius line in pg_hba.conf
Warning after md5 loginClients see a WARNING on every md5 authenticationRoles still stored with md5 hashes
io_workers removedA postgresql.conf that sets it needs editingReplace with io_min_workers and io_max_workers
max_locks_per_transaction default 64 to 128Lock sizing changed, so old custom values buy half the capacityAny value you raised by hand
standard_conforming_strings always onOld dumps made with it off will not loadHow your last dump was taken
Parallel vacuum shares one cost limitVACUUM with parallel workers and cost delay may run slowerMaintenance windows that rely on parallel vacuum

The postgresql 19 upgrade checklist: defaults to audit first

Work through these on a staging copy of production before you schedule the real move. Each item names the setting so you can grep your configuration for it.

JIT is disabled by default in PostgreSQL 19

The jit setting now defaults to off. The release notes give the reason plainly: JIT used to switch on based on optimizer cost estimates, and "this costing has been determined to be unreliable." For most transactional workloads that is good news, because queries the planner misjudged no longer pay JIT compilation time by accident.

The catch is analytics. The release notes say sites "doing many large analytical queries" must enable JIT manually. If you run reporting on the same cluster, capture timings for your heaviest queries on the old version, then compare on 19 with jit off and on. The jit_above_cost threshold defaults to 100000 and only matters once jit is on.

RADIUS authentication is removed in PostgreSQL 19

RADIUS support is gone entirely. The release notes explain that PostgreSQL only supported RADIUS over UDP, which they call "unfixably insecure." If any line in pg_hba.conf uses the radius method, those users will not be able to log in after the upgrade. Move them to another method, such as LDAP, certificate or SCRAM password authentication, before upgrading, not after.

The PostgreSQL 19 md5 password warning

md5 passwords were deprecated in PostgreSQL 18, and version 19 now issues a warning after every successful md5 login. The new md5_password_warnings setting defaults to on, and also warns when CREATE ROLE or ALTER ROLE sets an md5 password. You can switch it off, but the password authentication docs say md5 support "will be removed in a future release."

The documented path to SCRAM-SHA-256 is short. Confirm your client libraries support SCRAM, set password_encryption to scram-sha-256, have every role set a new password, then change md5 to scram-sha-256 in pg_hba.conf. Until a role's password is re-set, its stored hash is still md5, so the warning keeps firing for that role.

io_workers is replaced by io_min_workers and io_max_workers

The io_workers setting has been removed. With io_method set to worker, which is the default, PostgreSQL 19 scales its I/O worker pool on its own. The resource configuration docs give the new defaults: io_min_workers is 2, io_max_workers is 8, idle workers exit after 1 minute, and a new worker can start at most every 100ms. If you tuned io_workers on version 18, delete that line and decide whether your old value belongs in io_min_workers or io_max_workers.

max_locks_per_transaction now buys half as much

The default rose from 64 to 128, and that is not a free doubling. The release notes say lock size allocation changed, so "effectively settings must now be doubled to match their capacity in previous releases." If you raised it by hand, double your custom value too.

New values use lz4 TOAST compression

When PostgreSQL is built with LZ4, default_toast_compression now defaults to lz4 instead of pglz. The release notes file this under performance, not under migration, so it is easy to miss. If anything in your stack expects pglz, set the COMPRESSION option on those columns or set the default back.

standard_conforming_strings is forced on

The server now always runs with standard_conforming_strings on, and escape_string_warning is gone. The release notes warn that dumps made by a pre-19 pg_dump or pg_dumpall with standard_conforming_strings off will not load into version 19. If you migrate by dump and restore, take the dump with the version 19 client tools.

pg_upgrade to PostgreSQL 19: what blocks it and what to redo

The release notes give three supported routes for a major upgrade: dump and restore with pg_dumpall, pg_upgrade, or logical replication. pg_upgrade accepts clusters from 9.2 onward, and its --check option runs every pre-flight check without changing data, even while the old server is still running. Run it first. Version 19 adds two new reasons for it to refuse:

  • Database, role or tablespace names that contain carriage returns or line feeds.
  • Indexes using the btree_gist opclasses for inet or cidr, which the release notes say can exclude rows that should be returned. The default opclass for those types is now GiST.

Separately, the MULE_INTERNAL encoding was removed, and the release notes say databases using it "will need to be dumped and restored with a different encoding."

Two more steps come after the upgrade. The text search and pg_trgm modules now use the database's locale provider for character classification instead of always using libc, so reindex affected indexes and recompute stored tsvector values. Indexes built on fuzzystrmatch functions need a reindex too.

pg_upgrade transfers most optimizer statistics, but not extension statistics or the cumulative statistics that trigger autovacuum. The docs tell you to run vacuumdb --all --analyze-in-stages --missing-stats-only, then vacuumdb --all --analyze-only. Budget time for that in your maintenance window.

Logical replication of sequences helps the next upgrade more than this one

Logical replication is the low-downtime route: run version 19 as a subscriber, let it catch up, then switch over. Sequences have always been the weak point. The version 18 docs say "sequence data is not replicated," so a subscriber's sequences sit at their start values until you copy them across before switchover.

PostgreSQL 19 adds CREATE PUBLICATION ... FOR ALL SEQUENCES and an ALTER SUBSCRIPTION ... REFRESH SEQUENCES clause. But publications are defined on the publisher, and the version 19 docs state that REFRESH SEQUENCES "requires the publisher to be running PostgreSQL 19 or later." When you replicate from an 18 cluster into 19, the old cluster is the publisher, so plan to copy sequence values by hand as before. The new feature pays off on the move after this one. Even on 19, incremental sequence changes are not streamed; a subscriber keeps the last value it synchronized.

One change makes logical replication easier to set up: with wal_level set to replica, the server now raises the effective level to logical when the first logical slot is created, so you no longer need a restart for that. The new effective_wal_level setting shows which level is in force.

What replaces VACUUM FULL in PostgreSQL 19

The new REPACK command replaces VACUUM FULL and CLUSTER, which the release notes say "have been retained for compatibility," so existing maintenance scripts keep working. The bigger change is REPACK (CONCURRENTLY), which does most of its work under a SHARE UPDATE EXCLUSIVE lock and takes ACCESS EXCLUSIVE only to swap files at the end. Changes made in the meantime are captured through logical decoding.

Read the limits before you replace an extension-based workflow with it. CONCURRENTLY does not work on unlogged or partitioned tables, inside a transaction block, or on tables with no primary key or index-based replica identity. It needs a free slot under max_repack_replication_slots and disk space for a full copy of the table and its indexes. The docs also warn it is not MVCC-safe: transactions using snapshots taken before the repack began may see the table as empty afterwards.

PostgreSQL 19 new features worth planning for

These do not affect upgrade day, but they are reasons to schedule it. All are described in the official release announcement:

  • WAIT FOR LSN lets a session on a standby wait until WAL up to a given LSN has been replayed, which gives read-your-writes consistency when reads go to replicas. It accepts a TIMEOUT and errors if the standby does not catch up in time, unless you add NO_THROW.
  • pg_plan_advice and pg_stash_advice let you stabilize plan choices, and stashed advice is applied automatically by query identifier, without changing application code.
  • INSERT ... ON CONFLICT DO SELECT returns the existing conflicting row, which simplifies get-or-create code.
  • COPY can write JSON, and lead(), lag(), first_value(), last_value() and nth_value() accept IGNORE NULLS.
  • Autovacuum can process indexes with parallel workers, controlled by autovacuum_max_parallel_workers.
  • Server-side SNI lets one server present different TLS certificates per hostname, configured in pg_hosts.conf.

The announcement also says faster foreign key checks made bulk inserts "nearly 2x faster" in testing. Treat that as a reason to benchmark, not as a number to plan capacity around.

Does my managed Postgres support version 19 yet?

For most teams, the provider decides the upgrade date. Here is what each provider's own documentation said on November 20, 2026, the day after release:

ProviderPostgreSQL 19 status in its docsNewest version otherwise listed
NeonSupported, first available November 19, 202618 is still the default for new projects
Amazon RDSPreview environment only, up to 19 RC118.6
Azure Database for PostgreSQLNot listed (page dated August 26, 2026)18.6
Google Cloud SQLNot mentioned in release notes through November 9, 202618.6
SupabaseNot mentioned in its upgrade guideGuide covers upgrades to 17

Neon is the exception, with caveats on its own page. Version 19 projects do not support Neon's Data API yet, some extensions are not available, and pg_stash_advice stashes are not saved across compute restarts. Neon projects cannot be upgraded in place, so moving means creating a new project on 19 and migrating your data into it.

Skip the upgrade for now if you depend on an extension your provider has not certified for 19, or if your provider only offers it in preview. A preview environment is the right place to run the checklist above, not to hold production data.

Your next step

Run pg_upgrade --check against a copy of production on version 19 binaries, grep postgresql.conf and pg_hba.conf for jit, radius, md5, io_workers and max_locks_per_transaction, and time your heaviest analytical queries with JIT on and off. If you pool connections, our note on PgBouncer transaction mode covers what still breaks there.

If the upgrade is also a reason to reconsider hosting, Upfynd lists 830 products in Databases & Infrastructure as of November 20, 2026. Check any host's own version page before you commit, because availability of 19 will change week by week. Making a Postgres tool yourself? You can list it for free.

Frequently asked questions

When was PostgreSQL 19 released?

PostgreSQL 19 was released by the PostgreSQL Global Development Group on November 19, 2026. The release notes for version 19 carry the same date. Managed providers add new major versions on their own schedules, so a release date upstream does not mean your hosted database can be upgraded yet.

Is JIT disabled by default in PostgreSQL 19?

Yes. In PostgreSQL 19 the jit setting defaults to off. The release notes say the optimizer cost estimates that used to switch JIT on proved unreliable. Sites that run many large analytical queries need to enable JIT manually, so compare timings for those queries with jit on and off before upgrading.

What breaks when upgrading to PostgreSQL 19?

The PostgreSQL 19 changes most likely to break an upgrade are the removal of RADIUS authentication, the removal of the io_workers setting, standard_conforming_strings forced on, and a doubled max_locks_per_transaction default with changed lock sizing. pg_upgrade also refuses clusters with btree_gist inet or cidr indexes, or names containing line breaks.

What replaces VACUUM FULL in PostgreSQL 19?

PostgreSQL 19 adds the REPACK command, which replaces VACUUM FULL and CLUSTER. Both older commands are kept for compatibility. REPACK with the CONCURRENTLY option rebuilds a table while reads and writes continue, taking an exclusive lock only briefly at the end, but it has limits such as needing a primary key or replica identity.

Does PostgreSQL 19 still support md5 passwords?

PostgreSQL 19 still accepts md5 passwords but warns after every successful md5 login, controlled by the md5_password_warnings setting. The PostgreSQL documentation says md5 support will be removed in a future release. The documented fix is to set password_encryption to scram-sha-256, have roles set new passwords, and switch pg_hba.conf to scram-sha-256.

Does my managed Postgres support PostgreSQL 19 yet?

As of November 20, 2026, Neon's documentation lists PostgreSQL 19 as supported from release day. Amazon RDS offers it only in its preview environment, while the Azure, Google Cloud SQL and Supabase pages we checked did not mention version 19. Check your provider's own version page, since this changes quickly.

Products in this article

About the author

Nihar Rout profile photo

Nihar Rout

CEO & Co-founder

Connect on LinkedIn