Postgres 19 RC 2 Lands as Postgres 14 Goes End of Life
The November update release arrived on November 12 in a single announcement: PostgreSQL 18.7, 17.12, 16.16, 15.20, 14.25, and PostgreSQL 19 RC 2. It closes 52 security vulnerabilities and over 140 bugs, and it carries two dates that matter more than the version numbers: 14.25 is the final release of PostgreSQL 14, and PostgreSQL 19 is scheduled for general availability on November 19, 2026.
Most teams will replace the binaries and move on. The part worth reading is where the update does not finish when the server restarts. GIN indexes, pg_trgm GiST indexes, B-tree deduplication on bpchar columns, incremental backup chains, and NOT NULL constraints on partitioned tables all need a deliberate follow-up step. Skip them and a routine minor release turns into a quiet correctness problem weeks later.
What actually shipped on November 12
Five branches got a maintenance release, and 19 got its second release candidate. The largest security item is CVE-2026-6460, a type-confusion bug after ALTER ... SET DATA TYPE that allows arbitrary code execution at CVSS 8.8 — reachable by any role that can alter a table. Another cluster of holes around plperl, pltcl, pg_dump, pg_restore, and vacuumdb lands in the same 8.8 range, and CVE-2026-120154 is the one that should worry anyone leaning on row-level security: INSERT ... ON CONFLICT could read rows despite RLS policies.
On the non-security side, the bug list is unusually dense for a minor. Snapshot import/export from a standby could corrupt a parallel pg_dump and, more generally, return wrong results. Several GIN index fixes address corruption that makes queries miss rows or return unrelated ones. Partition pruning had multiple wrong-result bugs, including multi-column RANGE tables filtered on leading columns and DEFAULT partitions using MINVALUE/MAXVALUE bounds. Row loss during table rewrites after an aborted subtransaction is fixed. Incremental backups could produce corrupt output, so the project recommends taking a fresh full backup after updating and using it as the new incremental baseline.
Postgres 14 is now end of life
14.25 is the last release PostgreSQL 14 will ever get. The branch shipped in September 2021 and has now left the five-year support window. Five supported majors become four, then five again on November 19 when 19 goes GA — 15, 16, 17, 18, 19. We flagged the countdown to this end-of-life date a week ago; the difference today is that there is no longer a patched 14 to fall back to. If a 14 instance is still taking writes, its next CVE stays open indefinitely.
For Houston shops this is rarely one server. It is usually the reporting replica nobody upgraded, a vendor appliance, or the database behind a settlement or scheduling system that a small team inherited. Those are the instances that carry the risk precisely because they are not in anyone’s upgrade backlog.
The post-update steps most teams skip
The release notes list four conditions that require action after the binaries are in place. Inventory them before the maintenance window, not during it.
1. GIN indexes. The corruption fixes mean an affected GIN index keeps returning wrong answers until it is rebuilt. Find them first:
SELECT n.nspname AS schema,
c.relname AS table_name,
i.relname AS index_name
FROM pg_index x
JOIN pg_class c ON c.oid = x.indrelid
JOIN pg_class i ON i.oid = x.indexrelid
JOIN pg_am am ON am.oid = i.relam
JOIN pg_namespace n ON n.oid = c.relnamespace
WHERE am.amname = 'gin'
AND c.relkind = 'r'
ORDER BY 1, 2, 3;
2. pg_trgm GiST indexes. Any index using the gist_trgm_ops operator class needs a reindex after updating. On a large table, do it concurrently so writes keep flowing:
SELECT i.relname AS index_name, c.relname AS table_name
FROM pg_index x
JOIN pg_class i ON i.oid = x.indexrelid
JOIN pg_class c ON c.oid = x.indrelid
JOIN pg_opclass opc ON opc.oid = x.indclass[0]
WHERE opc.opcname = 'gist_trgm_ops';
REINDEX INDEX CONCURRENTLY index_name_from_above;
3. bpchar B-tree indexes. Deduplication is now disabled for columns declared as bpchar without a length, because index-only scans could return wrong results. Columns declared char(n) are unaffected. If you have such indexes, reindex them.
4. Partitioned tables with NOT NULL NO INHERIT. These constraints are now rejected outright because a partition could hold NULL where the parent promised it could not. Find the existing ones with the query from the release announcement:
SELECT conrelid::regclass, conname
FROM pg_constraint
WHERE contype = 'n' AND connoinherit
AND conrelid IN (SELECT oid FROM pg_class WHERE relkind = 'p');
Drop the ones you do not want; repair the ones you do with ALTER TABLE ... ALTER CONSTRAINT ... INHERIT, which will fail loudly if a partition actually contains a NULL.
One more: if you run extensions, this release changed the interface between the server and compiled extensions, so the extensions have to move with the server. The packaging mechanics behind that constraint are worth reading before you pin packages.
Postgres 19 RC 2 is three days from GA
If nothing blocks it, PostgreSQL 19 goes generally available on November 19. RC 2 is, by design, close to what 19.0 will be, which makes this the last cheap window to test. Two items from the RC 1 period deserve attention if you have been running betas:
- A change allowing
JSON_ARRAY()andJSON_OBJECT()over arrays, composites, and ranges in expression indexes, generated columns, and partition keys was reverted — those objects could go stale. If you created any, drop or redefine them before upgrading, orpg_dump/pg_upgradewill fail to recreate them. - With
pg_upgrade, indexes using full-text search,pg_trgm, orfuzzystrmatchmay need to be rebuilt after the upgrade. Budget the reindex time into the cutover plan rather than discovering it after.
Other RC 2 fixes cover wrong results in GROUPING SETS, an infinite loop in first_value()/nth_value() with IGNORE NULLS and certain EXCLUDE frames, standby promotion hanging on a stalled client, and hash tables sized far beyond work_mem. The timezone data moved to tzdata 2026f as well, with Northwest Territories and Manitoba going to year-round fixed offsets and Cuba staying on daylight time all winter.
What to do this week
- Patch every supported cluster to 18.7, 17.12, 16.16, or 15.20 this week — the CVSS 8.8 items are all authenticated-user reachable, and one is RLS-relevant.
- Run the four inventory queries above before the patch, so the reindex work is scoped and the concurrent indexes are scheduled instead of improvised.
- Take a new full backup after patching if you use incremental backups, and make it the new baseline.
- Put 14.25 on a hard date. Choose the target major — 17 and 18 are both comfortably mature — and test the upgrade path with real data volumes.
- Stand up a 19 RC 2 instance for your test suite now, so November 19 is a decision rather than a scramble.
If the Postgres 14 exit keeps sliding because nobody owns the upgrade path, that is the sort of thing worth handing to people who do it every quarter — get a fixed-date plan for retiring Postgres 14 and moving your stack to 18 or 19.
