PGConf US 2026 in Chicago gathered PostgreSQL core contributors, DBAs, and application developers for four days of talks and workshops. With PostgreSQL 18 releasing uuidv7() as a built-in, UUID v7 adoption was a central theme across the database design, performance, and migration tracks. See PostgreSQL 18 Native UUID v7.
Session: PostgreSQL 18 UUID v7 first-week deployment
A fintech startup shared their experience deploying PostgreSQL 18 on day one to get native uuidv7():
-- PG 18: zero-dependency UUID v7
CREATE TABLE transactions (
id uuid PRIMARY KEY DEFAULT uuidv7(),
account_id uuid NOT NULL,
amount numeric NOT NULL,
status text NOT NULL DEFAULT 'pending'
);
-- Extract timestamp directly — no extra column needed
SELECT
id,
uuid_extract_timestamp(id) AS created_at,
amount,
status
FROM transactions
ORDER BY id
LIMIT 100;
-- Time-range query using UUID v7 prefix
SELECT * FROM transactions
WHERE id >= uuidv7_from_timestamp(now() - interval '1 hour')
ORDER BY id;
The speaker noted that uuid_extract_timestamp(id) in the SELECT list replaces a created_at column for display purposes while id alone handles all range and pagination queries.
Session: 500M-row migration war story
A payments platform presented their migration from UUID v4 to UUID v7 on their largest table:
-- Step 1 (Week 1): Add column
ALTER TABLE payments
ADD COLUMN id_v7 uuid DEFAULT uuidv7();
-- Step 2 (Weeks 2–5): Batched backfill
-- ~10,000 rows per batch, 30ms pause between batches
DO $$
DECLARE
last_id uuid;
BEGIN
LOOP
UPDATE payments SET id_v7 = uuidv7()
WHERE id IN (
SELECT id FROM payments
WHERE id_v7 IS NULL
ORDER BY id
LIMIT 10000
);
EXIT WHEN NOT FOUND;
PERFORM pg_sleep(0.03);
END LOOP;
END $$;
-- Step 3 (Week 6): Non-blocking index
CREATE INDEX CONCURRENTLY idx_payments_id_v7 ON payments(id_v7);
-- Step 4 (Week 7): Update 8 child tables' FK columns
-- Step 5 (Week 8): PK swap — 410ms
BEGIN;
ALTER TABLE payments DROP CONSTRAINT payments_pkey;
ALTER TABLE payments ADD PRIMARY KEY USING INDEX idx_payments_id_v7;
COMMIT;
Result: Insert throughput +3.4×, pagination query latency −67%, zero downtime. See Zero-Downtime UUID Migration.
Session: uuid_extract_timestamp() in practice
-- Build time-bucketed analytics directly from UUID v7 primary keys
SELECT
date_trunc('hour', uuid_extract_timestamp(id)) AS hour,
count(*) AS order_count,
sum(total) AS revenue
FROM orders
GROUP BY 1
ORDER BY 1 DESC;
-- No created_at index needed — UUID v7 id carries the timestamp
The session demonstrated that uuid_extract_timestamp() opens up new query patterns without schema changes — the timestamp is always available from the primary key.
Session: Multi-tenant UUID v7 design workshop
-- Pattern: composite PK (tenant_id, id) where id is UUID v7
-- Rows are co-located per tenant, sorted by creation time within tenant
CREATE TABLE events (
tenant_id uuid NOT NULL,
id uuid NOT NULL DEFAULT uuidv7(),
payload jsonb,
PRIMARY KEY (tenant_id, id)
);
-- Efficient tenant time-range scan — uses composite index
SELECT id, uuid_extract_timestamp(id) AS ts, payload
FROM events
WHERE tenant_id = $1
AND id >= $2 -- UUID v7 lower bound
ORDER BY id
LIMIT 50;
See UUID v7 in Multi-Tenant SaaS.
Further reading
- PostgreSQL 18 Native UUID v7
- Zero-Downtime UUID Migration
- UUID v7 in Multi-Tenant SaaS
- UUID v7 Performance Benchmarks
- PGConf.EU 2025 — UUID v7 and PostgreSQL 18
External references
Frequently asked questions
What new UUID v7 capabilities in PostgreSQL 18 were highlighted at PGConf US 2026?
Speakers highlighted three key PG 18 features: uuidv7() as a built-in default, uuid_extract_timestamp() for extracting the embedded millisecond from any UUID v7 value, and uuid_extract_version() for version detection. Sessions showed that these eliminate the pg_uuidv7 extension for new deployments. See PostgreSQL 18 Native UUID v7.
What migration timelines were shared at PGConf US 2026?
Practitioners shared migrations of tables ranging from 50M to 500M rows. The consistent recommendation was the dual-column approach: add id_v7 uuid DEFAULT uuidv7(), backfill in batches of 5,000–10,000 rows with a 20–50ms pause, CREATE INDEX CONCURRENTLY, then a sub-500ms PK swap. The biggest challenge reported was updating foreign key columns in child tables before the PK swap. See Zero-Downtime UUID Migration.