Velocity is an O’Reilly web performance conference focused on scalability, reliability, and infrastructure engineering. The 2025 edition in New York featured a track on identifier design at scale — examining how primary key choice affects write throughput, cache efficiency, and distributed coordination at millions of requests per second. UUID v7’s adoption following RFC 9562 was a central theme across database, CDN, and edge sessions.
Database write throughput at scale
Sessions opened with production benchmarks from teams running 50–200M inserts per day:
The core finding matched existing PostgreSQL research: UUID v4 primary key tables accumulate 40–60% index bloat versus sequential IDs at 100M rows. UUID v7 reduces this to under 5% bloat because new values always append to the rightmost B-tree leaf. At 10,000 inserts/second sustained, the difference in write latency becomes measurable at p95.
PostgreSQL 18’s native uuidv7() eliminates application-layer generation overhead — the database generates the ID server-side, removing a round-trip for frameworks that previously had to generate IDs in application code. See PostgreSQL 18 Ships Native UUID v7.
Edge ID generation patterns
A session on edge infrastructure examined UUID generation in Cloudflare Workers and Vercel Edge Functions. Key points:
crypto.randomUUID()is available globally in all V8-based edge runtimes — zero overhead UUID v4- The uuid npm package runs in edge runtimes without Node.js polyfills
- UUID v7 at the edge as a correlation ID propagated via
x-correlation-idheader gives log aggregators chronological ordering across regions
See UUID in Edge Functions for the implementation patterns.
Cache key design with UUID v7
A CDN architecture session demonstrated time-windowed cache key patterns using UUID v7 timestamps. Because UUID v7’s first 12 hex characters encode Unix milliseconds, cache entries can be grouped by time bucket for bulk invalidation — without storing separate created_at metadata per key.
CDN ETags set to a UUID v7 version_id give monotonically increasing cache validators — a later write always produces a lexicographically larger ETag, making freshness comparisons unambiguous. See UUID v7 Caching Strategies.
RFC 9562 adoption timeline
A standards track session surveyed adoption since RFC 9562’s publication in May 2024:
- PostgreSQL 18 (September 2025):
uuidv7()built-in - Python 3.13 (October 2024):
uuid.uuid7()standard library - .NET 9 (November 2024):
Guid.CreateVersion7() - Hibernate 6.5+:
@UuidGenerator(style = TIME) - Prisma 5.x:
@default(dbgenerated("uuidv7()"))
The consensus: UUID v4 will remain valid but UUID v7 is now the sensible default for any new system using a relational database. See UUID v7 ORM Support in 2026.
Further reading
- UUID v7 Caching Strategies
- UUID in Edge Functions
- PostgreSQL 18 Ships Native UUID v7
- UUID v4 vs v7 — full comparison
- UUID in Databases
External references
Frequently asked questions
What UUID performance topics were covered at Velocity 2025?
Velocity 2025 sessions examined UUID v7 write throughput at high insert rates, comparing B-tree index fragmentation between UUID v4 and UUID v7 at 100M+ row tables. Speakers from large-scale web platforms shared production data showing UUID v7 reducing index page splits by 40–60% versus UUID v4. The RFC 9562 adoption timeline and native support in PostgreSQL 18 were key discussion points.
How does UUID v7 affect CDN and cache key performance?
UUID v7's millisecond timestamp prefix enables time-windowed cache key patterns that UUID v4 cannot support — cache entries can be grouped by time bucket and invalidated together without auxiliary metadata. Sessions at Velocity 2025 demonstrated how sorting cache keys by UUID v7 prefix gives approximate LRU ordering, reducing cache thrash at high request volumes. See UUID v7 Caching Strategies.