February 16, 2026
Cache me outside
How DSQL makes sure sequences scale
DSQL adds old-school numbers, sparking a UUID vs sequence brawl
TLDR: Aurora DSQL now supports simple numbered IDs (“sequences”) designed to scale by caching number blocks per worker. The crowd’s split: fans love readable IDs, skeptics fret about gaps and ask for real-world tests, while newcomers admit they’re just learning what DSQL even is—and why it matters.
Amazon’s Aurora DSQL just flipped on “sequences”—the classic number dispenser for IDs—and the comment section instantly turned into Team UUID vs Team Big Number. A bunch of readers admitted they’d never even heard of DSQL, prompting SoftTalker’s PSA that it’s AWS’s serverless distributed SQL released in late 2024. From there, it was open season: some cheered the simple, human-friendly IDs; others rolled their eyes and asked why we’re still debating this in 2026. The blog says DSQL hands out number chunks (via “cache”) to each worker so things scale, which had commenters joking it’s basically the deli counter ticket machine, just cloudier.
Then came the drama: gaps. If a worker crashes, cached numbers get tossed, meaning you might skip numbers. One camp declared gaps are harmless (IDs aren’t receipts!), another imagined accountants crying and dashboards glitching. Meme brigade weighed in with “hot key meltdown” jokes and “nextval, next drama” punchlines, while the cautious crowd asked for real-world benchmarks and failure-mode stories before betting their order IDs on this. Net-net: DSQL’s sequence support makes big apps feel less chaotic, but the community is split between loving the simplicity and warning that “cloud deli tickets” aren’t for every use case. Context for newbies? See Aurora and Postgres sequences.
Key Points
- •DSQL has launched support for SQL sequences and provides examples for creating sequences and identity columns using CACHE.
- •The authors still recommend UUIDs for extremely large-scale use cases, though numeric sequences are now supported.
- •In Postgres, sequences are stored in tables, updated via logs for crash recovery, and can be cached by backends to reduce I/O.
- •In DSQL’s distributed architecture, sequences become a hot key; nextval() operations involve storage reads, increments, and journal writes.
- •DSQL scales sequence allocation by using CACHE to pre-reserve chunks per backend, improving throughput but potentially causing gaps if a backend crashes.