Amazon Aurora DSQL

Original Article ↗ Hacker News Discussion ↗

Marketing, “Serverless”, and Operational Reality

  • Slogan claims like “virtually unlimited scale,” “highest availability,” and “zero infrastructure management” are widely viewed as exaggerated or misleading.
  • Critics note you still need solid AWS expertise (IAM, networking, cost controls) and there is no hard spending cap, so “serverless” does not mean zero ops risk.
  • Others argue “zero” is marketing shorthand for “far less than managing your own clustered DB,” not literally zero work.

Scalability, Limits, and Consistency

  • Quotas are seen as at odds with “limitless”: 100 GB per cluster (configurable), 10 MiB of data per write transaction, max 10k modified rows per transaction, 5-minute transaction time, 128 MiB temporary storage per query.
  • Isolation level is equivalent to PostgreSQL REPEATABLE READ; SERIALIZABLE is not supported.
  • Blog material claims strong consistency for cross-region transactions and constant latency vs statement count, which some find technically impressive.
  • Transaction limits are considered deal-breakers for “everything in the DB” architectures or heavy analytics.

PostgreSQL “Compatibility” and Missing Features

  • Wire/SQL compatibility is heavily debated.
  • Reported missing or limited features include: foreign keys, views, triggers, sequences, temporary tables, multiple databases per cluster, many types/UDFs/extensions, nested transactions, NOTIFY, and json/jsonb.
  • Many say this makes it unusable as a drop-in for existing Postgres apps; “compatibility” is seen as stretched marketing.
  • Some argue a strict definition of “PostgreSQL-compatible” is needed to curb such claims.

Positioning vs Other Databases

  • Frequently compared to:
    • Google Spanner (multi-region, strong consistency) but here “serverless” instead of pre-provisioned nodes.
    • CockroachDB and other distributed SQL systems (notably they offer SERIALIZABLE and FKs).
    • DynamoDB / “Dynamo with SQL front-end” and even Cassandra/CQL from a user-experience standpoint.
  • Some view it as closer to a distributed KV store with SQL semantics than a full relational system.

Use Cases and Audience

  • Critics struggle to see who needs both “virtually unlimited” scale and such limited SQL features.
  • Suggested fits: simple CRUD apps that want auto-scaling, multi-region financial/gaming workloads that are schema-simple but high-throughput, and teams burned by DynamoDB’s denormalization.
  • Lack of multiple DBs per cluster concerns multi-tenant and microservice designs, though some argue per-cluster isolation and usage-based pricing make that less relevant.

Pricing, Docs, and AWS Product Sprawl

  • Pricing is absent; several say they won’t seriously evaluate it without costs. Preview is noted as free.
  • Launch-time docs were initially missing or incomplete, reinforcing a “rushed”/cutting-edge feel.
  • Many complain AWS’s RDS/Aurora ecosystem (RDS, Aurora, Serverless v1/v2, Limitless, Global, DSQL) is confusing and poorly messaged, unlike the clearer S3 story.