--- title: "Aurora PostgreSQL 18 lands, Prisma Next hits 50% faster throughput" url: https://daily.dev/posts/aurora-postgresql-18-lands-prisma-next-hits-50-faster-throughput-bcpi2vef6 source_url: https://daily.dev/posts/aurora-postgresql-18-lands-prisma-next-hits-50-faster-throughput-bcpi2vef6 type: freeform source: "Databases Digest" published: 2026-06-12T04:17:56.776Z updated: 2026-06-12T04:18:19.409Z tags: ["aws", "backend", "postgresql", "prisma"] reading_time: 4 upvotes: 0 comments: 0 language: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Aurora PostgreSQL 18 lands, Prisma Next hits 50% faster throughput **[Databases Digest](https://daily.dev/sources/databases_digest)** · 4 min read · 0 upvotes · 0 comments ## Summary A curated digest of database news: Amazon Aurora now supports PostgreSQL 18.3 with B-tree skip scans, statistics retention across upgrades, and parallel logical replication. Prisma Next benchmarks show ~50% higher throughput and p95 latency dropping from 41.8ms to 4.4ms, with a 9x smaller bundle. A comparative analysis covers query planner design across PostgreSQL, Oracle, DuckDB, Snowflake, MySQL, and Db2, explaining why Postgres ships no query hints. Xata cut Postgres branch provisioning from 20+ seconds to 1-2 seconds using ZFS copy-on-write over NVMe-oF, making 1,000 short-lived branches cost ~$1. Also covered: PostgreSQL statistics tuning, Databricks Spatial SQL GA, Datadog+ClickHouse federated log querying, lakeFS agentic AI data isolation, and npm v12 breaking changes. ## Content **TLDR:** Amazon Aurora now supports PostgreSQL 18.3, bringing B-tree skip scans, statistics retention across major upgrades, and parallel logical replication streaming. Prisma Next benchmarks show ~50% higher peak throughput and p95 latency dropping from 41.8ms to 4.4ms at load, with a 9x smaller bundle. A deep comparative look at query planners across six databases — PostgreSQL, Oracle, DuckDB, Snowflake, MySQL, Db2 — is worth reading if you've ever wondered why Postgres ships zero query hints. Xata cut Postgres branch provisioning from 20+ seconds to 1-2 seconds using ZFS copy-on-write over NVMe-oF, making 1,000 short-lived branches cost roughly $1. --- ## Amazon Aurora PostgreSQL 18.3 now available with skip scans and parallel logical replication Aurora PostgreSQL-Compatible now supports major version 18, starting at 18.3. The headline features for query performance are B-tree skip scans (useful for multi-column indexes where the leading column has low cardinality) and retention of optimizer statistics across major version upgrades — no more cold planner after an upgrade. Logical replication gets parallel streaming for large transactions, which directly reduces replication lag on write-heavy workloads. A new pg_roaringbitmap extension handles fast set operations on large integer collections, targeting audience segmentation and permission checks. Upgrade paths include RDS Blue/Green deployments, in-place upgrade, and snapshot restore. [Read more](https://app.daily.dev/posts/F2VATMgdl) ## Prisma Next benchmarks: 50% more throughput, p95 latency down from 41.8ms to 4.4ms Prisma Next, the TypeScript-native rewrite currently in Early Access, hits ~12,500 req/s peak versus ~8,300 for Prisma 7 — about 50% higher throughput. More telling is the latency story: at 6,000-7,000 req/s, p95 drops from 41.8ms to 4.4ms, which is the difference between a noticeable hiccup and nothing. It now reaches 87% of raw pg driver throughput, up from 58%. Bundle size shrinks from 1.32MB to 148.5KB gzipped, which matters for serverless and edge. Prisma 7 remains the production recommendation for now. [Read more](https://app.daily.dev/posts/SSSzV40UK) ## Query planner comparison across six databases: how join order search and hint exposure differ A thorough breakdown of how PostgreSQL, Oracle, Db2, MySQL, DuckDB, and Snowflake each search the join order space and how much control they hand back to users. Oracle and Db2 offer both sophisticated search and rich hint surfaces. DuckDB uses modern DPccp enumeration with minimal override. Snowflake removes all user controls entirely. MySQL is mid-migration from a greedy left-deep optimizer to a hypergraph-based one. PostgreSQL implements textbook System R dynamic programming and deliberately ships zero query hints — a principled stance the piece frames as rare and contested. Worth reading if you've ever reached for a hint in Postgres and found nothing there. [Read more](https://app.daily.dev/posts/2eB26OBIC) ## Xata cuts Postgres branch provisioning to 1-2 seconds, 1,000 branches for $1 Xata rebuilt branch provisioning using ZFS copy-on-write over NVMe-oF (their custom Xatastor storage layer) combined with warm pools of pre-provisioned Kubernetes pods. Branches were taking 20+ seconds; they now take 1-2 seconds. Because branches only store diffs to the parent and billing is per-minute with scale-to-zero, 1,000 short-lived branches run about $1. Real customers — AI coding agents and CI/CD preview environments — are running thousands of branches per week for under $30/month. The architecture separates compute from storage so volumes hot-connect to waiting pods in milliseconds. [Read more](https://app.daily.dev/posts/NPf3LZD7k) --- ## Also notable - **PostgreSQL default_statistics_target deep dive: per-column tuning beats global changes:** Raising default_statistics_target globally slows ANALYZE across the board — the right move is per-column tuning via ALTER TABLE ... ALTER COLUMN ... SET STATISTICS, reserving the global change for analytics workloads only. [Read more](https://app.daily.dev/posts/XQdZ6PrJ3) - **Databricks Spatial SQL GA: 90+ ST_ functions, up to 15x faster, Iceberg v3 geospatial types:** Databricks Spatial SQL is now generally available with 90+ ST_ functions, up to 15x query speedups on some operations, native map visualizations in AI/BI dashboards, and Iceberg v3 interoperability with native geometry/geography types — GEOMETRY and GEOGRAPHY are also being contributed to Apache Spark 4.2 for summer 2026. [Read more](https://app.daily.dev/posts/ta19gRY1T) - **Datadog and ClickHouse federated log querying now in Preview:** Two new Preview features let teams route logs to ClickHouse via Datadog Observability Pipelines and query them directly from Log Explorer without re-ingesting into Datadog — relevant context given Datadog's DASH announcements covered in yesterday's digest. [Read more](https://app.daily.dev/posts/ryDEd8SBE) - **lakeFS launches agentic AI data isolation: branch-scoped credentials, immutable audit trails, policy-gated merges:** lakeFS for Agentic AI gives each agent an isolated zero-copy data sandbox with branch-scoped cryptographic credentials and policy-gated merges into production, directly addressing the class of incidents where agents wrote to production data with no rollback path. [Read more](https://app.daily.dev/posts/Ckx5S1Rqz) - **npm v12 will disable install scripts and remote URL resolution by default; npm 11.16.0 previews the changes:** npm 11.16.0 is available now to preview what v12 will block: preinstall/install/postinstall scripts and git/remote URL resolution will be off by default, a breaking change for any pipeline relying on those hooks. [Read more](https://app.daily.dev/posts/G6Wf3GIB6) --- Tags: [#aws](https://daily.dev/tags/aws), [#backend](https://daily.dev/tags/backend), [#postgresql](https://daily.dev/tags/postgresql), [#prisma](https://daily.dev/tags/prisma) [View this post on daily.dev](https://daily.dev/posts/aurora-postgresql-18-lands-prisma-next-hits-50-faster-throughput-bcpi2vef6)