SurrealDB 4.0 Adds a MySQL Listener and a PostgreSQL SQL Dialect to Its Postgres Port, but Its Blog and Release Notes Differ on Window Functions
SurrealDB 4.0 (November 3) adds PostgreSQL and MySQL SQL dialects that its blog calls a defined subset; the vendor advises existing 3.x users to wait for 4.1, and its 22x indexed-query gain is projected.
Overview
SurrealDB 4.0 was released on November 3, 2026, according to its release notes, and the notes say its --postgres-bind listener “now runs PostgreSQL SQL” while a new --mysql-bind listener runs MySQL. The date is the release notes’ own; GitHub’s release page shows the same release published at 2026-11-03T15:22:58Z. Blocks & Files reported the release on November 5 and noted that the company made no formal news announcement, relying instead on webpages, a blog post and posts on X. The Herald had previously reported that SurrealDB 3.3 added a Postgres wire-protocol listener.
What Changed for SQL Clients
The upgrade guide says a connection to the PostgreSQL listener now runs the PostgreSQL dialect and can switch to ISO GQL, whereas “On 3.3 a connection ran SurrealQL unless it selected GQL.” The protocol documentation adds that SurrealQL is no longer available over this protocol from 4.0, so a client that sent SurrealQL over the listener must move to HTTP, RPC or an SDK.
According to the release notes, “Joins, subqueries and window functions give the answers PostgreSQL and MySQL would,” and psql, mysql, drivers and ORMs such as Drizzle, Prisma and TypeORM “work with it as they are.” That is the vendor’s statement, and the blog describes the result as “a defined subset of SQL”; none of the cited sources reports an independent test of the dialects or of those ORMs. The protocol documentation lists unsupported items, among others: recursive common table expressions and FULL OUTER JOIN are refused with SQLSTATE 0A000, as are INTERSECT ALL and EXCEPT ALL.
The release notes list other changes, among others: SurrealKV becomes the only on-disk storage engine (RocksDB and IndexedDB are removed), SurrealML is removed, and Surrealism modules are renamed extensions with signer and capability declarations.
SurrealDB’s Pages Differ on Window Functions
SurrealDB’s launch blog post says “Window functions, recursive CTEs and FULL OUTER JOIN are refused with a named error (0A000).” The FAQ on the SurrealDB 4 page likewise says “recursive CTEs and window functions are refused.” The release notes and the protocol documentation instead list window functions (row_number, rank, lag, lead and others) as running, with PARTITION BY, ORDER BY and ROWS frames. The release notes add that “GROUPS frames and RANGE with a numeric offset are refused,” and the protocol page lists further refused forms. None of the pages explains the difference, so the Herald does not reconcile them. The blog page’s structured data gives a publication time of 2026-11-03T16:52:16Z and a modification time of 2026-11-10T11:25:20Z; the sentence is present in the version the Herald read. The SurrealDB 4 page’s FAQ answer was read on the same day.
Upgrade Guidance From the Vendor
The blog post says “SurrealDB 4 is recommended for new (greenfield) database deployments,” and for existing 3.x deployments “we recommend upgrading to SurrealDB 4.1 once it becomes available.” The upgrade guide describes the in-place upgrade as “experimental in 4.0” and says it is one-way: once a 4.0 node has opened a datastore, a 3.x build refuses to start on it. Blocks & Files reports the same recommendation to wait for 4.1.
A first patch, 4.0.1, followed on November 9 (release notes; GitHub’s release page shows 2026-11-09T16:56:21Z). Its notes say the LICENSE file now names SurrealDB 4.0 as the Licensed Work under the Business Source License 1.1 with a Change Date of 2030-11-03, and that “4.0.0 shipped with the parameters for SurrealDB 3.0.” They also say 4.0.1 refuses an approximate KNN search that no vector index serves, where 4.0.0 returned no rows.
Performance Figures Are the Vendor’s Own
SurrealDB’s benchmarks page reports results from the open-source crud-bench harness on one machine (an AMD Ryzen Threadripper 9970X, last run November 3, 2026). The page is published on SurrealDB’s own site, and the cited sources do not say whether anyone outside the company has run it. For the 3.x-to-4.x comparison the page lists gains of 42% for creates, reads, updates and deletes (mean), 130% for full-table scans and 2121% for indexed queries. The page states that “SurrealDB’s indexed query figures are projections: the measured throughput scaled by 1.63 (1.69 for the OR of two equalities).” The SurrealDB 4 page and Blocks & Files repeat the roughly 22x indexed-query figure as projected; Blocks & Files reports no measurement of its own.
The same page compares SurrealDB’s server engine with PostgreSQL 18, MySQL 26.10, MongoDB 9.0, ArangoDB 3.12 and Neo4j 2026.04. Against PostgreSQL 18 the page lists SurrealDB’s mean CRUD throughput as 12% slower, its point reads by record id as 47% slower, and its mean write throughput (creates, updates, deletes) as 1.8 times faster. The SurrealDB 4 page adds that a 1-million-vector, 768-dimension HNSW index built in 986 s instead of 7,290 s, “in parallel batches on a 64-core server.”
What We Don’t Know
- None of the cited sources reports an independent benchmark of SurrealDB 4.0 or an independent test of its PostgreSQL and MySQL dialects.
- It is not stated in the cited sources which patch builds the “3.x” and “4.x” rows on the benchmarks page correspond to.
- Whether the launch blog’s window-function sentence will be corrected is not stated.
- The descriptions and figures for SurrealDB Earth (the serverless offering that runs on Cloudflare) and SurrealDS come from the vendor; the cited independent report repeats the descriptions and reports no measurements.