Release 3.0

Bug fixes:

• Fixed computed fields being eagerly evaluated when not selected in SELECT queries

• Fixed $parent resolving to NONE inside graph traversal WHERE clauses

• Fixed LIMIT pushdown, plan-time parameter resolution, and computed field dependency analysis

• Fixed count() returning 1 for sets instead of counting items and added missing set function paths

Features & improvements:

• Added support for transaction conflict error reporting

• Increased v2 upgrade check frequency and added --upgrade flag for manual checks

Bug fixes:

• Fixed HNSW index compaction write conflicts

Features & improvements:

• Improved consistency for set and array operations

• Reduced import overhead and added benchmarks

• Added query planner and executor improvements

• Added GraphQL subscriptions

• Updated IndxDB to v0.12.0

Bug fixes:

• Fixed invalid time formatting strings causing a panic

• Fixed +=/-= silent no-ops with incompatible types

• Fixed START clause performance regression with pushdown KV skipping

• Fixed SurrealValue::from_value for all serde_json::Value variants

• Fixed BM25 search::score() returning 0 after index compaction

• Propagated actual query errors instead of misleading Connection uninitialised

• Fixed race condition in distributed task lease acquisition

• Fixed UPSERT SET with IF expressions always evaluating to the ELSE branch on existing records

• Fixed LIMIT producing fewer rows when WHERE is only partially resolved by the index

• Fixed subqueries not respecting nested fields with AS key.key

• Fixed index compaction KeyAlreadyExists and TransactionConflict errors

Features & improvements:

• Improved idempotency for transaction operations

• Updated SurrealKV to 0.21.0

• Added missing exports for API, BUCKET, MODULE, and CONFIG definitions

• Added NaiveDate implementation for SurrealValue in the SDK

• Fixed ORDER BY performance regressions with compound indexes

• Added root level field comments to GraphQL schema generation

• Implemented v2 subcommand to run the legacy binary

• Added retry logic for initial datastore transactions on concurrent startup

Bug fixes:

• Returned None for records that don't exist instead of a confusing error

• Fixed write-write conflicts on IP keys during IndexAppending

• Fixed serialization error of RecordIdKeyType::Object

• Ensured transaction timeouts were respected for queries

• Applied new executor bug fixes and optimizations

• Fixed DEFINE FUNCTION parsing from disk

• Fixed permission checks for DELETE live events

• Fixed compound unique indexing

Features & improvements:

• Added functions to set parts of datetimes

• Updated timestamp code to use boxed traits

• Made deriving SurrealValue more convenient when using the SDK

• Added error chaining infrastructure

• Added --tables-exclude flag to the surreal export command

Bug fixes:

• Fixed record IDs not being dereferenced when a field is computed on them

• Prevented creating indexes on computed fields

• Patch fixes for version errors and /sql endpoint

• Allow durations to be multiplied and divided by numbers, and incremented and decremented like numbers

• Fixed permission check condition for root users

• Improved environment variable values

• Parallelized index compaction across distinct indexes

• Fixed GraphQL literals for string enums

• Improved WASM compatibility

• Fixed error serialization

• Fixed new planner and deduplication logic

After numerous alpha and beta releases over more than a year of development, we are releasing SurrealDB v3.0.0! 🎉. The changes for this release pertain to those over all the alpha and beta versions before 3.0.0, starting with 3.0.0-alpha.1 and the last 3.0.0-beta.4 and finally the changes between the final beta version and stable 3.0.0.

Items in the release notes marked as 🆕 are the newest changes from the final beta and 3.0.0.

To migrate from a 2.x version to 3.0, please see the following page which goes into further detail on the breaking changes and how to use migration diagnostics to obtain a list of actions for your particular schema.

Performance and optimisations:

• 🆕 Internal streaming execution (new query planner)

• 🆕 Improve scanner for batched retrievals

• 🆕 Enable concurrent writes on HNSW vector index

• 🆕 Add support for skipping keys and values at storage engine layer

• 🆕 Implement multi-get fetches for graph edge scans

• Improved GROUP and ORDER BY execution correctness

• Improved edge lookups: Only fetch edge keys when values are unused

• Added criterion benchmarks for executor and functions

• Fetch ns, db, table, and fields once

• Reduced heap modifications in MemoryOrderedLimit

• Optimise index scans (descending) order and limit batch sizes

• Yield during CF/GC collection

• Optimise some array methods

• Invert expression value relation and move ast types out of value.

• Integrated CRUD benchmarking suite

• Enable query planner on DELETE, UPDATE and UPSERT statements

• Implemented reverse scan for table and range iterator

• Added memory limit threshold check to all statements

• Reduced storage and serialisation cost

• Optimised index iterators: stop scanning after LIMIT

• Added support for multiple writers in Full-Text search implementation

• Added support for range scans on compound indexes (prefix + range)

• Optimised START/LIMIT execution: correct semantics with WHERE/ORDER and early-stop iteration

• Added HTTP connection pooling to maximise throughput

• Removed Force Index mechanism and unify index building across all platforms

• Split AST & Logical Plan into separate components

• Modified allocation tracking behaviour and implementation

Features, additions and syntax improvements:

• Unified lookups for graph & references

• 🆕 Improve GraphQL integration and implementation

• 🆕 New datastore scheme with parameters

• Implemented permissions support for DEFINE API

• Hash-based vector deduplication for HNSW indexes

• Overhauled DEFINE API middleware logic

• Introduced Regex size limit configuration

• Added Filesystem support

• Added cbor encoding and decoding functions

• Added support for distributed sequences

• Added SQL base64 encoding/decoding with & without padding

• Added support for INFO FOR as a subquery

• Added support for filtering down graph selections with a range selection

• Exposed QueryType for live and kill responses

• Added RPC live statement options

• Added LQ Killed notification

• Allowed passing fields to RPC live

• Added support for partial array types in schemafull tables

• Introduced ALTER FIELD, update ALTER TABLE

• Enhanced Full-Text Search with OR Boolean Operation

• Added support for converting to/from protobufs and flatbuffers

• Introduced COUNT indexes for constant-time count() queries

• Introduced composable HTTP router factory for embedders

• Added Support of standard deviation and variance in foreign tables

• Use surrealdb_types::* in SDK and as core's public interface.

• Extended mock syntax to support full range

• Added support for variables in closures and live queries

• Allow access to $input parameter in EVENTS

• Added support for multi-session support in RPC protocol

• Added refresh token support to the RPC protocol and SDK

• New Surrealism plugins

• Added support for client-side transactions

• Memory-bounded LRU cache for HNSW vectors; remove deprecated MTREE index

• Moved strict mode to DB definition, added strict object type

• Added ALTER INDEX statement with PREPARE REMOVE for safe index decommissioning

• Added Default NS & DB

• Introduced a None Type

• Added record::is::edge() function

• Added string::capitalize() function

• Added array::sequence() function

• Added Value::Set / Set type

• Fixed CBOR and add type::of function.

• Added Infinity float value

• Added method syntax for record::is_edge()

• Added array functions for natural and/or lexical sorting

• Added set literal syntax

• Improved parameterisation of statements

• Added Finnish to snowball filter

Bug fixes:

• 🆕 Fix HNSW index memory explosion

• 🆕 Add support for versioned count queries

• 🆕 Reimplement ID generator to resolve concurrency freezes

• 🆕 Fix schemafull index validation for sub-field paths on structured field types

• 🆕 Fix count() AS optimisation ignoring field aliases with GROUP ALL

• 🆕 Fix WebSocket client memory leak in pending request cleanup

• Fixed async event retries to correctly roll back failed executions

• Fixed JWT decoding issues

• Fixed concurrent index builds dropping updates under heavy write load

• Fixed backwards compatibility issues in authentication limits

• Fixed HNSW layer popping logic

• Fixed edge cases in ONLY, OMIT, and SELECT VALUE

• Fixed edge cases in FETCH and ORDER BY

• Fixed ORDER BY behaviour for sets and Unicode characters

• Fixed INSERT RELATION preserving extra fields when using table variants

• Fixed parameterized clause validation

• Fixed Linux Docker builder issues with OpenSSL and ONNX Runtime linking

• Fixed datastore shutdown failures

• Fixed supported SDK version reporting

• Removed unsupported VERSION clause from CREATE and INSERT

• Ensured api::invoke omits execution context where appropriate

• Fixed race condition in Scanner causing duplicate range fetches

• Fixed panic in HNSW vector cache eviction under async contexts

• Fixed LIVE SELECT not returning UUID

• Fixed GraphQL parsing: excluded incorrectly treated as included

• Fixed bug in casting logic

• Fixed idiom continuity not broken by parentheses

• Fixed table serialisation bug

• Fixed LIVE DIFF response structure

• Fixed array slice iterating over own items

• Fixed arithmetic wrap panic in duration functions

• Fixed index out of bounds error

• Fixed overflow and panics in math functions

• Improved ALL/ANYINSIDE behaviour with strings and arrays of strings

• Improved over-fetch documents when LIMIT and START used together

• Fixed Tokeniser fail with successive punctuation characters

• Fixed the default endpoint for exports and imports

• Fixed Graph clauses ->(SELECT ... issue in Rust SDK

• Fixed select with start and limit for arrays

• Fixed allowing null bytes in strings

• Fixed duplicate ID in GraphQL

• Fixed a bug with conversions to union types like array<int> | bool

• Fixed rand::time and other time-related issues

• Canonicalised allowed paths

• Normalised decimal values to fix indexing results

• Fixed concurrent indexing on quoted table names

• Fixed indexing on filtered count results

• Ensured arguments for object idiom methods are computed

• Fixed geometry parsing bug

• Reworked subqueries/covered expressions

• Allowed idiom functions to be restricted by capabilities

• Fixed regex_non_string.surql language-test for UUIDs

• Fixed memory usage and performance of Garbage Collection on old databases

• Fixed method calls on negative numbers

• Fixed JSON multipoint parsing

• Fixed START clause with WHERE; safer optimisation

• Fixed LIVE SELECT parsing with type::table()

• Limit fields in live query delete notifications

• Fixed invalid table view results with sum() returning 0

• Std index shouldn't fail if value already exists

• Fixed potential deadlock in distributed sequence allocation

• Replaced BTreeDocIds with SeqDocIds in M-Tree, remove ANALYSE, use Strand for RecordKey::String

• Enabled referencing fields in default value for defined fields.

• Removed dec when casting decimal to a string, allow string casting from None and Null

• Restricted supported types on id field

• Fixed incorrect ORDER BY LIMIT results on multi-column index scans

• Added notification for caller when a connection is dropped in Rust SDK

• Fixed ignored insert with unique index creating a record when index violated

• Fixed strings, identifiers and escape sequences

• Fixed Windows file path handling in bucket operations

• Fixed WebSocket limits configurable and enforce max message size

• Fixed possible invalid formatting of idioms

• Fixed aliases not being respected in views

• Fixed invalid formatting of statement expressions within general expressions

• Ensured subqueries and blocks have consistent outputs

• Added CONTAINS/INSIDE planning consistent with substring semantics

• Ensured destructuring with computed fields returns correct output

• Improve behaviour of group by and improve performance and robustness of views

• Fixed formatting bugs related to empty identifiers

• fixed the info rpc

• Ensured all statements return a value

• Fixed TableCheck output for TYPE RELATION / NORMAL

• Fixed SELECT with SPLIT so ORDER BY and START/LIMIT apply after splitting

• Added exact hint when path is deprecated

• Fixed Number hash to use consistent value hashing

• Fixed OOM crash in COUNT index on large datasets by batching key retrieval

• Fixed provided vars not being attached to context in surrealism 'sql_with_vars' function

• Fixed literal objects allow unknown keys

Workflow, observability and developer experience improvements:

• 🆕 Improve error communication and serialisation

• Added /data and /logs directories to the Docker container

• Ensured a checksum is generated for Windows binaries

• Added debugger setup instructions

• Improved metrics consistency

• Overwrite by default during data import

• Improve context cancellation checking with adaptive back-off strategy for better OOM protection

• Improved Surrealism Wasm

• Implemented slow logs

• Tokio Console integration

• Added support for log files, json log format, and advanced log span filtering

• Added configurable slow-query logging with parameter allow/deny filtering and normalised single-line output

• Included event errors in output

Storage/index format and protocol:

• Normalised numeric key encoding to unify ordering

• Introduced DecimalLexEncoder for Number

• Used IndexId for Index keys

• Stored keys using IDs for Namespace/Database

• Improved key-value storage engine layer

Security/networking:

• Reject IPs resolving to forbidden net targets

• DNS resolution on Context::check_allowed_net

• Configurable user agent for HTTP requests