Repository navigation
Releases
v0.2.0
This release introduces a Raft-based replication track (new vibesql-consensus crate) and turns on the MVCC visibility + on-demand GC subsystem end-to-end. 369 commits since v0.1.4 also push SQLite compatibility forward — full SQLITE_MAX_TRIGGER_DEPTH (1000) trigger semantics, window-function correctness, ALTER TABLE RENAME COLUMN with trigger-body rewrite — and bring EXPLAIN QUERY PLAN output much closer to sqlite3 parity.
Replication & Consensus
A new vibesql-consensus workspace crate built on openraft introduces single-group whole-database replication. The MVCC state machine applies committed transactions from the Raft log; HTTP and prepared-statement writes can now be routed through consensus.
- Durable Raft log + vote persistence (#5357)
- In-process channel network + multi-node test cluster (#5364)
- TCP transport + cluster config +
make test-cluster(#5368) - Network snapshot transfer + durable snapshots + purge safety (#5369, #5374)
- MVCC state machine — apply committed txns from the Raft log (#5378, #5384)
- Freeze nondeterministic SQL at propose time (#5377, #5382)
- Freeze volatile DEFAULTs + validate replicated trigger bodies (#5396)
- Linearizable leader reads + stale-leader fencing (#5387)
- Bounded-staleness follower reads + read-your-writes tokens (#5390)
- Reject residual TIME→TIMESTAMP casts over non-target operands (#5398, #5405)
- Boot TCP test clusters on pre-bound ephemeral listeners (#5507, #5514)
Replicated server surface
The server now exposes the consensus surface across its HTTP and PostgreSQL-wire endpoints.
- Replicated operational surface — prepared-stmt writes, /health role, HTTP gating (#5411)
- Interactive transactions (BEGIN…COMMIT) in replicated sessions (#5402)
- Mid-transaction RYW + savepoints in replicated sessions (#5413)
- Route HTTP
/api/query+ CRUD-create writes through consensus (#5423) - Route HTTP CRUD by-id endpoints through consensus (#5441)
- Route HTTP GraphQL queries + mutations through consensus (#5447)
- Route HTTP blob storage writes through consensus (#5457)
- Buffer prepared EXECUTE inside a replicated transaction (#5430)
- Route replicated sessions' writes through MvccRaftNode (#5395)
- Feed HTTP SSE subscriptions from applied consensus entries (#5452)
- Coalesce apply-path change events for replicated subscriptions (#5456, #5467)
- Names-only replicated Describe to avoid materializing rows (#5503)
- Prune replicated subscription re-queries by changed PK (#5496)
- Resolve real column names in replicated SELECT results (#5428)
MVCC
VibeSQL's MVCC subsystem now covers the read path end-to-end, with on-demand garbage collection wired to VACUUM.
- Stamp xmin/xmax on INSERT/UPDATE/DELETE behind
mvcc_enabled(#5193) - TxnSnapshot + Row::visible_to predicate (Phase 1b) (#5184)
- Add xmin/xmax to Row + bump vbsql to v7 (Phase 1a) (#5142)
- Phase 1d read-path visible_to filtering + FK deferred-replay coordination (#5209)
- Extend visible_to to index scans, PK lookup, UNIQUE scans (#5204, #5222)
- Widen snapshots for autocommit + in-txn self-write visibility (#5223)
- On-demand
vacuum_mvccfor old-version garbage collection (#5224) - Thread visibility filter through DML read sites (#5205, #5225)
- Enable SIMD/columnar fast paths under
mvcc_enabled(#5227) - Parse
VACUUM/VACUUM INTOand map execution to MVCC GC (#5226)
SQL Compatibility
Triggers
Trigger semantics got a heavy pass to align with sqlite3 3.51.
- Honor
sqlite3_limit(SQLITE_LIMIT_TRIGGER_DEPTH)at runtime (#5605) - Reach SQLite's full SQLITE_MAX_TRIGGER_DEPTH (1000) via on-demand stack growth (#5610)
- Raise recursion depth cap 16 → 700 toward SQLite (#5533)
- Honor
PRAGMA recursive_triggers = off(#5550) - Schema-aware trigger firing for temp vs. main triggers (#5592)
- Parse
CREATE TEMP/TEMPORARY TRIGGER+ schema-qualified names (#5532) - Parse
CREATE TRIGGER:IF NOT EXISTS, rejectFOR EACH STATEMENT, surface body syntax errors (#5495) - Reject
CREATE TRIGGERwith bound parameter in body /WHEN(#5499) - Accept unparenthesized
UPDATE OFcolumn list (#5580) - Accept subquery in trigger
WHENclause (#5586) - Support
RAISE()in trigger bodies (#5415) - Reject
RAISE()outside trigger programs at parse time (#5424) - Resolve
NEW/OLDin triggerWHEN-clause subqueries (#5585, #5593) - Resolve
NEW.rowid/OLD.rowid(oid / rowid) in trigger bodies (#5519) - Live columnar values for BEFORE/AFTER UPDATE triggers (#5543, #5546)
- Live
count(*)for combined BEFORE+AFTER DELETE triggers (#5542) - Interleave BEFORE/AFTER row triggers per-row in multi-row DML (#5504)
- Cascade-drop triggers ON a dropped table (temp lifecycle) (#5597)
- Cascade-drop INSTEAD OF triggers when their view is dropped (#5604)
- List temp views + temp triggers in
sqlite_temp_master(#5582) - Transactional DDL rollback +
DROP TRIGGER IF EXISTS(#5506) UPDATE OR REPLACE/IGNOREconflict resolution + DELETE triggers on WITHOUT ROWID (#5522)- Fire child UPDATE triggers on
ON DELETE SET NULL/SET DEFAULTcascades (#5518) - Statement-end orphan FK check after cascade-fired
RAISE(IGNORE)(#5498) - Auto-commit statement atomicity for cascade /
RAISE(ABORT)and multi-row DML (#5473) OLD/NEWinINSERT…SELECTtrigger bodies + trigger*.test triage (#5493)- Rewrite trigger table refs + verbatim SQL on
ALTER TABLE RENAME(#5510)
Window Functions
- Match SQLite last-pass row order and sort
GROUP BYoutput by key (#5311) - Propagate
PARTITION BYevaluation errors instead of NULL-keying (#5305) - Honor
NULLS FIRST/LASTand collation in sort and RANGE frames (#5297) - Resolve aggregate args of embedded window functions via hidden columns (#5280)
- Evaluate window functions mixed with aggregates in one
SELECT(#5265) - Window IN-subqueries and ordinal
ORDER BYthrough wildcards (#5256) - window1.test residuals — nested scope binding, UNION-arm misuse, text coercion (#5273)
- WHERE push-down into window subqueries + covering-index window-sort EQP (#5349)
- Pre-compute aggregates in
OVER (PARTITION BY/ORDER BY/frame)(#5127) - Reject non-window aggregates in
ORDER BYof window queries (#5094, #5119) - Detect window-function misuse in positional
GROUP BY(#5111) COUNT()/COUNT(*)over post-aggregate frames now counts all frame rows (#5089)- Preserve REAL type in
sum()over float inputs GROUP_CONCATdispatch andRANGE ORDER BYvalidation (#5058)- Thread
case_sensitive_likePRAGMA through all filter paths - Recurse through all expression variants for window detection (#5095, #5121)
ALTER TABLE
ALTER TABLE RENAME COLUMNwith trigger-body column rewrite (#5602)- Abort
RENAME COLUMNon ambiguous trigger-body column ref (#5607) - Edit verbatim
sqlite_master.sqlin place forDROP COLUMN/RENAME TO(#5635) - Sync catalog schema on ALTER + edit verbatim SQL in place (#5633)
- Match sqlite3 wording for
ALTER TABLE RENAME TOcollision (#5560)
Foreign Keys
- Partial-index awareness unblocks fkey1-6.0/6.1/6.2 (#5213)
- Wire
sqlite3_db_status DBSTATUS_DEFERRED_FKSbridge (#5211) - Persist DEFERRABLE clause and skip auto-save inside txn (#5185)
- Self-ref multi-row INSERT + tcl-shim multi-table parse (#5180)
- Collation-aware comparison in CASCADE/SET NULL/SET DEFAULT helpers (#5176)
- Self-FK + deferral introspection (Phase C3 of #5085) (#5141)
- Deferred FK violation queue + COMMIT-time enforcement (Phase C2 of #5085) (#5133)
- Catalog metadata +
PRAGMA defer_foreign_keys(Phase C1 of #5085) (#5124) - Emit "foreign key mismatch" error class for invalid FK targets (#5120)
- Fire child row triggers on FK ON DELETE/UPDATE CASCADE (#5463)
Parser
- Parse
VACUUM/VACUUM INTO(#5226) - Support
WITHclause beforeVALUESstatement (#5358) - Support
date()andtime()as scalar function calls (#5316) - Accept
TEMP/TEMPORARYmodifier onCREATE TRIGGER(#5229) - Support partial-index
WHEREclauses inCREATE INDEX(#5109) - Support
RESTRICT, post-constraintDEFAULT, empty statements (#5088)
EXPLAIN QUERY PLAN
EQP output moved much closer to sqlite3 parity this cycle.
- Expand views / subqueries in EQP output (windowpushd) (#5354)
- Flatten plain views in EQP output (#5360)
- Render blocked view bodies as CO-ROUTINE blocks in EQP (#5366)
- Emit temp B-tree annotations for GROUP BY/DISTINCT/compound (#5370)
- Ordering-index scan lines + compound
ORDER BYshape (#5372) - Suppress
ORDER BYtemp B-tree for rowid-alias natural-order scans (#5380) - Emit EQP window-sort entries for
OVER-clause sorting passes (#5244) - Suppress statement
ORDER BYtemp-B-tree entry when covered by first window sort key (#5249) - Restrict window-sort index suppression to the innermost pass (#5250)
Storage & Indexes
- Schema-aware index manager — same-named indexes coexist across schemas (#5540, #5555)
- Schema-aware spatial indexes — same-name cross-schema coexistence (#5558, #5576)
- Schema-aware indexes +
sqlite_temp_master(#5513, #5539) - HNSW threshold-triggered graph compaction to restore recall (#5461)
- Vector index per-row maintenance — keep IVFFlat/HNSW in sync on non-compacting INSERT/UPDATE/DELETE (#5453)
- Rebuild IVFFlat/HNSW vector indexes on table compaction (#5449)
- Make full-tree index rebuild reversible mid-transaction (#5443)
- Undo disk-backed (spilled) index mutations on ROLLBACK (#5433)
- Scope spilled-index undo to statement savepoint (RAISE(ABORT)) (#5458)
- **Copy-on-writ...
v0.1.4 - SQLite Compatibility & Parallel Execution
This release focuses on SQLite compatibility and morsel-driven parallel execution. With 878 commits since v0.1.3, highlights include 100% TCL test pass rate on Priority 1 tests, memory-bounded operators with spill-to-disk, UPDATE FROM syntax, and CHECK constraint enforcement.
Highlights
🎯 100% TCL Test Pass Rate (Priority 1)
Complete SQLite compatibility for core SQL operations: SELECT, WHERE, JOIN, aggregate functions, and more.
⚡ Morsel-Driven Parallel Execution (Phase 10)
- Work-stealing parallel dispatcher with adaptive morsel sizing
- Parallel GROUP BY, hash join (build + probe), sort, and nested loop joins
- Per-operation morsel sizes tuned for each operation type
💾 Memory-Bounded Operators
- External sort with spill-to-disk for datasets larger than memory
- Memory-bounded aggregation and hash join with configurable limits
🔄 SQLite Compatibility
- UPDATE FROM - Multi-table UPDATE syntax
- DELETE with ORDER BY LIMIT - Ordered deletion
- CHECK constraint enforcement - Runtime validation
- Row value comparisons - Tuple expressions
- DEFERRABLE constraints - Foreign key deferral
- ON CONFLICT DO NOTHING/UPDATE - Upsert support
- json() function, RTRIM collation, and more
🛠️ CLI Improvements
- SQLite-style
.tables,.schema,.helpdot-commands - Positional database argument:
vibesql mydb.vbsql - Automatic database file format detection
🐛 488 Bug Fixes
Extensive fixes for NATURAL/USING JOINs, window functions, type affinity, LIKE/GLOB patterns, triggers, and more.
📖 Full changelog: CHANGELOG.md
🌐 Live demo: https://rjwalters.github.io/vibesql/
v0.1.3
[0.1.3] - 2025-12-08
This release focuses on Phase 9 performance optimization and cross-connection subscription enhancements. With over 500 merged PRs since v0.1.2, highlights include O(1) row deletion, skip-scan optimization, Bloom filter joins, selective column updates for subscriptions, and comprehensive observability metrics.
Performance
Phase 9b Core Optimizations
- Deletion bitmap - O(1) row deletion via bitmap-tracked slots (#3789)
- SimpleFastPath column caching - Cache column names for prepared statements (#3788)
- Streaming range scan - Iterator-based range scans avoiding full materialization (#3793)
- Single-row PK DELETE fast path - Optimized deletion for single-row primary key lookups (#3801)
- Skip unaffected indexes - Skip updating user-defined indexes when UPDATE doesn't touch indexed columns (#3800)
- Covering index scan - Return results directly from indexes without table lookup (#3804)
Query Execution
- Skip-scan optimization - Non-prefix index usage for range queries (#4081, #4088, #4089)
- Bloom filter joins - Build-side Bloom filters for hash join optimization (#4079)
- ONEPASS UPDATE - Single-pass optimization for single-row UPDATE operations (#4080)
- Streaming range SELECT - Pre-allocation for streaming range queries (#4073)
- Lazy FromResult materialization - Deferred row cloning in query results (#4068)
- DISTINCT containment check - Avoid cloning when checking DISTINCT containment (#4070)
- Streaming aggregation - Fast path for SUM range queries (#3818)
- Early projection - Project columns early in PK range scans (#3814)
- Range scan row cloning - Reduce cloning overhead in range scan fast path (#3848)
- Column-to-column predicates - Support column comparisons in columnar filtering (#4048)
- Columnar deduplication - DISTINCT queries in columnar join path (#3787)
- LIMIT/OFFSET in columnar - Support LIMIT/OFFSET in columnar join path (#3782)
- Column resolution caching - 34% faster Sysbench point lookups (#3593)
Optimizer Improvements
- Semi-join filter pushdown - Early semi-join filter pushdown for aggregate IN subqueries (#3685)
- Composite key selectivity - Fix composite key selectivity estimation (#3680)
- IN subquery conversion - Convert IN subqueries with GROUP BY/HAVING to semi-joins (#3666)
- Case-insensitive column lookup - Optimize column_index_cache for case-insensitive lookups (#3726)
Storage Engine
- SmallVec for Row.values - Reduce heap allocations for rows with ≤8 columns (#3954)
- String interning - Intern low-cardinality string columns (#3973)
- Arc<str> for strings - Use Arc<str> for Varchar/Character values (#3904)
- Batch index updates - Batch optimize B+tree and spatial index operations (#3877, #3882, #3888, #3896, #3916)
- DELETE hot path - Direct delete_by_pk_fast avoiding double-cloning (#3860)
- Lazy row ID adjustment - O(1) single-row deletes (#3725)
- Index point lookup - Reduce allocation overhead in index point lookups (#3701)
- Skip trigger overhead - Skip trigger overhead when no triggers defined (#3704)
- LIMIT-aware range scan - Early termination for LIMIT queries (#3699)
- Batch DELETE - Batch index updates for DELETE operations (#3693)
- Incremental index adjustment - Incremental adjustment for DELETE operations (#3416)
Index Optimizations
- O(n²) index creation fix - Fix O(n²) index creation scalability issue (#3679)
- Hash join row combination - Optimize row combination in hash joins (#3677)
- Plan caching - SimpleFastPath plan caching for prepared statements (#3676)
Data Loading
- TPC data loading - Optimize TPC data loading for large scale factors (#3640)
- Fix O(n²) batch insert - Fix O(n²) data loading in batch insert (#3671)
Added
Cross-Connection Subscriptions
- Real-time notifications - Notify subscribers across connections when data changes (#3825)
- Delta updates - Send only changed rows for cross-connection notifications (#3834)
- Async select optimization - Use async select for efficient notification dispatch (#3832)
- Selective column updates - Send only changed columns in subscription updates (#3843, #3854)
- Partial event type - New SSE event type for selective column updates (#3952)
- PartialRowUpdate messages - Wire protocol support for selective updates (#3930)
- Per-subscription config - Configure selective update thresholds per subscription (#3996)
- PK-based delta computation - Use primary key columns for efficient delta detection (#3895)
- Subscription filtering - Add filtering expressions to subscriptions (#3846)
- Protocol extensions - Add Ack, Pause, Resume subscription protocol messages (#3837)
- PK detection - Detect primary key columns for selective subscription updates (#3861)
Observability & Metrics
- Subscription metrics - Active subscriptions gauge metric (#4015)
- Bytes saved counter - Track bytes saved by selective updates (#4017)
- Eligibility breakdown - Metrics for selective update eligibility reasons (#4009)
- Partial update efficiency - Metrics for partial vs full update ratio (#3951, #3893)
- Selective-eligible gauge - Track subscriptions eligible for selective updates (#3922)
- HTTP efficiency endpoint -
/stats/subscriptions/efficiencyendpoint (#3986) - Observability documentation - Comprehensive metrics documentation (#4003)
Server Configuration
- Environment variable overrides - Configure server via environment variables (#4020)
- API key env vars - Set API keys via
VIBESQL_API_KEY(#4027) - HTTP auth env vars - Configure HTTP auth via environment (#4024)
- Selective update thresholds - Make column update thresholds configurable (#3969)
Developer Experience
- VIBESQL_DEBUG flag - Umbrella flag for enabling all debug output (#4056)
- JSON debug output - Structured JSON format for debug output (#4057)
- Profiling decision tree - Documentation for choosing profiling tools (#4058)
- Skip-scan in EXPLAIN - Show skip-scan plans in EXPLAIN output (#4087)
- CPU profiling guide - Guide for using samply profiler
- DELETE hot path profiling - Profiling instrumentation for DELETE operations (#3873)
- Range scan profiling - Add range scan profiling instrumentation (#3830)
SQL Compatibility
- CREATE TABLE IF NOT EXISTS - Standard SQL syntax support (#3820)
- PostgreSQL type conversion - Expanded type conversion coverage (#3897)
- Window functions in CASE - Support window functions in CASE and IS NULL expressions (#3813)
Internationalization
- 19 language translations - Complete translations for web demo:
- Locale switching - Runtime locale switching for web demo (#3819)
- Conformance page i18n - Locale support for conformance page
TypeScript Client
- Subscription events - Emit subscription events from Connection class (#3881)
- PartialData parser - Parse SubscriptionPartialData (0xF7) messages (#3870, #3886)
- Type narrowing fix - Resolve TypeScript type narrowing in Connection.query() (#3878)
Changed
- DML cost estimation - Use statistics-based cost estimates for DML optimization (#3968, #3974)
- Row size WAL cost - Consider row size in WAL write cost estimation (#3976)
- Statistics fallback - Prefer actual avg_row_bytes over schema estimates (#4018)
- Switched to samply - Replace flamegraph with samply (no sudo required)
- Separate benchmark features - Separate benchmark comparison features to avoid DuckDB overhead (#3594)
Fixed
Benchmark Integrity
- Remove SQL bypasses - Remove fast-path SQL bypasses from TPC-C benchmark (#4046)
- Sysbench SQL - Replace direct API bypass with SQL in sysbench update_index (#4047)
- DELETE timing - Fix sysbench DELETE benchmark to measure actual DELETE time (#4037)
- TPC-DS deduplication - Deduplicate TPC-DS benchmark results by averaging iterations (#3817)
Query Correctness
- Columnar cache invalidation - Add columnar cache invalidation for ALTER TABLE, INSERT, TRUNCATE (#3931, #3941, #3946, #3989)
- Missing columnar invalidation - Fix missing columnar cache invalidation in REPLACE path (#3890)
- Ambiguous column resolution - Resolve ambiguous columns to leftmost table in LEFT JOINs (#3783)
- Constant folding - Add constant folding before predicate extraction in columnar join (#3775)
- Semi-join equijoin extraction - Extract equijoin predicates for comma-separated tables in semi-join (#3744)
- Multi-column LEFT JOIN - Add multi-column LEFT OUTER JOIN support (#3723)
- Multi-column hash join - Add multi-column hash join for WHERE clause conditions (#3670)
- Table name normalization - Normalize table names to lowercase for case-insensitive lookups (#3661)
- Predicate pushdown validation - Validate predicate pushdown in covering index scan (#3811)
- Window functions in aggregates - Apply window functions to aggregate queries for AVG(SUM(...)) patterns (#3706)
Storage & Indexes
- Deleted row filtering - SELECT queries filter deleted rows from deletion bitmap (#3791)
- Index rebuild after compaction - Rebuild user-defined indexes after table compaction (#3808)
- Index rebuild after load - Rebuild indexes after loading data from binary format (#3607)
Server
v0.1.2
What's New in v0.1.2
Vector Search
- Distance operators -
<->(L2/Euclidean),<#>(inner product),<=>(cosine) - Distance functions -
l2_distance(),inner_product(),cosine_distance(),cosine_similarity() - IVFFlat index - Inverted file index for approximate nearest neighbor search
- HNSW index - Hierarchical Navigable Small World graph for high-performance ANN
HTTP Server & API
- REST API - Auto-generated CRUD endpoints for all tables
- GraphQL endpoint - Full query support with schema introspection
- Relationship resolution - Nested queries following foreign key relationships
- Authentication - API key and token-based HTTP authentication
- SSE subscriptions - Real-time data streaming with Server-Sent Events
- Delta updates - Efficient change notifications for subscriptions
- Backpressure handling - Configurable channel buffers and rate limiting
- Subscription limits - Per-connection and global quotas
- Retry with exponential backoff - Automatic recovery from transient errors
- Pagination -
limitandoffsetquery parameters
Blob Storage
- SQL integration -
vibesql_storagesystem table for blob metadata - HTTP endpoints - Upload and download blobs via REST API
- OpenDAL integration - Cloud storage backends (S3, GCS, Azure, local filesystem)
- TypeScript SDK - Storage methods in
@vibesql/client
SDK & Tooling
- Drizzle ORM adapter -
@vibesql/drizzlepackage using sqlite-proxy driver - TypeScript codegen -
vibesql-cli codegencommand for type-safe database access
Scheduled Functions
- Cron scheduling - Execute SQL statements on a schedule
- Job management - Create, list, pause, resume scheduled jobs
Changed
- Unified benchmark CLI (
scripts/bench) with consistent interface - Consolidated benchmark result processing into
process_results.py - Improved SQLite dialect compatibility for dogfooding scenarios
Fixed
- Clippy warning for approximate PI constant in tests
- WASM UUID generation with
jsfeature for RNG support
Installation
Rust (crates.io)
cargo add vibesqlnpm
npm install @vibesql/client
npm install @vibesql/drizzle # For Drizzle ORM integrationFull Changelog: v0.1.1...v0.1.2