Deciphering Kùzu v0.13.6: Structural Improvements, Core Memory Layout Refactoring, and Stability Patches
The v0.136 release is essential for developers seeking stability, particularly those working across different operating systems. While it introduces a notable breaking change, it also resolves several key bugs that could cause unexpected behavior and errors.
Handling unstructured or semi-structured data in graph schemas requires strong UNION data type handling. kuzu v0 136 fixed
The v0.13.6 release is a targeted patch designed to harden the system against runtime crashes, incorrect query evaluations, and memory leak vulnerabilities. Below is a breakdown of the most significant fixes introduced in this version. 1. Memory Management and Leak Fixes
The improvements implemented in the v0.13.6 codebase established a clear structural divergence from previous legacy iterations: Feature/Metric Legacy Kùzu Versions Kùzu v0.13.6 (Fixed & Optimized) Restricted precision Native DOUBLE precision columns In-Memory HNSW Uncompressed node indexing Compressed neighbor offsets Match Clauses Sequential, independent scans Merged consecutive MATCH logic Data Ingestion Prone to COPY FROM subquery bugs Safe, type-checked subquery streaming Buffer Management Unoptimized individual evictions Batch-processed eviction candidates The Legacy of Kùzu v0.13.6 Deciphering Kùzu v0
For those unfamiliar with the project, Kùzu is a graph database written in C++ that operates in-process, similar to how SQLite or DuckDB work for relational data. It is explicitly built to handle graph workloads efficiently on modern multi-core CPUs. Key Features of Kùzu
| Metric | Kuzu v0.135 (unstable) | Kuzu v0.136 (fixed) | Improvement | | --- | --- | --- | --- | | Average query latency | 340 ms | 212 ms | | | Memory usage (peak) | 5.2 GB | 1.8 GB | 65% reduction | | Multi-threaded throughput | 1,200 ops/sec | 2,450 ops/sec | 104% increase | | Crash rate (24 hours) | 1 crash per 6 hours | 0 crashes | 100% stable | The v0
Kùzu utilizes vectorized and factorized query processing to parallelize execution across multi-core systems.
Shortly after the release of this codebase, Kuzu Inc. announced they would no longer actively maintain the project under its original commercial umbrella. However, because the technology is licensed under the permissive , the community quickly rallied to preserve its advanced vectorized join algorithms.