MatrixOne v26.4.1.1 Release Notes

Release Date: July 16, 2026 MatrixOne Version: v26.4.1.1 (v4.1.1)

MatrixOne v26.4.1.1 is a patch release that expands Parquet cross-type conversion coverage for LOAD DATA, fixes data branch merge support for BLOB/BINARY/VARBINARY columns, corrects publication table lookup for databases with quoted identifier names, fixes a regular-index ORDER BY LIMIT cursor result correctness issue, fixes uint64-to-int64 division type promotion, and adjusts the default incremental checkpoint interval to 5 minutes.

Bug Fixes

  • Data branch merge with BLOB columns (#25569): DATA BRANCH MERGE now correctly handles tables that contain BLOB, BINARY, and VARBINARY columns. Previously, binary column values were serialized using SQL string escaping, which could corrupt bytes containing control characters during merge. Binary values are now encoded as hex literals (x'...'), preserving every byte exactly.

  • Publication table lookup for quoted database names (#25532): Creating and altering publications now correctly quotes database names in internal SHOW TABLES queries. Databases with Chinese names or names containing special characters (such as backticks) that require identifier quoting now work correctly with CREATE PUBLICATION and ALTER PUBLICATION.

  • Regular index ORDER BY LIMIT cursor fix (#25543): When using a regular (non-primary) index with ORDER BY + LIMIT combined with an additional cursor filter on the primary key (e.g., AND id < 'some_value'), the optimizer previously could push down an incorrect reading limit to the index scan, potentially returning incomplete or wrong results. The optimizer now correctly skips the ordered limit pushdown when a residual filter exists beyond the index prefix equality conditions.

  • UINT64 / INT64 division type promotion fix (#25566): Division between BIGINT UNSIGNED and BIGINT columns now correctly promotes both operands to DECIMAL128 to preserve precision. Previously, an incorrect casting path could cause precision loss in the result.

Improvements

  • Parquet cross-type conversions for LOAD DATA (#25510): When loading Parquet files via LOAD DATA INFILE, MatrixOne now supports automatic cross-type conversion between a broad range of Parquet source types and MatrixOne target column types. New conversions include: floating-point to integer (with standard rounding), boolean to float/decimal/JSON, integer to boolean/JSON/enum, decimal to integer, date↔timestamp, timestamp→date/time, time→datetime, string to bool/bit, float list↔vecf32/vecf64, and integer/float/boolean/date-time to JSON. Out-of-range values are rejected with descriptive overflow errors.

  • Default incremental checkpoint interval changed to 5 minutes (#25432): The default incremental-interval for checkpoints has been changed from 60–180 seconds to 300 seconds (5 minutes) across all deployment configurations. This reduces checkpoint overhead while maintaining data durability.

Compatibility Notes

  • DATA BRANCH MERGE now supports BLOB, BINARY, and VARBINARY columns. Tables with these column types that previously could not be merged will now work correctly.

  • The LOAD DATA INFILE Parquet importer now accepts many more source-to-target type combinations. Loads that previously failed with type-mismatch errors may now succeed with automatic conversion.

  • The incremental-interval checkpoint default has changed. Deployments that previously relied on the 60s or 180s default will now use 300s unless explicitly configured.

About This Document

This document is generated by the docs sync agent based on the MatrixOne source diff and reviewed by humans via pull request.

Full Changelog

v4.1.0…v4.1.1