bump version to 2.3.0

[lalepee]

Release Notes - CosmoTech Acceleration Library v2.3.0

Release Date: April 2026

Overview

Version 2.3.0 introduces a new input collection framework for lazily resolving datasets, workspace files, and parameters, as well as native Parquet support in the Store and a new CLI command. This release also includes several important bug fixes for PostgreSQL metadata management and foreign key constraint handling.

✨ New Features

Input Collector Framework

• New InputCollector abstraction (cosmotech/coal/utils/input_collector.py) — Provides a unified interface for lazily resolving simulation inputs from environment-configured paths
• DatasetCollector — Resolves dataset files by name from the configured dataset path
• WorkspaceCollector — Resolves workspace files from the configured workspace path
• ParameterCollector — Resolves parameter values by name from the configured parameters path
• ENVIRONMENT_INPUT_COLLECTOR singleton — A ready-to-use collector instance pre-configured from environment variables

Parquet Store Support

• store_parquet_file — Reads a .parquet file into a Store table with sanitized column names
• convert_store_table_to_parquet — Exports a Store table to a .parquet file
• load-parquet-folder CLI command — New csm-data command that loads all .parquet files from a folder into the Store (options: --store-folder, --parquet-folder)

Configuration Enhancements

• Added output and tmp Cosmo environment variable mappings to the configuration object

🔧 Improvements

Dataset API

• Dataset part name uses stem — When uploading dataset parts, part_name is now derived from the file name without its extension (Path(file_name).stem), producing cleaner part identifiers

🐛 Bug Fixes

PostgreSQL — Runner Metadata

• Fixed foreign key constraint violations — Runner metadata deletion now cascades correctly; the add_foreign_key method drops any pre-existing metadata constraint before re-adding it (DROP CONSTRAINT IF EXISTS metadata)
• Fixed ADD CONSTRAINT SQL syntax — Corrected malformed ALTER TABLE statement in add_foreign_key
• Added ON DELETE CASCADE to the runner metadata foreign key to automatically clean up dependent rows
• Schema-qualified referenced table — Foreign key now correctly references the table name with the right schema-qualifier
• Fixed autocommit flag — add_foreign_key now uses autocommit=False to properly participate in transactions
• Replaced upsert with DELETE + INSERT — Runner metadata update logic now deletes the existing row by id before inserting to avoid conflicts with the UNIQUE constraint on last_csm_run_id
• Added UNIQUE constraint on last_csm_run_id in the runner metadata table

Store

• Fixed execute_query with empty parameters — Queries with no bind parameters no longer raise an error

Code Quality

• Removed unused log message in runner metadata handling
• Minor consistency fixes across csv.py, native_python.py, pandas.py, parquet.py, and pyarrow.py store modules following Copilot review

🔄 Breaking Changes

None. This release maintains backward compatibility with version 2.2.x.