v0.2.1

Fixes

• MySQL/ClickHouse CAST: CAST(AS VARCHAR) → CAST(AS CHAR) for MySQL, CAST(AS STRING) for BigQuery/Databricks, CAST(AS String) for ClickHouse. Affected checks: not_missing, min_length, max_length.
• Regex backslash on MySQL/ClickHouse: \d, \w, \. patterns were silently stripped by string parser. Now doubled for dialects that consume backslashes.
• reference check in SQL runner: result was silently dropped (missing parser). Now returns orphan_count correctly.
• agg_row crash: defensive handling when aggregation query returns empty or unexpected result.

CI

Integration tests against 4 real engines on every push:

Engine	Tests	How
PostgreSQL	18	Service container
MySQL	16	Service container
DuckDB	17	In-process
Polars	19	DataFrame/parquet/CSV

All tests use the same 10-row dataset with exact fail count assertions.

workflow_dispatch dropdown to run one specific dialect when fixing a bug.

v0.2.0

New

• [Feature] Lazy Polars import -- SQL path cold start 20ms instead of 176ms (Polars loads only when DataFrames used)
• [Feature] Default sample queries select only checked columns instead of SELECT * -- cheaper on BQ/Athena with no configuration
• [Feature] sample_id and sample_columns YAML options for full control over sample query cost
• [Feature] Benchmark script supports 259M rows via --full

Fixes

• [Fix] SQL errors report status=error, exit_code=2 (was silently passing)
• [Fix] Regex backslash preservation in all 8 SQL dialects
• [Fix] Connector dialect correctly routed to SQL generation
• [Fix] HAVING COUNT(*) > 1 for Postgres/Redshift/Snowflake unique check
• [Fix] COUNT(CASE...THEN 1 END) aggregation (26% faster than SUM)

Verified benchmarks

• Athena: 2x faster than Soda Core (179M-492M rows, parallel samples + 20ms cold start)
• BigQuery: similar speed to Soda (slot scheduling dominates), up to 92% cheaper with sample_id on wide tables
• DataFrames: 16x faster than Pandera on 126M rows (both check everything)
• Streaming: 259M rows from disk in 1.1s, under 1 GB RAM

pip install eliza-dq --upgrade

v0.1.3

• [Feature] sample_id and sample_columns for cheaper DWH sample queries (up to 92% cost reduction on wide BQ tables)
• [Feature] Parallel sample collection via ThreadPoolExecutor
• [Fix] SQL errors now report status=error, exit_code=2
• [Fix] Regex backslash preservation in all SQL dialects
• [Fix] Connector dialect correctly routed to SQL generation
• [Fix] HAVING COUNT(*) > 1 for Postgres/Redshift/Snowflake unique check
• [Fix] COUNT(CASE...THEN 1 END) optimized aggregation (26% faster)
• [Docs] Verified benchmarks on 10 fresh tables (BQ + Athena, cache disabled)

pip install eliza-dq --upgrade