Auxten Wang

## Performance issue on c6a.metal vs c6a.4xlarge chDB (Parquet, partitioned) c6a.metal is slower than c6a.4xlarge on [Q0-Q7, Q36-Q42](https://benchmark.clickhouse.com/#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) chDB (MergeTree) c6a.metal is slower than c6a.4xlarge on [Q1-Q7, Q36-Q42](https://benchmark.clickhouse.com/#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) ## Performance...

Support complex type in Python Table Engine like Array and Map, Object

2

Code snippet to reproduce: ```python df = pd.read_parquet(f'{GRAPHRAG_FOLDER}/create_final_community_reports.parquet', columns=["id","community","level","title","summary", "findings","rank","rank_explanation"]) batched_import(statement, df) df.head(2) df.dtypes import chdb df=df.astype({'id': 'string'}) df2=df[["id","level","rank"]] chdb.query("FROM Python(df2) SELECT max(id),sum(level),avg(rank) group by all").show() df.dtypes ``` [parquet-test-data.zip](https://github.com/user-attachments/files/16361689/parquet-test-data.zip)

Some of these are not papers, but still very good to study database theory: ## Database Systems History and Evolution [RDBMS_Genealogy_V6.pdf](https://github.com/user-attachments/files/16811957/RDBMS_Genealogy_V6.pdf) ## Query Engines and Query Optimization [Assembling a Query...