inst/tpch/README.md

In ursa-labs/arrowbench: Tools for Continuous and Interactive Benchmarking

Why is there a separate duckdb data directory here?

The DuckDB data generator actually produces data that is slightly out of spec. Specifically, the _address columns generate slightly different data. Generally speaking, this isn't a big deal since the queries don't use pattern matches on those columns, but they do show up in some of the answers.

But it's plain to see if you look at the official answers and the duckdb ones for a query that includes a _address column:

https://github.com/databricks/tpch-dbgen/blob/6985da461c641fd0d255b214f2d693f1bf08bc33/answers/q2.out https://github.com/duckdb/duckdb/blob/c0a4ab96c626426961c207f49c19aa81448e91da/extension/tpch/dbgen/answers/sf1/q02.csv

Additionally, DuckDB >= 0.8 has also changed slightly some of the s_comment columns. This doesn't impact queries themselves, but the answers changed. Note the answer differences before PR #136.

ursa-labs/arrowbench documentation built on July 8, 2023, 11:36 a.m.