sm4rtm4art commented on code in PR #18051: URL: https://github.com/apache/datafusion/pull/18051#discussion_r2435962352
########## docs/source/user-guide/arrow-introduction.md: ########## @@ -0,0 +1,301 @@ +<!--- + Licensed to the Apache Software Foundation (ASF) under one + or more contributor license agreements. See the NOTICE file + distributed with this work for additional information + regarding copyright ownership. The ASF licenses this file + to you under the Apache License, Version 2.0 (the + "License"); you may not use this file except in compliance + with the License. You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, + software distributed under the License is distributed on an + "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + KIND, either express or implied. See the License for the + specific language governing permissions and limitations + under the License. +--> + +# A Gentle Introduction to Arrow & RecordBatches (for DataFusion users) + +```{contents} +:local: +:depth: 2 +``` + +This guide helps DataFusion users understand Arrow and its RecordBatch format. While you may never need to work with Arrow directly, this knowledge becomes valuable when using DataFusion's extension points or debugging performance issues. + +**Why Arrow is central to DataFusion**: Arrow provides the unified type system that makes DataFusion possible. When you query a CSV file, join it with a Parquet file, and aggregate results from JSON—it all works seamlessly because every data source is converted to Arrow's common representation. This unified type system, combined with Arrow's columnar format, enables DataFusion to execute efficient vectorized operations across any combination of data sources while benefiting from zero-copy data sharing between query operators. + +## Why Columnar? The Arrow Advantage + +Apache Arrow is an open **specification** that defines how analytical data should be organized in memory. Think of it as a blueprint that different systems agree to follow, not a database or programming language. + +### Row-oriented vs Columnar Layout + +Traditional databases often store data row-by-row: + +``` +Row 1: [id: 1, name: "Alice", age: 30] +Row 2: [id: 2, name: "Bob", age: 25] +Row 3: [id: 3, name: "Carol", age: 35] +``` + +Arrow organizes the same data by column: + +``` +Column "id": [1, 2, 3] +Column "name": ["Alice", "Bob", "Carol"] +Column "age": [30, 25, 35] +``` + +Visual comparison: + +``` +Traditional Row Storage: Arrow Columnar Storage: +┌──────────────────┐ ┌─────────┬─────────┬──────────┐ +│ id │ name │ age │ │ id │ name │ age │ +├────┼──────┼──────┤ ├─────────┼─────────┼──────────┤ +│ 1 │ A │ 30 │ │ [1,2,3] │ [A,B,C] │[30,25,35]│ +│ 2 │ B │ 25 │ └─────────┴─────────┴──────────┘ +│ 3 │ C │ 35 │ ↑ ↑ ↑ +└──────────────────┘ Int32Array StringArray Int32Array +(read entire rows) (process entire columns at once) +``` + +### Why This Matters + +- **Vectorized Execution**: Process entire columns at once using SIMD instructions +- **Better Compression**: Similar values stored together compress more efficiently +- **Cache Efficiency**: Scanning specific columns doesn't load unnecessary data +- **Zero-Copy Data Sharing**: Systems can share Arrow data without conversion overhead + +DataFusion, DuckDB, Polars, and Pandas all speak Arrow natively—they can exchange data without expensive serialization/deserialization steps. + +## What is a RecordBatch? (And Why Batch?) + +A **[`RecordBatch`]** represents a horizontal slice of a table—a collection of equal-length columnar arrays sharing the same schema. + +### Why Not Process Entire Tables? + +- **Memory Constraints**: A billion-row table might not fit in RAM +- **Pipeline Processing**: Start producing results before reading all data +- **Parallel Execution**: Different threads can process different batches + +### Why Not Process Single Rows? + +- **Lost Vectorization**: Can't use SIMD instructions on single values +- **Poor Cache Utilization**: Jumping between rows defeats CPU cache optimization +- **High Overhead**: Managing individual rows has significant bookkeeping costs Review Comment: You're right - I essentially repeated the same points. My intention was to show the progression from "too big" (entire table) → "too small" (single rows) → "just right" (batches), but I see it reads as repetitive. I'll consolidate into a single "Why batches are the sweet spot" section that covers both extremes concisely without redundancy. Do you have suggestions, I might not see ? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
