[ARROW-11745] [C++] Improve configurability of random data generation

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.0.0
• Fix Version/s: 4.0.0
• Component/s: C++
• Labels: pull-request-available
• External issue URL: https://github.com/apache/arrow/issues/27601
• Assignee: Ben Kietzman
• Reporter: Ben Kietzman
• Created: 23/Feb/21 15:46
• Updated: 11/Jan/23 08:21
• Resolved: 19/Mar/21 15:53

Description

arrow::random::RandomArrayGenerator is useful for stress testing and benchmarking. Arrays of primitives can be generated with little boilerplate, however it is cumbersome to specify creation of nested arrays or record batches which are necessary for testing $n column operations such as group_by.

My ideal API for random generation takes only a FieldVector, a number of rows, and a seed as arguments. Other options (such as minimum, maximum, unique count, null probability, etc) are specified using field metadata so that they can be provided uniformly or granularly as necessary for a given test case:

auto random_batch = Generate({
  field("i32", int32()), // i32 may take any value between INT_MAX and INT_MIN
                         // and will be null with default probability 0.01
  field("f32", float32(), false), // f32 will be entirely valid
  field("probability", float64(), true, key_value_metadata({
    // custom random generation properties:
    {"min", "0.0"},
    {"max", "1.0"},
    {"null_probability", "0.0001"},
  }),
  field("list_i32", list(
    field("item", int32(), true, key_value_metadata({
      // custom random generation properties can also be specified for nested fields:
      {"min", "0"},
      {"max", "1"},
    })
  )),
}, num_rows, 0xdeadbeef);

Issue Links

• supercedes: ARROW-9149 [C++] Improve configurability of RandomArrayGenerator::ArrayOf
• supercedes: ARROW-11674 [C++] Add tests for RandomArrayGenerator
• links to: GitHub Pull Request #9715