Skip to content

[AURON #2279] Implement native Levenshtein with threshold support for Spark 4.0+ #2280

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
lyne7-sc wants to merge 9 commits into
base: master
Choose a base branch
from
Open

[AURON #2279] Implement native Levenshtein with threshold support for Spark 4.0+ #2280

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
67fdebd
implement levenshtein with threshold parameter for Spark 4.x
lyne7-sc May 18, 2026
7963af0
Refactor spark levenshtein implementation
lyne7-sc May 19, 2026
8fcc669
Simplify levenshtein array threshold handling
lyne7-sc May 19, 2026
1b7439c
Comment out levenshtein proto enum
lyne7-sc May 19, 2026
f426f03
Clean up levenshtein planner fallback
lyne7-sc May 19, 2026
be74b5f
Clean up levenshtein planner fallback
lyne7-sc May 19, 2026
c7bdafc
Merge remote-tracking branch 'origin/fix/spark_levenshtein' into fix/…
lyne7-sc May 19, 2026
b1b35c5
Fix null handling in levenshtein function
lyne7-sc May 21, 2026
96e0c6e
lint
lyne7-sc May 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 0 additions & 2 deletions auron-spark-tests/spark35/src/test/scala/org/apache/auron/utils/AuronSparkTestSettings.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,8 +59,6 @@ class AuronSparkTestSettings extends SparkTestSettings {
enableSuite[AuronMiscFunctionsSuite]

enableSuite[AuronStringFunctionsSuite]
// Native levenshtein has a Spark 3.5+ result or schema comparison mismatch.
.exclude("string Levenshtein distance")
// Native substr does not support BinaryType inputs.
// See https://github.com/apache/auron/issues/1724
.exclude("string / binary substring function")
Expand Down
3 changes: 0 additions & 3 deletions auron-spark-tests/spark40/src/test/scala/org/apache/auron/utils/AuronSparkTestSettings.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,9 +59,6 @@ class AuronSparkTestSettings extends SparkTestSettings {
enableSuite[AuronMiscFunctionsSuite]

enableSuite[AuronStringFunctionsSuite]
// Spark 4 adds the threshold argument, but native levenshtein currently supports only
// two arguments.
.exclude("string Levenshtein distance")
// Native substr does not support BinaryType inputs.
.exclude("string / binary substring function")

Expand Down
3 changes: 0 additions & 3 deletions auron-spark-tests/spark41/src/test/scala/org/apache/auron/utils/AuronSparkTestSettings.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,9 +59,6 @@ class AuronSparkTestSettings extends SparkTestSettings {
enableSuite[AuronMiscFunctionsSuite]

enableSuite[AuronStringFunctionsSuite]
// Spark 4 adds the threshold argument, but native levenshtein currently supports only
// two arguments.
.exclude("string Levenshtein distance")
// Native substr does not support BinaryType inputs.
.exclude("string / binary substring function")

Expand Down
2 changes: 1 addition & 1 deletion native-engine/auron-planner/proto/auron.proto
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -281,7 +281,7 @@ enum ScalarFunction {
Hex=66;
Power=67;
IsNaN=69;
Levenshtein=80;
// Levenshtein=80;
FindInSet=81;
Nvl=82;
Nvl2=83;
Expand Down
3 changes: 1 addition & 2 deletions native-engine/auron-planner/src/planner.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1292,8 +1292,7 @@ impl From<protobuf::ScalarFunction> for Arc<ScalarUDF> {
ScalarFunction::Rpad => f::unicode::rpad(),
ScalarFunction::SplitPart => f::string::split_part(),
ScalarFunction::StartsWith => f::string::starts_with(),
ScalarFunction::Levenshtein => f::string::levenshtein(),

// ScalarFunction::Levenshtein => f::string::levenshtein(),
ScalarFunction::FindInSet => f::unicode::find_in_set(),
ScalarFunction::Strpos => f::unicode::strpos(),
ScalarFunction::Substr => f::unicode::substr(),
Expand Down
1 change: 1 addition & 0 deletions native-engine/datafusion-ext-functions/src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,6 +75,7 @@ pub fn create_auron_ext_function(
"Spark_StringConcatWs" => Arc::new(spark_strings::string_concat_ws),
"Spark_StringLower" => Arc::new(spark_strings::string_lower),
"Spark_StringUpper" => Arc::new(spark_strings::string_upper),
"Spark_Levenshtein" => Arc::new(spark_strings::spark_levenshtein),
"Spark_InitCap" => Arc::new(spark_initcap::string_initcap),
"Spark_Year" => Arc::new(spark_dates::spark_year),
"Spark_Month" => Arc::new(spark_dates::spark_month),
Expand Down
242 changes: 237 additions & 5 deletions native-engine/datafusion-ext-functions/src/spark_strings.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,9 @@
use std::sync::Arc;

use arrow::{
array::{Array, ArrayRef, AsArray, ListArray, ListBuilder, StringArray, StringBuilder},
array::{
Array, ArrayRef, AsArray, Int32Array, ListArray, ListBuilder, StringArray, StringBuilder,
},
datatypes::DataType,
};
use datafusion::{
Expand Down Expand Up @@ -114,6 +116,118 @@ pub fn string_split(args: &[ColumnarValue]) -> Result<ColumnarValue> {
Ok(ColumnarValue::Array(Arc::new(splitted_builder.finish())))
}

pub fn spark_levenshtein(args: &[ColumnarValue]) -> Result<ColumnarValue> {
Copy link
Copy Markdown
Contributor

@ShreyeshArangath ShreyeshArangath May 20, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Correct me if I'm wrong but, in spark_levenshtein, a null threshold is coerced to Some(0), which then returns -1 for any non-zero distance and 0 for equal strings:

  Some(ColumnarValue::Scalar(scalar)) if scalar.is_null() => Some(0),
  ...
  Some(array) if array.data_type() == &DataType::Null => Some(0),
  Some(_) => thresholds.map(|array| if array.is_valid(i) { array.value(i) } else { 0 }),

I think that doesn't match Spark. In Spark, Levenshtein.eval only null-checks left/right..a null threshold at runtime causes an NPE during v.asInstanceOf[Int]. The defensible options are (a) propagate null when threshold is null, or (b) mirror Spark and error

Copy link
Copy Markdown
Contributor Author

@lyne7-sc lyne7-sc May 21, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for catching this! You're correct. I double-confirmed this and have fixed it by propagating null. Updated tests accordingly.

if args.len() != 2 && args.len() != 3 {
df_execution_err!(
"levenshtein was called with {} arguments. It requires 2 or 3.",
args.len(),
)?;
}

if args
.iter()
.all(|arg| matches!(arg, ColumnarValue::Scalar(_)))
{
let left = match &args[0] {
ColumnarValue::Scalar(ScalarValue::Utf8(value)) => value.as_deref(),
_ => df_execution_err!("levenshtein only supports utf8 string arguments")?,
};
let right = match &args[1] {
ColumnarValue::Scalar(ScalarValue::Utf8(value)) => value.as_deref(),
_ => df_execution_err!("levenshtein only supports utf8 string arguments")?,
};
let threshold = match args.get(2) {
Some(ColumnarValue::Scalar(ScalarValue::Int32(Some(value)))) => Some(*value),
Some(ColumnarValue::Scalar(scalar)) if scalar.is_null() => {
return Ok(ColumnarValue::Scalar(ScalarValue::Int32(None)));
}
Some(_) => df_execution_err!("levenshtein threshold only supports int32")?,
None => None,
};
return Ok(ColumnarValue::Scalar(ScalarValue::Int32(
compute_levenshtein(left, right, threshold),
)));
}

let array_len = args
.iter()
.find_map(|arg| match arg {
ColumnarValue::Array(array) => Some(array.len()),
ColumnarValue::Scalar(_) => None,
})
.expect("levenshtein arguments include an array");
let left_array = args[0].clone().into_array(array_len)?;
let right_array = args[1].clone().into_array(array_len)?;
let threshold_array = args
.get(2)
.map(|threshold| threshold.clone().into_array(array_len))
.transpose()?;

let left_strings = as_string_array(&left_array)?;
let right_strings = as_string_array(&right_array)?;
let thresholds = threshold_array
.as_ref()
.filter(|array| array.data_type() != &DataType::Null)
.map(|array| as_int32_array(array))
.transpose()?;

let result = Int32Array::from_iter((0..array_len).map(|i| {
let threshold = match &threshold_array {
Some(array) if array.data_type() == &DataType::Null => return None,
Some(_) => match thresholds {
Some(arr) if arr.is_valid(i) => Some(arr.value(i)),
_ => return None,
},
None => None,
};
compute_levenshtein(
left_strings.is_valid(i).then(|| left_strings.value(i)),
right_strings.is_valid(i).then(|| right_strings.value(i)),
threshold,
)
}));
Ok(ColumnarValue::Array(Arc::new(result)))
}

fn compute_levenshtein(
left: Option<&str>,
right: Option<&str>,
threshold: Option<i32>,
) -> Option<i32> {
let left = left?;
let right = right?;
let distance = if left == right {
0
} else {
let left_chars = left.chars().collect::<Vec<_>>();
let right_chars = right.chars().collect::<Vec<_>>();
if left_chars.is_empty() {
right_chars.len() as i32
} else if right_chars.is_empty() {
left_chars.len() as i32
} else {
let mut previous = (0..=right_chars.len()).collect::<Vec<_>>();
let mut current = vec![0; right_chars.len() + 1];

for (i, left_char) in left_chars.iter().enumerate() {
current[0] = i + 1;
for (j, right_char) in right_chars.iter().enumerate() {
let substitution_cost = usize::from(left_char != right_char);
current[j + 1] = (current[j] + 1)
.min(previous[j + 1] + 1)
.min(previous[j] + substitution_cost);
}
std::mem::swap(&mut previous, &mut current);
}
previous[right_chars.len()] as i32
}
};
Some(match threshold {
Some(threshold) if distance > threshold => -1,
_ => distance,
})
}

/// concat() function compatible with spark (returns null if any param is null)
/// concat('abcde', 2, 22) = 'abcde222
/// concat('abcde', 2, NULL, 22) = NULL
Expand Down Expand Up @@ -322,19 +436,19 @@ pub fn string_concat_ws(args: &[ColumnarValue]) -> Result<ColumnarValue> {
mod test {
use std::sync::Arc;

use arrow::array::{Int32Array, ListBuilder, StringArray, StringBuilder};
use arrow::array::{Int32Array, ListBuilder, NullArray, StringArray, StringBuilder};
use datafusion::{
common::{
Result, ScalarValue,
cast::{as_list_array, as_string_array},
cast::{as_int32_array, as_list_array, as_string_array},
},
physical_plan::ColumnarValue,
};
use datafusion_ext_commons::df_execution_err;

use crate::spark_strings::{
string_concat, string_concat_ws, string_lower, string_repeat, string_space, string_split,
string_upper,
spark_levenshtein, string_concat, string_concat_ws, string_lower, string_repeat,
string_space, string_split, string_upper,
};

#[test]
Expand Down Expand Up @@ -395,6 +509,124 @@ mod test {
}
}

#[test]
fn test_spark_levenshtein_array() -> Result<()> {
let r = spark_levenshtein(&vec![
ColumnarValue::Array(Arc::new(StringArray::from_iter(vec![
Some("kitten".to_string()),
Some("frog".to_string()),
Some("千世".to_string()),
None,
]))),
ColumnarValue::Array(Arc::new(StringArray::from_iter(vec![
Some("sitting".to_string()),
Some("fog".to_string()),
Some("世界千世".to_string()),
Some("abc".to_string()),
]))),
])?;
let s = r.into_array(4)?;
assert_eq!(
as_int32_array(&s)?.into_iter().collect::<Vec<_>>(),
vec![Some(3), Some(1), Some(2), None]
);
Ok(())
}

#[test]
fn test_spark_levenshtein_threshold() -> Result<()> {
let r = spark_levenshtein(&vec![
ColumnarValue::Array(Arc::new(StringArray::from_iter(vec![
Some("kitten".to_string()),
Some("kitten".to_string()),
Some("abc".to_string()),
Some("abc".to_string()),
Some("".to_string()),
Some("abc".to_string()),
]))),
ColumnarValue::Array(Arc::new(StringArray::from_iter(vec![
Some("sitting".to_string()),
Some("sitting".to_string()),
Some("abc".to_string()),
Some("xyz".to_string()),
Some("abc".to_string()),
Some("abc".to_string()),
]))),
ColumnarValue::Array(Arc::new(Int32Array::from_iter(vec![
Some(2),
Some(3),
Some(0),
None,
Some(3),
Some(-1),
]))),
])?;
let s = r.into_array(6)?;
assert_eq!(
as_int32_array(&s)?.into_iter().collect::<Vec<_>>(),
vec![Some(-1), Some(3), Some(0), None, Some(3), Some(-1)]
);

let r = spark_levenshtein(&vec![
ColumnarValue::Array(Arc::new(StringArray::from_iter(vec![Some(
"abc".to_string(),
)]))),
ColumnarValue::Array(Arc::new(StringArray::from_iter(vec![Some(
"xyz".to_string(),
)]))),
ColumnarValue::Array(Arc::new(NullArray::new(1))),
])?;
let s = r.into_array(1)?;
assert_eq!(
as_int32_array(&s)?.into_iter().collect::<Vec<_>>(),
vec![None]
);
Ok(())
}

#[test]
fn test_spark_levenshtein_scalar() -> Result<()> {
let r = spark_levenshtein(&vec![
ColumnarValue::Scalar(ScalarValue::Utf8(Some("kitten".to_string()))),
ColumnarValue::Scalar(ScalarValue::Utf8(Some("sitting".to_string()))),
ColumnarValue::Scalar(ScalarValue::Int32(Some(2))),
])?;
match r {
ColumnarValue::Scalar(ScalarValue::Int32(Some(-1))) => {}
other => df_execution_err!("Expected Int32(-1) scalar, got: {:?}", other)?,
}

let r = spark_levenshtein(&vec![
ColumnarValue::Scalar(ScalarValue::Utf8(Some("kitten".to_string()))),
ColumnarValue::Scalar(ScalarValue::Utf8(Some("sitting".to_string()))),
ColumnarValue::Scalar(ScalarValue::Int32(Some(3))),
])?;
match r {
ColumnarValue::Scalar(ScalarValue::Int32(Some(3))) => {}
other => df_execution_err!("Expected Int32(3) scalar, got: {:?}", other)?,
}

let r = spark_levenshtein(&vec![
ColumnarValue::Scalar(ScalarValue::Utf8(Some("kitten".to_string()))),
ColumnarValue::Scalar(ScalarValue::Utf8(Some("sitting".to_string()))),
ColumnarValue::Scalar(ScalarValue::Null),
])?;
match r {
ColumnarValue::Scalar(ScalarValue::Int32(None)) => {}
other => df_execution_err!("Expected null Int32 scalar, got: {:?}", other)?,
}

let r = spark_levenshtein(&vec![
ColumnarValue::Scalar(ScalarValue::Utf8(None)),
ColumnarValue::Scalar(ScalarValue::Utf8(Some("sitting".to_string()))),
])?;
match r {
ColumnarValue::Scalar(ScalarValue::Int32(None)) => {}
other => df_execution_err!("Expected null Int32 scalar, got: {:?}", other)?,
}
Ok(())
}

#[test]
fn test_string_repeat() -> Result<()> {
// positive case
Expand Down
2 changes: 1 addition & 1 deletion spark-extension/src/main/scala/org/apache/spark/sql/auron/NativeConverters.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -975,7 +975,7 @@ object NativeConverters extends Logging {
buildTimePartExt("Spark_Quarter", child, isPruningExpr, fallback)

case e: Levenshtein =>
buildScalarFunction(pb.ScalarFunction.Levenshtein, e.children, e.dataType)
buildExtScalarFunction("Spark_Levenshtein", e.children, e.dataType)

case e: Hour if datetimeExtractEnabled =>
buildTimePartExt("Spark_Hour", e.children.head, isPruningExpr, fallback)
Expand Down
Loading