1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912
// Copyright 2024 RisingWave Labs
//
// Licensed 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.
#![doc = include_str!("../README.md")]
// Notice for developers:
// This library uses the sub-interpreter and per-interpreter GIL introduced in Python 3.12
// to support concurrent execution of different functions in multiple threads.
// However, pyo3 has not yet safely supported sub-interpreter. We use the raw FFI API of pyo3 to implement sub-interpreter.
// Therefore, special attention is needed:
// **All PyObject created in a sub-interpreter must be destroyed in the same sub-interpreter.**
// Otherwise, it will cause a crash the next time Python is called.
// Special attention is needed for PyErr in PyResult.
// Remember to convert `PyErr` using the `pyerr_to_anyhow` function before passing it out of the sub-interpreter.
use self::interpreter::SubInterpreter;
pub use self::into_field::IntoField;
use anyhow::{bail, Context, Result};
use arrow_array::builder::{ArrayBuilder, Int32Builder, StringBuilder};
use arrow_array::{Array, ArrayRef, BooleanArray, RecordBatch};
use arrow_schema::{DataType, Field, FieldRef, Schema, SchemaRef};
use pyo3::types::{PyAnyMethods, PyIterator, PyModule, PyTuple};
use pyo3::{Py, PyObject};
use std::collections::HashMap;
use std::fmt::Debug;
use std::sync::Arc;
// #[cfg(Py_3_12)]
mod interpreter;
mod into_field;
mod pyarrow;
/// A runtime to execute user defined functions in Python.
///
/// # Usages
///
/// - Create a new runtime with [`Runtime::new`] or [`Runtime::builder`].
/// - For scalar functions, use [`add_function`] and [`call`].
/// - For table functions, use [`add_function`] and [`call_table_function`].
/// - For aggregate functions, create the function with [`add_aggregate`], and then
/// - create a new state with [`create_state`],
/// - update the state with [`accumulate`] or [`accumulate_or_retract`],
/// - merge states with [`merge`],
/// - finally get the result with [`finish`].
///
/// Click on each function to see the example.
///
/// # Parallelism
///
/// As we know, Python has a Global Interpreter Lock (GIL) that prevents multiple threads from executing Python code simultaneously.
/// To work around this limitation, each runtime creates a sub-interpreter with its own GIL. This feature requires Python 3.12 or later.
///
/// [`add_function`]: Runtime::add_function
/// [`add_aggregate`]: Runtime::add_aggregate
/// [`call`]: Runtime::call
/// [`call_table_function`]: Runtime::call_table_function
/// [`create_state`]: Runtime::create_state
/// [`accumulate`]: Runtime::accumulate
/// [`accumulate_or_retract`]: Runtime::accumulate_or_retract
/// [`merge`]: Runtime::merge
/// [`finish`]: Runtime::finish
pub struct Runtime {
interpreter: SubInterpreter,
functions: HashMap<String, Function>,
aggregates: HashMap<String, Aggregate>,
converter: pyarrow::Converter,
}
impl Debug for Runtime {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("Runtime")
.field("functions", &self.functions.keys())
.field("aggregates", &self.aggregates.keys())
.finish()
}
}
/// A user defined function.
struct Function {
function: PyObject,
return_field: FieldRef,
mode: CallMode,
}
/// A user defined aggregate function.
struct Aggregate {
state_field: FieldRef,
output_field: FieldRef,
mode: CallMode,
create_state: PyObject,
accumulate: PyObject,
retract: Option<PyObject>,
finish: Option<PyObject>,
merge: Option<PyObject>,
}
/// A builder for `Runtime`.
#[derive(Default, Debug)]
pub struct Builder {
sandboxed: bool,
removed_symbols: Vec<String>,
}
impl Builder {
/// Set whether the runtime is sandboxed.
///
/// When sandboxed, only a limited set of modules can be imported, and some built-in functions are disabled.
/// This is useful for running untrusted code.
///
/// Allowed modules: `json`, `decimal`, `re`, `math`, `datetime`, `time`.
///
/// Disallowed builtins: `breakpoint`, `exit`, `eval`, `help`, `input`, `open`, `print`.
///
/// The default is `false`.
pub fn sandboxed(mut self, sandboxed: bool) -> Self {
self.sandboxed = sandboxed;
self.remove_symbol("__builtins__.breakpoint")
.remove_symbol("__builtins__.exit")
.remove_symbol("__builtins__.eval")
.remove_symbol("__builtins__.help")
.remove_symbol("__builtins__.input")
.remove_symbol("__builtins__.open")
.remove_symbol("__builtins__.print")
}
/// Remove a symbol from builtins.
///
/// # Examples
///
/// ```
/// # use arrow_udf_python::Runtime;
/// let builder = Runtime::builder().remove_symbol("__builtins__.eval");
/// ```
pub fn remove_symbol(mut self, symbol: &str) -> Self {
self.removed_symbols.push(symbol.to_string());
self
}
/// Build the `Runtime`.
pub fn build(self) -> Result<Runtime> {
let interpreter = SubInterpreter::new()?;
interpreter.run(
r#"
# internal use for json types
import json
import pickle
import decimal
# an internal class used for struct input arguments
class Struct:
pass
"#,
)?;
if self.sandboxed {
let mut script = r#"
# limit the modules that can be imported
original_import = __builtins__.__import__
def limited_import(name, globals=None, locals=None, fromlist=(), level=0):
# FIXME: 'sys' should not be allowed, but it is required by 'decimal'
# FIXME: 'time.sleep' should not be allowed, but 'time' is required by 'datetime'
allowlist = (
'json',
'decimal',
're',
'math',
'datetime',
'time',
'operator',
'numbers',
'abc',
'sys',
'contextvars',
'_io',
'_contextvars',
'_pydecimal',
'_pydatetime',
)
if level == 0 and name in allowlist:
return original_import(name, globals, locals, fromlist, level)
raise ImportError(f'import {name} is not allowed')
__builtins__.__import__ = limited_import
del limited_import
"#
.to_string();
for symbol in self.removed_symbols {
script.push_str(&format!("del {}\n", symbol));
}
interpreter.run(&script)?;
}
Ok(Runtime {
interpreter,
functions: HashMap::new(),
aggregates: HashMap::new(),
converter: pyarrow::Converter::new(),
})
}
}
impl Runtime {
/// Create a new `Runtime`.
pub fn new() -> Result<Self> {
Builder::default().build()
}
/// Return a new builder for `Runtime`.
pub fn builder() -> Builder {
Builder::default()
}
/// Add a new scalar function or table function.
///
/// # Arguments
///
/// - `name`: The name of the function.
/// - `return_type`: The data type of the return value.
/// - `mode`: Whether the function will be called when some of its arguments are null.
/// - `code`: The Python code of the function.
///
/// The code should define a function with the same name as the function.
/// The function should return a value for scalar functions, or yield values for table functions.
///
/// # Example
///
/// ```
/// # use arrow_udf_python::{Runtime, CallMode};
/// # use arrow_schema::DataType;
/// let mut runtime = Runtime::new().unwrap();
/// // add a scalar function
/// runtime
/// .add_function(
/// "gcd",
/// DataType::Int32,
/// CallMode::ReturnNullOnNullInput,
/// r#"
/// def gcd(a: int, b: int) -> int:
/// while b:
/// a, b = b, a % b
/// return a
/// "#,
/// )
/// .unwrap();
/// // add a table function
/// runtime
/// .add_function(
/// "series",
/// DataType::Int32,
/// CallMode::ReturnNullOnNullInput,
/// r#"
/// def series(n: int):
/// for i in range(n):
/// yield i
/// "#,
/// )
/// .unwrap();
/// ```
pub fn add_function(
&mut self,
name: &str,
return_type: impl IntoField,
mode: CallMode,
code: &str,
) -> Result<()> {
self.add_function_with_handler(name, return_type, mode, code, name)
}
/// Add a new scalar function or table function with custom handler name.
///
/// # Arguments
///
/// - `handler`: The name of function in Python code to be called.
/// - others: Same as [`add_function`].
///
/// [`add_function`]: Runtime::add_function
pub fn add_function_with_handler(
&mut self,
name: &str,
return_type: impl IntoField,
mode: CallMode,
code: &str,
handler: &str,
) -> Result<()> {
let function = self.interpreter.with_gil(|py| {
Ok(PyModule::from_code_bound(py, code, name, name)?
.getattr(handler)?
.into())
})?;
let function = Function {
function,
return_field: return_type.into_field(name).into(),
mode,
};
self.functions.insert(name.to_string(), function);
Ok(())
}
/// Add a new aggregate function from Python code.
///
/// # Arguments
///
/// - `name`: The name of the function.
/// - `state_type`: The data type of the internal state.
/// - `output_type`: The data type of the aggregate value.
/// - `mode`: Whether the function will be called when some of its arguments are null.
/// - `code`: The Python code of the aggregate function.
///
/// The code should define at least two functions:
///
/// - `create_state() -> state`: Create a new state object.
/// - `accumulate(state, *args) -> state`: Accumulate a new value into the state, returning the updated state.
///
/// optionally, the code can define:
///
/// - `finish(state) -> value`: Get the result of the aggregate function.
/// If not defined, the state is returned as the result.
/// In this case, `output_type` must be the same as `state_type`.
/// - `retract(state, *args) -> state`: Retract a value from the state, returning the updated state.
/// - `merge(state, state) -> state`: Merge two states, returning the merged state.
///
/// # Example
///
/// ```
/// # use arrow_udf_python::{Runtime, CallMode};
/// # use arrow_schema::DataType;
/// let mut runtime = Runtime::new().unwrap();
/// runtime
/// .add_aggregate(
/// "sum",
/// DataType::Int32, // state_type
/// DataType::Int32, // output_type
/// CallMode::ReturnNullOnNullInput,
/// r#"
/// def create_state():
/// return 0
///
/// def accumulate(state, value):
/// return state + value
///
/// def retract(state, value):
/// return state - value
///
/// def merge(state1, state2):
/// return state1 + state2
/// "#,
/// )
/// .unwrap();
/// ```
pub fn add_aggregate(
&mut self,
name: &str,
state_type: impl IntoField,
output_type: impl IntoField,
mode: CallMode,
code: &str,
) -> Result<()> {
let aggregate = self.interpreter.with_gil(|py| {
let module = PyModule::from_code_bound(py, code, name, name)?;
Ok(Aggregate {
state_field: state_type.into_field(name).into(),
output_field: output_type.into_field(name).into(),
mode,
create_state: module.getattr("create_state")?.into(),
accumulate: module.getattr("accumulate")?.into(),
retract: module.getattr("retract").ok().map(|f| f.into()),
finish: module.getattr("finish").ok().map(|f| f.into()),
merge: module.getattr("merge").ok().map(|f| f.into()),
})
})?;
if aggregate.finish.is_none() && aggregate.state_field != aggregate.output_field {
bail!("`output_type` must be the same as `state_type` when `finish` is not defined");
}
self.aggregates.insert(name.to_string(), aggregate);
Ok(())
}
/// Remove a scalar or table function.
pub fn del_function(&mut self, name: &str) -> Result<()> {
let function = self.functions.remove(name).context("function not found")?;
_ = self.interpreter.with_gil(|_| {
drop(function);
Ok(())
});
Ok(())
}
/// Remove an aggregate function.
pub fn del_aggregate(&mut self, name: &str) -> Result<()> {
let aggregate = self.functions.remove(name).context("function not found")?;
_ = self.interpreter.with_gil(|_| {
drop(aggregate);
Ok(())
});
Ok(())
}
/// Call a scalar function.
///
/// # Example
///
/// ```
#[doc = include_str!("doc_create_function.txt")]
/// // suppose we have created a scalar function `gcd`
/// // see the example in `add_function`
///
/// let schema = Schema::new(vec![
/// Field::new("x", DataType::Int32, true),
/// Field::new("y", DataType::Int32, true),
/// ]);
/// let arg0 = Int32Array::from(vec![Some(25), None]);
/// let arg1 = Int32Array::from(vec![Some(15), None]);
/// let input = RecordBatch::try_new(Arc::new(schema), vec![Arc::new(arg0), Arc::new(arg1)]).unwrap();
///
/// let output = runtime.call("gcd", &input).unwrap();
/// assert_eq!(&**output.column(0), &Int32Array::from(vec![Some(5), None]));
/// ```
pub fn call(&self, name: &str, input: &RecordBatch) -> Result<RecordBatch> {
let function = self.functions.get(name).context("function not found")?;
// convert each row to python objects and call the function
let (output, error) = self.interpreter.with_gil(|py| {
let mut results = Vec::with_capacity(input.num_rows());
let mut errors = vec![];
let mut row = Vec::with_capacity(input.num_columns());
for i in 0..input.num_rows() {
if function.mode == CallMode::ReturnNullOnNullInput
&& input.columns().iter().any(|column| column.is_null(i))
{
results.push(py.None());
continue;
}
row.clear();
for (column, field) in input.columns().iter().zip(input.schema().fields()) {
let pyobj = self.converter.get_pyobject(py, field, column, i)?;
row.push(pyobj);
}
let args = PyTuple::new_bound(py, row.drain(..));
match function.function.call1(py, args) {
Ok(result) => results.push(result),
Err(e) => {
results.push(py.None());
errors.push((i, e.to_string()));
}
}
}
let output = self
.converter
.build_array(&function.return_field, py, &results)?;
let error = build_error_array(input.num_rows(), errors);
Ok((output, error))
})?;
if let Some(error) = error {
let schema = Schema::new(vec![
function.return_field.clone(),
Field::new("error", DataType::Utf8, true).into(),
]);
Ok(RecordBatch::try_new(Arc::new(schema), vec![output, error])?)
} else {
let schema = Schema::new(vec![function.return_field.clone()]);
Ok(RecordBatch::try_new(Arc::new(schema), vec![output])?)
}
}
/// Call a table function.
///
/// # Example
///
/// ```
#[doc = include_str!("doc_create_function.txt")]
/// // suppose we have created a table function `series`
/// // see the example in `add_function`
///
/// let schema = Schema::new(vec![Field::new("x", DataType::Int32, true)]);
/// let arg0 = Int32Array::from(vec![Some(1), None, Some(3)]);
/// let input = RecordBatch::try_new(Arc::new(schema), vec![Arc::new(arg0)]).unwrap();
///
/// let mut outputs = runtime.call_table_function("series", &input, 10).unwrap();
/// let output = outputs.next().unwrap().unwrap();
/// let pretty = arrow_cast::pretty::pretty_format_batches(&[output]).unwrap().to_string();
/// assert_eq!(pretty, r#"
/// +-----+--------+
/// | row | series |
/// +-----+--------+
/// | 0 | 0 |
/// | 2 | 0 |
/// | 2 | 1 |
/// | 2 | 2 |
/// +-----+--------+"#.trim());
/// ```
pub fn call_table_function<'a>(
&'a self,
name: &'a str,
input: &'a RecordBatch,
chunk_size: usize,
) -> Result<RecordBatchIter<'a>> {
assert!(chunk_size > 0);
let function = self.functions.get(name).context("function not found")?;
// initial state
Ok(RecordBatchIter {
interpreter: &self.interpreter,
input,
function,
schema: Arc::new(Schema::new(vec![
Field::new("row", DataType::Int32, true).into(),
function.return_field.clone(),
])),
chunk_size,
row: 0,
generator: None,
converter: &self.converter,
})
}
/// Create a new state for an aggregate function.
///
/// # Example
/// ```
#[doc = include_str!("doc_create_aggregate.txt")]
/// let state = runtime.create_state("sum").unwrap();
/// assert_eq!(&*state, &Int32Array::from(vec![0]));
/// ```
pub fn create_state(&self, name: &str) -> Result<ArrayRef> {
let aggregate = self.aggregates.get(name).context("function not found")?;
let state = self.interpreter.with_gil(|py| {
let state = aggregate.create_state.call0(py)?;
let state = self
.converter
.build_array(&aggregate.state_field, py, &[state])?;
Ok(state)
})?;
Ok(state)
}
/// Call accumulate of an aggregate function.
///
/// # Example
/// ```
#[doc = include_str!("doc_create_aggregate.txt")]
/// let state = runtime.create_state("sum").unwrap();
///
/// let schema = Schema::new(vec![Field::new("value", DataType::Int32, true)]);
/// let arg0 = Int32Array::from(vec![Some(1), None, Some(3), Some(5)]);
/// let input = RecordBatch::try_new(Arc::new(schema), vec![Arc::new(arg0)]).unwrap();
///
/// let state = runtime.accumulate("sum", &state, &input).unwrap();
/// assert_eq!(&*state, &Int32Array::from(vec![9]));
/// ```
pub fn accumulate(
&self,
name: &str,
state: &dyn Array,
input: &RecordBatch,
) -> Result<ArrayRef> {
let aggregate = self.aggregates.get(name).context("function not found")?;
// convert each row to python objects and call the accumulate function
let new_state = self.interpreter.with_gil(|py| {
let mut state = self
.converter
.get_pyobject(py, &aggregate.state_field, state, 0)?;
let mut row = Vec::with_capacity(1 + input.num_columns());
for i in 0..input.num_rows() {
if aggregate.mode == CallMode::ReturnNullOnNullInput
&& input.columns().iter().any(|column| column.is_null(i))
{
continue;
}
row.clear();
row.push(state.clone());
for (column, field) in input.columns().iter().zip(input.schema().fields()) {
let pyobj = self.converter.get_pyobject(py, field, column, i)?;
row.push(pyobj);
}
let args = PyTuple::new_bound(py, row.drain(..));
state = aggregate.accumulate.call1(py, args)?;
}
let output = self
.converter
.build_array(&aggregate.state_field, py, &[state])?;
Ok(output)
})?;
Ok(new_state)
}
/// Call accumulate or retract of an aggregate function.
///
/// The `ops` is a boolean array that indicates whether to accumulate or retract each row.
/// `false` for accumulate and `true` for retract.
///
/// # Example
/// ```
#[doc = include_str!("doc_create_aggregate.txt")]
/// let state = runtime.create_state("sum").unwrap();
///
/// let schema = Schema::new(vec![Field::new("value", DataType::Int32, true)]);
/// let arg0 = Int32Array::from(vec![Some(1), None, Some(3), Some(5)]);
/// let ops = BooleanArray::from(vec![false, false, true, false]);
/// let input = RecordBatch::try_new(Arc::new(schema), vec![Arc::new(arg0)]).unwrap();
///
/// let state = runtime.accumulate_or_retract("sum", &state, &ops, &input).unwrap();
/// assert_eq!(&*state, &Int32Array::from(vec![3]));
/// ```
pub fn accumulate_or_retract(
&self,
name: &str,
state: &dyn Array,
ops: &BooleanArray,
input: &RecordBatch,
) -> Result<ArrayRef> {
let aggregate = self.aggregates.get(name).context("function not found")?;
let retract = aggregate
.retract
.as_ref()
.context("function does not support retraction")?;
// convert each row to python objects and call the accumulate function
let new_state = self.interpreter.with_gil(|py| {
let mut state = self
.converter
.get_pyobject(py, &aggregate.state_field, state, 0)?;
let mut row = Vec::with_capacity(1 + input.num_columns());
for i in 0..input.num_rows() {
if aggregate.mode == CallMode::ReturnNullOnNullInput
&& input.columns().iter().any(|column| column.is_null(i))
{
continue;
}
row.clear();
row.push(state.clone());
for (column, field) in input.columns().iter().zip(input.schema().fields()) {
let pyobj = self.converter.get_pyobject(py, field, column, i)?;
row.push(pyobj);
}
let args = PyTuple::new_bound(py, row.drain(..));
let func = if ops.is_valid(i) && ops.value(i) {
retract
} else {
&aggregate.accumulate
};
state = func.call1(py, args)?;
}
let output = self
.converter
.build_array(&aggregate.state_field, py, &[state])?;
Ok(output)
})?;
Ok(new_state)
}
/// Merge states of an aggregate function.
///
/// # Example
/// ```
#[doc = include_str!("doc_create_aggregate.txt")]
/// let states = Int32Array::from(vec![Some(1), None, Some(3), Some(5)]);
///
/// let state = runtime.merge("sum", &states).unwrap();
/// assert_eq!(&*state, &Int32Array::from(vec![9]));
/// ```
pub fn merge(&self, name: &str, states: &dyn Array) -> Result<ArrayRef> {
let aggregate = self.aggregates.get(name).context("function not found")?;
let merge = aggregate.merge.as_ref().context("merge not found")?;
let output = self.interpreter.with_gil(|py| {
let mut state = self
.converter
.get_pyobject(py, &aggregate.state_field, states, 0)?;
for i in 1..states.len() {
if aggregate.mode == CallMode::ReturnNullOnNullInput && states.is_null(i) {
continue;
}
let state2 = self
.converter
.get_pyobject(py, &aggregate.state_field, states, i)?;
let args = PyTuple::new_bound(py, [state, state2]);
state = merge.call1(py, args)?;
}
let output = self
.converter
.build_array(&aggregate.state_field, py, &[state])?;
Ok(output)
})?;
Ok(output)
}
/// Get the result of an aggregate function.
///
/// If the `finish` function is not defined, the state is returned as the result.
///
/// # Example
/// ```
#[doc = include_str!("doc_create_aggregate.txt")]
/// let states: ArrayRef = Arc::new(Int32Array::from(vec![Some(1), None, Some(3), Some(5)]));
///
/// let outputs = runtime.finish("sum", &states).unwrap();
/// assert_eq!(&outputs, &states);
/// ```
pub fn finish(&self, name: &str, states: &ArrayRef) -> Result<ArrayRef> {
let aggregate = self.aggregates.get(name).context("function not found")?;
let Some(finish) = &aggregate.finish else {
return Ok(states.clone());
};
let output = self.interpreter.with_gil(|py| {
let mut results = Vec::with_capacity(states.len());
for i in 0..states.len() {
if aggregate.mode == CallMode::ReturnNullOnNullInput && states.is_null(i) {
results.push(py.None());
continue;
}
let state = self
.converter
.get_pyobject(py, &aggregate.state_field, states, i)?;
let args = PyTuple::new_bound(py, [state]);
let result = finish.call1(py, args)?;
results.push(result);
}
let output = self
.converter
.build_array(&aggregate.output_field, py, &results)?;
Ok(output)
})?;
Ok(output)
}
}
/// An iterator over the result of a table function.
pub struct RecordBatchIter<'a> {
interpreter: &'a SubInterpreter,
input: &'a RecordBatch,
function: &'a Function,
schema: SchemaRef,
chunk_size: usize,
// mutable states
/// Current row index.
row: usize,
/// Generator of the current row.
generator: Option<Py<PyIterator>>,
converter: &'a pyarrow::Converter,
}
impl RecordBatchIter<'_> {
/// Get the schema of the output.
pub fn schema(&self) -> &Schema {
&self.schema
}
fn next(&mut self) -> Result<Option<RecordBatch>> {
if self.row == self.input.num_rows() {
return Ok(None);
}
let batch = self.interpreter.with_gil(|py| {
let mut indexes = Int32Builder::with_capacity(self.chunk_size);
let mut results = Vec::with_capacity(self.input.num_rows());
let mut errors = vec![];
let mut row = Vec::with_capacity(self.input.num_columns());
while self.row < self.input.num_rows() && results.len() < self.chunk_size {
let generator = if let Some(g) = self.generator.as_ref() {
g
} else {
// call the table function to get a generator
if self.function.mode == CallMode::ReturnNullOnNullInput
&& (self.input.columns().iter()).any(|column| column.is_null(self.row))
{
self.row += 1;
continue;
}
row.clear();
for (column, field) in
(self.input.columns().iter()).zip(self.input.schema().fields())
{
let val = self.converter.get_pyobject(py, field, column, self.row)?;
row.push(val);
}
let args = PyTuple::new_bound(py, row.drain(..));
match self.function.function.bind(py).call1(args) {
Ok(result) => {
let iter = result.iter()?.into();
self.generator.insert(iter)
}
Err(e) => {
// append a row with null value and error message
indexes.append_value(self.row as i32);
results.push(py.None());
errors.push((indexes.len(), e.to_string()));
self.row += 1;
continue;
}
}
};
match generator.bind(py).clone().next() {
Some(Ok(value)) => {
indexes.append_value(self.row as i32);
results.push(value.into());
}
Some(Err(e)) => {
indexes.append_value(self.row as i32);
results.push(py.None());
errors.push((indexes.len(), e.to_string()));
self.row += 1;
self.generator = None;
}
None => {
self.row += 1;
self.generator = None;
}
}
}
if results.is_empty() {
return Ok(None);
}
let indexes = Arc::new(indexes.finish());
let output = self
.converter
.build_array(&self.function.return_field, py, &results)
.context("failed to build arrow array from return values")?;
let error = build_error_array(indexes.len(), errors);
if let Some(error) = error {
Ok(Some(
RecordBatch::try_new(
Arc::new(append_error_to_schema(&self.schema)),
vec![indexes, output, error],
)
.unwrap(),
))
} else {
Ok(Some(
RecordBatch::try_new(self.schema.clone(), vec![indexes, output]).unwrap(),
))
}
})?;
Ok(batch)
}
}
impl Iterator for RecordBatchIter<'_> {
type Item = Result<RecordBatch>;
fn next(&mut self) -> Option<Self::Item> {
self.next().transpose()
}
}
impl Drop for RecordBatchIter<'_> {
fn drop(&mut self) {
if let Some(generator) = self.generator.take() {
_ = self.interpreter.with_gil(|_| {
drop(generator);
Ok(())
});
}
}
}
/// Whether the function will be called when some of its arguments are null.
#[derive(Debug, Default, PartialEq, Eq, PartialOrd, Ord)]
pub enum CallMode {
/// The function will be called normally when some of its arguments are null.
/// It is then the function author's responsibility to check for null values if necessary and respond appropriately.
#[default]
CalledOnNullInput,
/// The function always returns null whenever any of its arguments are null.
/// If this parameter is specified, the function is not executed when there are null arguments;
/// instead a null result is assumed automatically.
ReturnNullOnNullInput,
}
impl Drop for Runtime {
fn drop(&mut self) {
// `PyObject` must be dropped inside the interpreter
_ = self.interpreter.with_gil(|_| {
self.functions.clear();
self.aggregates.clear();
Ok(())
});
}
}
fn build_error_array(num_rows: usize, errors: Vec<(usize, String)>) -> Option<ArrayRef> {
if errors.is_empty() {
return None;
}
let data_capacity = errors.iter().map(|(i, _)| i).sum();
let mut builder = StringBuilder::with_capacity(num_rows, data_capacity);
for (i, msg) in errors {
while builder.len() + 1 < i {
builder.append_null();
}
builder.append_value(&msg);
}
while builder.len() < num_rows {
builder.append_null();
}
Some(Arc::new(builder.finish()))
}
/// Append an error field to the schema.
fn append_error_to_schema(schema: &Schema) -> Schema {
let mut fields = schema.fields().to_vec();
fields.push(Arc::new(Field::new("error", DataType::Utf8, true)));
Schema::new(fields)
}