polars_io/ipc/
ipc_file.rs

1//! # (De)serializing Arrows IPC format.
2//!
3//! Arrow IPC is a [binary format](https://arrow.apache.org/docs/python/ipc.html).
4//! It is the recommended way to serialize and deserialize Polars DataFrames as this is most true
5//! to the data schema.
6//!
7//! ## Example
8//!
9//! ```rust
10//! use polars_core::prelude::*;
11//! use polars_io::prelude::*;
12//! use std::io::Cursor;
13//!
14//!
15//! let s0 = Column::new("days".into(), &[0, 1, 2, 3, 4]);
16//! let s1 = Column::new("temp".into(), &[22.1, 19.9, 7., 2., 3.]);
17//! let mut df = DataFrame::new_infer_height(vec![s0, s1]).unwrap();
18//!
19//! // Create an in memory file handler.
20//! // Vec<u8>: Read + Write
21//! // Cursor<T>: Seek
22//!
23//! let mut buf: Cursor<Vec<u8>> = Cursor::new(Vec::new());
24//!
25//! // write to the in memory buffer
26//! IpcWriter::new(&mut buf).finish(&mut df).expect("ipc writer");
27//!
28//! // reset the buffers index after writing to the beginning of the buffer
29//! buf.set_position(0);
30//!
31//! // read the buffer into a DataFrame
32//! let df_read = IpcReader::new(buf).finish().unwrap();
33//! assert!(df.equals(&df_read));
34//! ```
35use std::io::{Read, Seek};
36use std::path::PathBuf;
37
38use arrow::datatypes::{ArrowSchemaRef, Metadata};
39use arrow::io::ipc::read::{self, get_row_count};
40use arrow::record_batch::RecordBatch;
41use polars_core::prelude::*;
42use polars_utils::pl_str::PlRefStr;
43#[cfg(feature = "serde")]
44use serde::{Deserialize, Serialize};
45
46use crate::RowIndex;
47use crate::hive::materialize_hive_partitions;
48use crate::mmap::MmapBytesReader;
49use crate::predicates::PhysicalIoExpr;
50use crate::prelude::*;
51use crate::shared::{ArrowReader, finish_reader};
52
53#[derive(Clone, Debug, PartialEq, Hash)]
54#[cfg_attr(feature = "serde", derive(Serialize, Deserialize))]
55#[cfg_attr(feature = "dsl-schema", derive(schemars::JsonSchema))]
56pub struct IpcScanOptions;
57
58#[expect(clippy::derivable_impls)]
59impl Default for IpcScanOptions {
60    fn default() -> Self {
61        Self {}
62    }
63}
64
65/// Read Arrows IPC format into a DataFrame
66///
67/// # Example
68/// ```
69/// use polars_core::prelude::*;
70/// use std::fs::File;
71/// use polars_io::ipc::IpcReader;
72/// use polars_io::SerReader;
73///
74/// fn example() -> PolarsResult<DataFrame> {
75///     let file = File::open("file.ipc").expect("file not found");
76///
77///     IpcReader::new(file)
78///         .finish()
79/// }
80/// ```
81#[must_use]
82pub struct IpcReader<R: MmapBytesReader> {
83    /// File or Stream object
84    pub(super) reader: R,
85    /// Aggregates chunks afterwards to a single chunk.
86    rechunk: bool,
87    pub(super) n_rows: Option<usize>,
88    pub(super) projection: Option<Vec<usize>>,
89    pub(crate) columns: Option<Vec<String>>,
90    hive_partition_columns: Option<Vec<Series>>,
91    include_file_path: Option<(PlSmallStr, PlRefStr)>,
92    pub(super) row_index: Option<RowIndex>,
93    // Stores the as key semaphore to make sure we don't write to the memory mapped file.
94    pub(super) memory_map: Option<PathBuf>,
95    metadata: Option<read::FileMetadata>,
96    schema: Option<ArrowSchemaRef>,
97}
98
99fn check_mmap_err(err: PolarsError) -> PolarsResult<()> {
100    if let PolarsError::ComputeError(s) = &err {
101        if s.as_ref() == "memory_map can only be done on uncompressed IPC files" {
102            eprintln!(
103                "Could not memory_map compressed IPC file, defaulting to normal read. \
104                Toggle off 'memory_map' to silence this warning."
105            );
106            return Ok(());
107        }
108    }
109    Err(err)
110}
111
112impl<R: MmapBytesReader> IpcReader<R> {
113    fn get_metadata(&mut self) -> PolarsResult<&read::FileMetadata> {
114        if self.metadata.is_none() {
115            let metadata = read::read_file_metadata(&mut self.reader)?;
116            self.schema = Some(metadata.schema.clone());
117            self.metadata = Some(metadata);
118        }
119        Ok(self.metadata.as_ref().unwrap())
120    }
121
122    /// Get arrow schema of the Ipc File.
123    pub fn schema(&mut self) -> PolarsResult<ArrowSchemaRef> {
124        self.get_metadata()?;
125        Ok(self.schema.as_ref().unwrap().clone())
126    }
127
128    /// Get schema-level custom metadata of the Ipc file
129    pub fn custom_metadata(&mut self) -> PolarsResult<Option<Arc<Metadata>>> {
130        self.get_metadata()?;
131        Ok(self
132            .metadata
133            .as_ref()
134            .and_then(|meta| meta.custom_schema_metadata.clone()))
135    }
136
137    /// Stop reading when `n` rows are read.
138    pub fn with_n_rows(mut self, num_rows: Option<usize>) -> Self {
139        self.n_rows = num_rows;
140        self
141    }
142
143    /// Columns to select/ project
144    pub fn with_columns(mut self, columns: Option<Vec<String>>) -> Self {
145        self.columns = columns;
146        self
147    }
148
149    pub fn with_hive_partition_columns(mut self, columns: Option<Vec<Series>>) -> Self {
150        self.hive_partition_columns = columns;
151        self
152    }
153
154    pub fn with_include_file_path(
155        mut self,
156        include_file_path: Option<(PlSmallStr, PlRefStr)>,
157    ) -> Self {
158        self.include_file_path = include_file_path;
159        self
160    }
161
162    /// Add a row index column.
163    pub fn with_row_index(mut self, row_index: Option<RowIndex>) -> Self {
164        self.row_index = row_index;
165        self
166    }
167
168    /// Set the reader's column projection. This counts from 0, meaning that
169    /// `vec![0, 4]` would select the 1st and 5th column.
170    pub fn with_projection(mut self, projection: Option<Vec<usize>>) -> Self {
171        self.projection = projection;
172        self
173    }
174
175    /// Set if the file is to be memory_mapped. Only works with uncompressed files.
176    /// The file name must be passed to register the memory mapped file.
177    pub fn memory_mapped(mut self, path_buf: Option<PathBuf>) -> Self {
178        self.memory_map = path_buf;
179        self
180    }
181
182    // todo! hoist to lazy crate
183    #[cfg(feature = "lazy")]
184    pub fn finish_with_scan_ops(
185        mut self,
186        predicate: Option<Arc<dyn PhysicalIoExpr>>,
187        verbose: bool,
188    ) -> PolarsResult<DataFrame> {
189        if self.memory_map.is_some() && self.reader.to_file().is_some() {
190            if verbose {
191                eprintln!("memory map ipc file")
192            }
193            match self.finish_memmapped(predicate.clone()) {
194                Ok(df) => return Ok(df),
195                Err(err) => check_mmap_err(err)?,
196            }
197        }
198        let rechunk = self.rechunk;
199        let metadata = read::read_file_metadata(&mut self.reader)?;
200
201        // NOTE: For some code paths this already happened. See
202        // https://github.com/pola-rs/polars/pull/14984#discussion_r1520125000
203        // where this was introduced.
204        if let Some(columns) = &self.columns {
205            self.projection = Some(columns_to_projection(columns, &metadata.schema)?);
206        }
207
208        let schema = if let Some(projection) = &self.projection {
209            Arc::new(apply_projection(&metadata.schema, projection))
210        } else {
211            metadata.schema.clone()
212        };
213
214        let reader = read::FileReader::new(self.reader, metadata, self.projection, self.n_rows);
215
216        finish_reader(reader, rechunk, None, predicate, &schema, self.row_index)
217    }
218}
219
220impl<R: MmapBytesReader> ArrowReader for read::FileReader<R>
221where
222    R: Read + Seek,
223{
224    fn next_record_batch(&mut self) -> PolarsResult<Option<RecordBatch>> {
225        self.next().map_or(Ok(None), |v| v.map(Some))
226    }
227}
228
229impl<R: MmapBytesReader> SerReader<R> for IpcReader<R> {
230    fn new(reader: R) -> Self {
231        IpcReader {
232            reader,
233            rechunk: true,
234            n_rows: None,
235            columns: None,
236            hive_partition_columns: None,
237            include_file_path: None,
238            projection: None,
239            row_index: None,
240            memory_map: None,
241            metadata: None,
242            schema: None,
243        }
244    }
245
246    fn set_rechunk(mut self, rechunk: bool) -> Self {
247        self.rechunk = rechunk;
248        self
249    }
250
251    fn finish(mut self) -> PolarsResult<DataFrame> {
252        let reader_schema = if let Some(ref schema) = self.schema {
253            schema.clone()
254        } else {
255            self.get_metadata()?.schema.clone()
256        };
257        let reader_schema = reader_schema.as_ref();
258
259        let hive_partition_columns = self.hive_partition_columns.take();
260        let include_file_path = self.include_file_path.take();
261
262        // In case only hive columns are projected, the df would be empty, but we need the row count
263        // of the file in order to project the correct number of rows for the hive columns.
264        let mut df = (|| {
265            if self.projection.as_ref().is_some_and(|x| x.is_empty()) {
266                let row_count = if let Some(v) = self.n_rows {
267                    v
268                } else {
269                    get_row_count(&mut self.reader)? as usize
270                };
271                let mut df = DataFrame::empty_with_height(row_count);
272
273                if let Some(ri) = &self.row_index {
274                    unsafe { df.with_row_index_mut(ri.name.clone(), Some(ri.offset)) };
275                }
276                return PolarsResult::Ok(df);
277            }
278
279            if self.memory_map.is_some() && self.reader.to_file().is_some() {
280                match self.finish_memmapped(None) {
281                    Ok(df) => {
282                        return Ok(df);
283                    },
284                    Err(err) => check_mmap_err(err)?,
285                }
286            }
287            let rechunk = self.rechunk;
288            let schema = self.get_metadata()?.schema.clone();
289
290            if let Some(columns) = &self.columns {
291                let prj = columns_to_projection(columns, schema.as_ref())?;
292                self.projection = Some(prj);
293            }
294
295            let schema = if let Some(projection) = &self.projection {
296                Arc::new(apply_projection(schema.as_ref(), projection))
297            } else {
298                schema
299            };
300
301            let metadata = self.get_metadata()?.clone();
302
303            let ipc_reader =
304                read::FileReader::new(self.reader, metadata, self.projection, self.n_rows);
305            let df = finish_reader(ipc_reader, rechunk, None, None, &schema, self.row_index)?;
306            Ok(df)
307        })()?;
308
309        if let Some(hive_cols) = hive_partition_columns {
310            materialize_hive_partitions(&mut df, reader_schema, Some(hive_cols.as_slice()));
311        };
312
313        if let Some((col, value)) = include_file_path {
314            unsafe {
315                df.push_column_unchecked(Column::new_scalar(
316                    col,
317                    Scalar::new(
318                        DataType::String,
319                        AnyValue::StringOwned(value.as_str().into()),
320                    ),
321                    df.height(),
322                ))
323            };
324        }
325
326        Ok(df)
327    }
328}
polars_io/ipc/ipc_file.rs

polars_io/ipc/
ipc_file.rs
