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

polars_io/ipc/
ipc_file.rs
