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

polars_io/ipc/
ipc_file.rs
