forked from apache/arrow-rs
/
encoder.rs
217 lines (178 loc) · 6.5 KB
/
encoder.rs
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
use crate::basic::Encoding;
use crate::column::writer::{compare_greater, fallback_encoding, has_dictionary_support, is_nan, update_max, update_min};
use crate::data_type::private::ParquetValueType;
use crate::data_type::DataType;
use crate::encodings::encoding::{get_encoder, DictEncoder, Encoder};
use crate::errors::{ParquetError, Result};
use crate::file::properties::WriterProperties;
use crate::schema::types::{ColumnDescPtr, ColumnDescriptor};
use crate::util::memory::ByteBufferPtr;
pub trait ColumnValues {
type T: ParquetValueType;
fn len(&self) -> usize;
fn min_max(&self, descr: &ColumnDescriptor) -> Option<(&Self::T, &Self::T)>;
}
pub struct DictionaryPage {
pub buf: ByteBufferPtr,
pub num_values: usize,
pub is_sorted: bool,
}
/// The encoded values for a data page, with optional statistics
pub struct DataPageValues<T> {
pub buf: ByteBufferPtr,
pub num_values: usize,
pub encoding: Encoding,
pub min_value: Option<T>,
pub max_value: Option<T>,
}
pub trait ColumnValueEncoder {
type T: ParquetValueType;
type Values: ColumnValues<T = Self::T> + ?Sized;
/// Create a new [`ColumnValueEncoder`]
fn try_new(descr: &ColumnDescPtr, props: &WriterProperties) -> Result<Self>
where
Self: Sized;
/// Write the corresponding values to this [`ColumnValueEncoder`]
fn write(&mut self, values: &Self::Values, offset: usize, len: usize) -> Result<()>;
/// Returns the number of buffered values
fn num_values(&self) -> usize;
/// Returns true if this encoder has a dictionary page
fn has_dictionary(&self) -> bool;
/// Returns an estimate of the dictionary page size in bytes, or `None` if no dictionary
fn estimated_dict_page_size(&self) -> Option<usize>;
/// Returns an estimate of the data page size in bytes
fn estimated_data_page_size(&self) -> usize;
/// Flush the dictionary page for this column chunk if any. Any subsequent calls to
/// [`Self::write`] will not be dictionary encoded
///
/// Note: [`Self::flush_data_page`] must be called first, as this will error if there
/// are any pending page values
fn flush_dict_page(&mut self) -> Result<Option<DictionaryPage>>;
/// Flush the next data page for this column chunk
fn flush_data_page(&mut self) -> Result<DataPageValues<Self::T>>;
}
pub struct ColumnValueEncoderImpl<T: DataType> {
encoder: Box<dyn Encoder<T>>,
dict_encoder: Option<DictEncoder<T>>,
descr: ColumnDescPtr,
num_values: usize,
min_value: Option<T::T>,
max_value: Option<T::T>,
}
impl<T: DataType> ColumnValueEncoder for ColumnValueEncoderImpl<T> {
type T = T::T;
type Values = [T::T];
fn try_new(descr: &ColumnDescPtr, props: &WriterProperties) -> Result<Self> {
let dict_supported = props.dictionary_enabled(descr.path())
&& has_dictionary_support(T::get_physical_type(), &props);
let dict_encoder = dict_supported.then(|| DictEncoder::new(descr.clone()));
// Set either main encoder or fallback encoder.
let encoder = get_encoder(
descr.clone(),
props
.encoding(descr.path())
.unwrap_or_else(|| fallback_encoding(T::get_physical_type(), &props)),
)?;
Ok(Self {
encoder,
dict_encoder,
descr: descr.clone(),
num_values: 0,
min_value: None,
max_value: None,
})
}
fn write(&mut self, values: &[T::T], offset: usize, len: usize) -> Result<()> {
self.num_values += len;
let slice = values.get(offset..offset + len).ok_or_else(|| {
general_err!(
"Expected to write {} values, but have only {}",
len,
values.len() - offset
)
})?;
if let Some((min, max)) = slice.min_max(&self.descr) {
update_min(&self.descr, min, &mut self.min_value);
update_max(&self.descr, max, &mut self.max_value);
}
match &mut self.dict_encoder {
Some(encoder) => encoder.put(slice),
_ => self.encoder.put(slice),
}
}
fn num_values(&self) -> usize {
self.num_values
}
fn has_dictionary(&self) -> bool {
self.dict_encoder.is_some()
}
fn estimated_dict_page_size(&self) -> Option<usize> {
Some(self.dict_encoder.as_ref()?.dict_encoded_size())
}
fn estimated_data_page_size(&self) -> usize {
match &self.dict_encoder {
Some(encoder) => encoder.estimated_data_encoded_size(),
_ => self.encoder.estimated_data_encoded_size(),
}
}
fn flush_dict_page(&mut self) -> Result<Option<DictionaryPage>> {
match self.dict_encoder.take() {
Some(encoder) => {
if self.num_values != 0 {
return Err(general_err!(
"Must flush data pages before flushing dictionary"
));
}
let buf = encoder.write_dict()?;
Ok(Some(DictionaryPage {
buf,
num_values: encoder.num_entries(),
is_sorted: encoder.is_sorted(),
}))
}
_ => Ok(None),
}
}
fn flush_data_page(&mut self) -> Result<DataPageValues<T::T>> {
let (buf, encoding) = match &mut self.dict_encoder {
Some(encoder) => (encoder.write_indices()?, Encoding::RLE_DICTIONARY),
_ => (self.encoder.flush_buffer()?, self.encoder.encoding()),
};
Ok(DataPageValues {
buf,
encoding,
num_values: std::mem::take(&mut self.num_values),
min_value: self.min_value.take(),
max_value: self.max_value.take(),
})
}
}
impl<T: ParquetValueType> ColumnValues for [T] {
type T = T;
fn len(&self) -> usize {
self.len()
}
fn min_max(&self, descr: &ColumnDescriptor) -> Option<(&T, &T)> {
let mut iter = self.iter();
let first = loop {
let next = iter.next()?;
if !is_nan(next) {
break next
}
};
let mut min = first;
let mut max = first;
for val in iter {
if is_nan(val) {
continue
}
if compare_greater(descr, min, val) {
min = val;
}
if compare_greater(descr, val, max) {
max = val;
}
}
Some((min, max))
}
}