connectorx/
arrow_batch_iter.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
use crate::prelude::*;
use arrow::record_batch::RecordBatch;
use itertools::Itertools;
use log::debug;
use rayon::prelude::*;
use std::marker::PhantomData;

pub fn set_global_num_thread(num: usize) {
    rayon::ThreadPoolBuilder::new()
        .num_threads(num)
        .build_global()
        .unwrap();
}

/// The iterator that returns arrow in `RecordBatch`
pub struct ArrowBatchIter<S, TP>
where
    S: Source,
    TP: Transport<
        TSS = S::TypeSystem,
        TSD = ArrowStreamTypeSystem,
        S = S,
        D = ArrowStreamDestination,
    >,
    <S as Source>::Partition: 'static,
    <S as Source>::TypeSystem: 'static,
    <TP as Transport>::Error: 'static,
{
    dst: ArrowStreamDestination,
    dst_parts: Option<Vec<ArrowStreamPartitionWriter>>,
    src_parts: Option<Vec<S::Partition>>,
    dorder: DataOrder,
    src_schema: Vec<S::TypeSystem>,
    dst_schema: Vec<ArrowStreamTypeSystem>,
    _phantom: PhantomData<TP>,
}

impl<'a, S, TP> ArrowBatchIter<S, TP>
where
    S: Source + 'a,
    TP: Transport<
        TSS = S::TypeSystem,
        TSD = ArrowStreamTypeSystem,
        S = S,
        D = ArrowStreamDestination,
    >,
{
    pub fn new(
        src: S,
        mut dst: ArrowStreamDestination,
        origin_query: Option<String>,
        queries: &[CXQuery<String>],
    ) -> Result<Self, TP::Error> {
        let dispatcher = Dispatcher::<_, _, TP>::new(src, &mut dst, queries, origin_query);
        let (dorder, src_parts, dst_parts, src_schema, dst_schema) = dispatcher.prepare()?;

        Ok(Self {
            dst,
            dst_parts: Some(dst_parts),
            src_parts: Some(src_parts),
            dorder,
            src_schema,
            dst_schema,
            _phantom: PhantomData,
        })
    }

    fn run(&mut self) {
        let src_schema = self.src_schema.clone();
        let dst_schema = self.dst_schema.clone();
        let src_partitions = self.src_parts.take().unwrap();
        let dst_partitions = self.dst_parts.take().unwrap();
        let dorder = self.dorder;

        std::thread::spawn(move || -> Result<(), TP::Error> {
            let schemas: Vec<_> = src_schema
                .iter()
                .zip_eq(&dst_schema)
                .map(|(&src_ty, &dst_ty)| (src_ty, dst_ty))
                .collect();

            debug!("Start writing");
            // parse and write
            dst_partitions
                .into_par_iter()
                .zip_eq(src_partitions)
                .enumerate()
                .try_for_each(|(i, (mut dst, mut src))| -> Result<(), TP::Error> {
                    let mut parser = src.parser()?;

                    match dorder {
                        DataOrder::RowMajor => loop {
                            let (n, is_last) = parser.fetch_next()?;
                            dst.aquire_row(n)?;
                            for _ in 0..n {
                                #[allow(clippy::needless_range_loop)]
                                for col in 0..dst.ncols() {
                                    {
                                        let (s1, s2) = schemas[col];
                                        TP::process(s1, s2, &mut parser, &mut dst)?;
                                    }
                                }
                            }
                            if is_last {
                                break;
                            }
                        },
                        DataOrder::ColumnMajor => loop {
                            let (n, is_last) = parser.fetch_next()?;
                            dst.aquire_row(n)?;
                            #[allow(clippy::needless_range_loop)]
                            for col in 0..dst.ncols() {
                                for _ in 0..n {
                                    {
                                        let (s1, s2) = schemas[col];
                                        TP::process(s1, s2, &mut parser, &mut dst)?;
                                    }
                                }
                            }
                            if is_last {
                                break;
                            }
                        },
                    }

                    debug!("Finalize partition {}", i);
                    dst.finalize()?;
                    debug!("Partition {} finished", i);
                    Ok(())
                })?;

            debug!("Writing finished");

            Ok(())
        });
    }
}

impl<'a, S, TP> Iterator for ArrowBatchIter<S, TP>
where
    S: Source + 'a,
    TP: Transport<
        TSS = S::TypeSystem,
        TSD = ArrowStreamTypeSystem,
        S = S,
        D = ArrowStreamDestination,
    >,
{
    type Item = RecordBatch;
    /// NOTE: not thread safe
    fn next(&mut self) -> Option<Self::Item> {
        self.dst.record_batch().unwrap()
    }
}

pub trait RecordBatchIterator {
    fn get_schema(&self) -> (RecordBatch, &[String]);
    fn prepare(&mut self);
    fn next_batch(&mut self) -> Option<RecordBatch>;
}

impl<'a, S, TP> RecordBatchIterator for ArrowBatchIter<S, TP>
where
    S: Source + 'a,
    TP: Transport<
        TSS = S::TypeSystem,
        TSD = ArrowStreamTypeSystem,
        S = S,
        D = ArrowStreamDestination,
    >,
{
    fn get_schema(&self) -> (RecordBatch, &[String]) {
        (self.dst.empty_batch(), self.dst.names())
    }

    fn prepare(&mut self) {
        self.run();
    }

    fn next_batch(&mut self) -> Option<RecordBatch> {
        self.next()
    }
}