connectorx/
dispatcher.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
///! This module provides [`dispatcher::Dispatcher`], the core struct in ConnectorX
///! that drives the data loading from a source to a destination.
use crate::{
    data_order::{coordinate, DataOrder},
    destinations::{Destination, DestinationPartition},
    errors::{ConnectorXError, Result as CXResult},
    sources::{PartitionParser, Source, SourcePartition},
    sql::CXQuery,
    typesystem::Transport,
};
use itertools::Itertools;
use log::debug;
use rayon::prelude::*;
use std::marker::PhantomData;

/// A dispatcher takes a `S: Source`, a `D: Destination`, a `TP: Transport` and a vector of `queries` as input to
/// load data from `S` to `D` using the queries.
pub struct Dispatcher<'a, S, D, TP> {
    src: S,
    dst: &'a mut D,
    queries: Vec<CXQuery<String>>,
    origin_query: Option<String>,
    _phantom: PhantomData<TP>,
}

impl<'w, S, D, TP> Dispatcher<'w, S, D, TP>
where
    S: Source,
    D: Destination,
    TP: Transport<TSS = S::TypeSystem, TSD = D::TypeSystem, S = S, D = D>,
{
    /// Create a new dispatcher by providing a source, a destination and the queries.
    pub fn new<Q>(src: S, dst: &'w mut D, queries: &[Q], origin_query: Option<String>) -> Self
    where
        for<'a> &'a Q: Into<CXQuery>,
    {
        Self {
            src,
            dst,
            queries: queries.iter().map(Into::into).collect(),
            origin_query,
            _phantom: PhantomData,
        }
    }

    pub fn prepare(
        mut self,
    ) -> Result<
        (
            DataOrder,
            Vec<S::Partition>,
            Vec<D::Partition<'w>>,
            Vec<S::TypeSystem>,
            Vec<D::TypeSystem>,
        ),
        TP::Error,
    > {
        debug!("Prepare");
        let dorder = coordinate(S::DATA_ORDERS, D::DATA_ORDERS)?;
        self.src.set_data_order(dorder)?;
        self.src.set_queries(self.queries.as_slice());
        self.src.set_origin_query(self.origin_query);

        debug!("Fetching metadata");
        self.src.fetch_metadata()?;
        let src_schema = self.src.schema();
        let dst_schema = src_schema
            .iter()
            .map(|&s| TP::convert_typesystem(s))
            .collect::<CXResult<Vec<_>>>()?;
        let names = self.src.names();

        let mut total_rows = if self.dst.needs_count() {
            // return None if cannot derive total count
            debug!("Try get row rounts for entire result");
            self.src.result_rows()?
        } else {
            debug!("Do not need counts in advance");
            Some(0)
        };
        let mut src_partitions: Vec<S::Partition> = self.src.partition()?;
        if self.dst.needs_count() && total_rows.is_none() {
            debug!("Manually count rows of each partitioned query and sum up");
            // run queries
            src_partitions
                .par_iter_mut()
                .try_for_each(|partition| -> Result<(), S::Error> { partition.result_rows() })?;

            // get number of row of each partition from the source
            let part_rows: Vec<usize> = src_partitions
                .iter()
                .map(|partition| partition.nrows())
                .collect();
            total_rows = Some(part_rows.iter().sum());
        }
        let total_rows = total_rows.ok_or_else(ConnectorXError::CountError)?;

        debug!(
            "Allocate destination memory: {}x{}",
            total_rows,
            src_schema.len()
        );
        self.dst.allocate(total_rows, &names, &dst_schema, dorder)?;

        debug!("Create destination partition");
        let dst_partitions = self.dst.partition(self.queries.len())?;

        Ok((
            dorder,
            src_partitions,
            dst_partitions,
            src_schema,
            dst_schema,
        ))
    }

    /// Start the data loading process.
    pub fn run(self) -> Result<(), TP::Error> {
        debug!("Run dispatcher");
        let (dorder, src_partitions, dst_partitions, src_schema, dst_schema) = self.prepare()?;

        #[cfg(all(not(feature = "branch"), not(feature = "fptr")))]
        compile_error!("branch or fptr, pick one");

        #[cfg(feature = "branch")]
        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> {
                #[cfg(feature = "fptr")]
                let f: Vec<_> = src_schema
                    .iter()
                    .zip_eq(&dst_schema)
                    .map(|(&src_ty, &dst_ty)| TP::processor(src_ty, dst_ty))
                    .collect::<CXResult<Vec<_>>>()?;

                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() {
                                #[cfg(feature = "fptr")]
                                f[col](&mut parser, &mut dst)?;

                                #[cfg(feature = "branch")]
                                {
                                    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 {
                                #[cfg(feature = "fptr")]
                                f[col](&mut parser, &mut dst)?;
                                #[cfg(feature = "branch")]
                                {
                                    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(())
    }

    /// Only fetch the metadata (header) of the destination.
    pub fn get_meta(&mut self) -> Result<(), TP::Error> {
        let dorder = coordinate(S::DATA_ORDERS, D::DATA_ORDERS)?;
        self.src.set_data_order(dorder)?;
        self.src.set_queries(self.queries.as_slice());
        self.src.set_origin_query(self.origin_query.clone());
        self.src.fetch_metadata()?;
        let src_schema = self.src.schema();
        let dst_schema = src_schema
            .iter()
            .map(|&s| TP::convert_typesystem(s))
            .collect::<CXResult<Vec<_>>>()?;
        let names = self.src.names();
        self.dst.allocate(0, &names, &dst_schema, dorder)?;
        Ok(())
    }
}