sqlload 0.55.0

SQL data loading tooling.
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
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
//! Rows read value by value with `SQLGetData`: every column of a table with a column too long to bind, each value at its own size.

use std::sync::Arc;

use arrow_array::builder::{
    BinaryBuilder, BooleanBuilder, Date32Builder, FixedSizeBinaryBuilder, Float32Builder,
    Float64Builder, Int8Builder, Int16Builder, Int32Builder, Int64Builder, StringBuilder,
    Time32MillisecondBuilder, Time32SecondBuilder, Time64MicrosecondBuilder,
    Time64NanosecondBuilder, TimestampMicrosecondBuilder, TimestampNanosecondBuilder, UInt8Builder,
};
use arrow_array::{ArrayRef, RecordBatch};
use arrow_odbc::odbc_api::buffers::Indicator;
use arrow_odbc::odbc_api::parameter::{Binary, Text, VarCell, VarKind, WideText};
use arrow_odbc::odbc_api::sys::{Date, Time, Timestamp};
use arrow_odbc::odbc_api::{Bit, Cursor, CursorRow, Nullable, Pod};
use arrow_schema::{DataType, Field, SchemaRef, TimeUnit};
use chrono::{NaiveDate, NaiveTime, Timelike};

use crate::odbc_source::constants::{KEPT_VALUE_BUFFER, VALUE_OVERHEAD};
use crate::odbc_source::errors::OdbcSourceError;
use crate::odbc_source::models::ValueLimits;

/// One column read value by value: what it appends to, and the buffer its text or binary values are read into.
struct ValueColumn {
    name: String,
    values: Values,
    bytes: Vec<u8>,
    wide: Vec<u16>,
}

/// The Arrow array a column's values are appended to, by its fetched type.
enum Values {
    Boolean(BooleanBuilder),
    Int8(Int8Builder),
    Int16(Int16Builder),
    Int32(Int32Builder),
    Int64(Int64Builder),
    UInt8(UInt8Builder),
    Float32(Float32Builder),
    Float64(Float64Builder),
    Date(Date32Builder),
    TimeSeconds(Time32SecondBuilder),
    TimeMilliseconds(Time32MillisecondBuilder),
    TimeMicroseconds(Time64MicrosecondBuilder),
    TimeNanoseconds(Time64NanosecondBuilder),
    TimestampMicroseconds(TimestampMicrosecondBuilder, Option<Arc<str>>),
    TimestampNanoseconds(TimestampNanosecondBuilder, Option<Arc<str>>),
    Text(StringBuilder),
    Binary(BinaryBuilder),
    FixedBinary(FixedSizeBinaryBuilder, i32),
}

/// Fetch `cursor`'s rows one at a time, reading every value of `schema`'s columns (as `fetch_schema` fetches them) with `SQLGetData` in column order, and hand them to `send` in batches that end at `max_rows` rows or once their values hold `budget` bytes; a value longer than `max_text_size` or `max_binary_size` fails the read rather than being cut. A single value bigger than the budget loads, ends its batch, and is logged with `table`'s name. Stops when the rows end or `send` says no more.
pub(crate) fn read_values<C: Cursor>(
    (mut cursor, schema): (C, &SchemaRef),
    (budget, max_rows): (usize, usize),
    (limits, table): (&ValueLimits, &str),
    mut send: impl FnMut(RecordBatch) -> bool,
) -> Result<(), OdbcSourceError> {
    let mut columns: Vec<ValueColumn> = schema
        .fields()
        .iter()
        .map(|field| ValueColumn::new(field))
        .collect::<Result<_, _>>()?;
    loop {
        let (mut rows, mut bytes, mut ended) = (0usize, 0usize, false);
        while rows < max_rows.max(1) && bytes < budget {
            let Some(mut row) = cursor.next_row()? else {
                ended = true;
                break;
            };
            for (column, number) in columns.iter_mut().zip(1u16..) {
                let (next, size) = column.read(row, number, limits)?;
                row = next;
                if size > budget {
                    tracing::warn!(
                        table,
                        column = %column.name,
                        bytes = size,
                        budget,
                        "a value is larger than the table's fetch buffer (max_batch_memory); it loads, and its fetch ends after its row"
                    );
                }
                bytes = bytes.saturating_add(size + VALUE_OVERHEAD);
            }
            rows += 1;
        }
        if rows > 0 {
            tracing::debug!(table, rows, bytes, budget, "fetched rows value by value");
            let arrays: Vec<ArrayRef> = columns.iter_mut().map(ValueColumn::finish).collect();
            let batch = RecordBatch::try_new(schema.clone(), arrays)
                .map_err(|error| OdbcSourceError::Invalid(error.to_string()))?;
            if !send(batch) {
                return Ok(());
            }
        }
        if ended {
            return Ok(());
        }
    }
}

impl ValueColumn {
    /// An empty column of `field`'s fetched type.
    fn new(field: &Field) -> Result<Self, OdbcSourceError> {
        let values = match field.data_type() {
            DataType::Boolean => Values::Boolean(BooleanBuilder::new()),
            DataType::Int8 => Values::Int8(Int8Builder::new()),
            DataType::Int16 => Values::Int16(Int16Builder::new()),
            DataType::Int32 => Values::Int32(Int32Builder::new()),
            DataType::Int64 => Values::Int64(Int64Builder::new()),
            DataType::UInt8 => Values::UInt8(UInt8Builder::new()),
            DataType::Float32 => Values::Float32(Float32Builder::new()),
            DataType::Float64 => Values::Float64(Float64Builder::new()),
            DataType::Date32 => Values::Date(Date32Builder::new()),
            DataType::Time32(TimeUnit::Second) => Values::TimeSeconds(Time32SecondBuilder::new()),
            DataType::Time32(TimeUnit::Millisecond) => {
                Values::TimeMilliseconds(Time32MillisecondBuilder::new())
            }
            DataType::Time64(TimeUnit::Microsecond) => {
                Values::TimeMicroseconds(Time64MicrosecondBuilder::new())
            }
            DataType::Time64(TimeUnit::Nanosecond) => {
                Values::TimeNanoseconds(Time64NanosecondBuilder::new())
            }
            DataType::Timestamp(TimeUnit::Microsecond, zone) => {
                Values::TimestampMicroseconds(TimestampMicrosecondBuilder::new(), zone.clone())
            }
            DataType::Timestamp(TimeUnit::Nanosecond, zone) => {
                Values::TimestampNanoseconds(TimestampNanosecondBuilder::new(), zone.clone())
            }
            DataType::Utf8 => Values::Text(StringBuilder::new()),
            DataType::Binary => Values::Binary(BinaryBuilder::new()),
            DataType::FixedSizeBinary(length) => {
                Values::FixedBinary(FixedSizeBinaryBuilder::new(*length), *length)
            }
            other => {
                return Err(OdbcSourceError::Invalid(format!(
                    "column {} is fetched as {other}, which SQLLoad doesn't read value by value",
                    field.name()
                )));
            }
        };
        Ok(Self {
            name: field.name().clone(),
            values,
            bytes: Vec::new(),
            wide: Vec::new(),
        })
    }

    /// Read the current row's value of column `number` from `row` and append it; returns the row, and the value's bytes (a text or binary value's own length).
    fn read<'r>(
        &mut self,
        mut row: CursorRow<'r>,
        number: u16,
        limits: &ValueLimits,
    ) -> Result<(CursorRow<'r>, usize), OdbcSourceError> {
        let row_ref = &mut row;
        let name = self.name.as_str();
        let failed = |error: arrow_odbc::odbc_api::Error| {
            OdbcSourceError::Invalid(format!("column {name}: ODBC: {error}"))
        };
        let size = match &mut self.values {
            Values::Boolean(builder) => {
                builder.append_option(
                    fixed::<Bit>(row_ref, number)
                        .map_err(failed)?
                        .map(Bit::as_bool),
                );
                size_of::<bool>()
            }
            Values::Int8(builder) => {
                builder.append_option(fixed(row_ref, number).map_err(failed)?);
                size_of::<i8>()
            }
            Values::Int16(builder) => {
                builder.append_option(fixed(row_ref, number).map_err(failed)?);
                size_of::<i16>()
            }
            Values::Int32(builder) => {
                builder.append_option(fixed(row_ref, number).map_err(failed)?);
                size_of::<i32>()
            }
            Values::Int64(builder) => {
                builder.append_option(fixed(row_ref, number).map_err(failed)?);
                size_of::<i64>()
            }
            Values::UInt8(builder) => {
                builder.append_option(fixed(row_ref, number).map_err(failed)?);
                size_of::<u8>()
            }
            Values::Float32(builder) => {
                builder.append_option(fixed(row_ref, number).map_err(failed)?);
                size_of::<f32>()
            }
            Values::Float64(builder) => {
                builder.append_option(fixed(row_ref, number).map_err(failed)?);
                size_of::<f64>()
            }
            Values::Date(builder) => {
                let days = match fixed::<Date>(row_ref, number).map_err(failed)? {
                    Some(date) => Some(epoch_days(&date).ok_or_else(|| invalid(name, "date"))?),
                    None => None,
                };
                builder.append_option(days);
                size_of::<i32>()
            }
            Values::TimeSeconds(builder) => {
                let time = fixed::<Time>(row_ref, number).map_err(failed)?;
                builder.append_option(time.map(|time| {
                    (i32::from(time.hour) * 60 + i32::from(time.minute)) * 60
                        + i32::from(time.second)
                }));
                size_of::<i32>()
            }
            Values::TimeMilliseconds(builder) => {
                let time = time_text((&mut self.bytes, &self.name), row_ref, number)?;
                builder.append_option(time.map(|(seconds, nanoseconds)| {
                    i32::try_from(seconds * 1_000 + nanoseconds / 1_000_000).unwrap_or(i32::MAX)
                }));
                size_of::<i32>()
            }
            Values::TimeMicroseconds(builder) => {
                let time = time_text((&mut self.bytes, &self.name), row_ref, number)?;
                builder.append_option(
                    time.map(|(seconds, nanoseconds)| seconds * 1_000_000 + nanoseconds / 1_000),
                );
                size_of::<i64>()
            }
            Values::TimeNanoseconds(builder) => {
                let time = time_text((&mut self.bytes, &self.name), row_ref, number)?;
                builder.append_option(
                    time.map(|(seconds, nanoseconds)| seconds * 1_000_000_000 + nanoseconds),
                );
                size_of::<i64>()
            }
            Values::TimestampMicroseconds(builder, _) => {
                let micros = match fixed::<Timestamp>(row_ref, number).map_err(failed)? {
                    Some(time) => {
                        Some(epoch_micros(&time).ok_or_else(|| invalid(name, "timestamp"))?)
                    }
                    None => None,
                };
                builder.append_option(micros);
                size_of::<i64>()
            }
            Values::TimestampNanoseconds(builder, _) => {
                let nanos = match fixed::<Timestamp>(row_ref, number).map_err(failed)? {
                    Some(time) => Some(
                        epoch_nanos(&time)
                            .ok_or_else(|| invalid(name, "timestamp in nanoseconds"))?,
                    ),
                    None => None,
                };
                builder.append_option(nanos);
                size_of::<i64>()
            }
            Values::Text(_) => self.read_text(row_ref, number, limits.max_text_size)?,
            Values::Binary(_) | Values::FixedBinary(..) => {
                self.read_binary(row_ref, number, limits.max_binary_size)?
            }
        };
        Ok((row, size))
    }

    /// The current row's text value of column `number`, as UTF-16 on Windows and UTF-8 elsewhere (as arrow-odbc reads text), appended as UTF-8; fails when it is longer than `cap` bytes of UTF-8 (on Windows too) or isn't valid text. The read stops as soon as the pieces read pass `cap`: a UTF-16 unit is at least one byte of UTF-8.
    fn read_text(
        &mut self,
        row: &mut CursorRow<'_>,
        number: u16,
        cap: Option<usize>,
    ) -> Result<usize, OdbcSourceError> {
        let Values::Text(builder) = &mut self.values else {
            return Ok(0);
        };
        let name = self.name.as_str();
        let size = match cfg!(target_os = "windows") {
            true => {
                let capped = (cap, "max_text_size");
                let wide = std::mem::take(&mut self.wide);
                let present;
                (self.wide, present) = row.read_capped::<WideText>(number, wide, (name, capped))?;
                if !present {
                    builder.append_null();
                    return Ok(0);
                }
                let text = String::from_utf16(&self.wide).map_err(|_| invalid(name, "text"))?;
                over_cap(name, (text.len(), true), cap, "max_text_size")?;
                builder.append_value(&text);
                self.wide = kept(std::mem::take(&mut self.wide));
                text.len()
            }
            false => {
                let capped = (cap, "max_text_size");
                let bytes = std::mem::take(&mut self.bytes);
                let present;
                (self.bytes, present) = row.read_capped::<Text>(number, bytes, (name, capped))?;
                if !present {
                    builder.append_null();
                    return Ok(0);
                }
                over_cap(name, (self.bytes.len(), true), cap, "max_text_size")?;
                let text = std::str::from_utf8(&self.bytes).map_err(|_| {
                    OdbcSourceError::Invalid(format!("column {name} holds text that is not UTF-8"))
                })?;
                builder.append_value(text);
                let size = self.bytes.len();
                self.bytes = kept(std::mem::take(&mut self.bytes));
                size
            }
        };
        Ok(size)
    }

    /// The current row's binary value of column `number`, appended; fails when it is longer than `cap` (stopping as soon as the pieces read pass it), or than a fixed-size column's width.
    fn read_binary(
        &mut self,
        row: &mut CursorRow<'_>,
        number: u16,
        cap: Option<usize>,
    ) -> Result<usize, OdbcSourceError> {
        let name = self.name.as_str();
        let capped = (cap, "max_binary_size");
        let bytes = std::mem::take(&mut self.bytes);
        let present;
        (self.bytes, present) = row.read_capped::<Binary>(number, bytes, (name, capped))?;
        over_cap(name, (self.bytes.len(), true), cap, "max_binary_size")?;
        let size = self.bytes.len();
        match (&mut self.values, present) {
            (Values::Binary(builder), true) => builder.append_value(&self.bytes),
            (Values::Binary(builder), false) => builder.append_null(),
            (Values::FixedBinary(builder, width), true) => {
                builder.append_value(&self.bytes).map_err(|_| {
                    OdbcSourceError::Invalid(format!(
                        "column {name} holds a value of {size} bytes, not its {width}"
                    ))
                })?;
            }
            (Values::FixedBinary(builder, _), false) => builder.append_null(),
            _ => {}
        }
        self.bytes = kept(std::mem::take(&mut self.bytes));
        Ok(size)
    }

    /// The column's values read since the last call, as one array; the column starts empty again.
    fn finish(&mut self) -> ArrayRef {
        match &mut self.values {
            Values::Boolean(builder) => Arc::new(builder.finish()),
            Values::Int8(builder) => Arc::new(builder.finish()),
            Values::Int16(builder) => Arc::new(builder.finish()),
            Values::Int32(builder) => Arc::new(builder.finish()),
            Values::Int64(builder) => Arc::new(builder.finish()),
            Values::UInt8(builder) => Arc::new(builder.finish()),
            Values::Float32(builder) => Arc::new(builder.finish()),
            Values::Float64(builder) => Arc::new(builder.finish()),
            Values::Date(builder) => Arc::new(builder.finish()),
            Values::TimeSeconds(builder) => Arc::new(builder.finish()),
            Values::TimeMilliseconds(builder) => Arc::new(builder.finish()),
            Values::TimeMicroseconds(builder) => Arc::new(builder.finish()),
            Values::TimeNanoseconds(builder) => Arc::new(builder.finish()),
            Values::TimestampMicroseconds(builder, zone) => {
                Arc::new(builder.finish().with_timezone_opt(zone.clone()))
            }
            Values::TimestampNanoseconds(builder, zone) => {
                Arc::new(builder.finish().with_timezone_opt(zone.clone()))
            }
            Values::Text(builder) => Arc::new(builder.finish()),
            Values::Binary(builder) => Arc::new(builder.finish()),
            Values::FixedBinary(builder, _) => Arc::new(builder.finish()),
        }
    }
}

/// The current row's time of column `number` (of column `name`), read into `bytes` as `HH:MM:SS.fff…` text, as arrow-odbc reads a time with fractional seconds: its seconds since midnight and the nanoseconds after them.
fn time_text(
    (bytes, name): (&mut Vec<u8>, &str),
    row: &mut CursorRow<'_>,
    number: u16,
) -> Result<Option<(i64, i64)>, OdbcSourceError> {
    let present = row
        .get_text(number, bytes)
        .map_err(|error| OdbcSourceError::Invalid(format!("column {name}: ODBC: {error}")))?;
    if !present {
        return Ok(None);
    }
    let text = String::from_utf8_lossy(bytes);
    let time =
        NaiveTime::parse_from_str(text.trim(), "%H:%M:%S%.f").map_err(|_| invalid(name, "time"))?;
    Ok(Some((
        i64::from(time.num_seconds_from_midnight()),
        i64::from(time.nanosecond()),
    )))
}

/// The current row's fixed-size value of column `number`; `None` when NULL.
fn fixed<T: Pod>(
    row: &mut CursorRow<'_>,
    number: u16,
) -> Result<Option<T>, arrow_odbc::odbc_api::Error> {
    let mut value = Nullable::<T>::null();
    row.get_data(number, &mut value)?;
    Ok(value.into_opt())
}

/// The date of an ODBC date's or timestamp's fields; `None` when there is no such day.
fn day(year: i16, month: u16, day: u16) -> Option<NaiveDate> {
    NaiveDate::from_ymd_opt(i32::from(year), u32::from(month), u32::from(day))
}

/// The wall-clock time of an ODBC timestamp; `None` when its fields make none.
fn moment(time: &Timestamp) -> Option<chrono::NaiveDateTime> {
    day(time.year, time.month, time.day)?.and_hms_nano_opt(
        u32::from(time.hour),
        u32::from(time.minute),
        u32::from(time.second),
        time.fraction,
    )
}

/// An ODBC date as days since 1970-01-01; `None` when there is no such day.
fn epoch_days(date: &Date) -> Option<i32> {
    Some(day(date.year, date.month, date.day)?.to_epoch_days())
}

/// An ODBC timestamp as microseconds since 1970; `None` when its fields make no time.
fn epoch_micros(time: &Timestamp) -> Option<i64> {
    Some(moment(time)?.and_utc().timestamp_micros())
}

/// An ODBC timestamp as nanoseconds since 1970; `None` when its fields make no time or it is outside 1677 to 2262.
fn epoch_nanos(time: &Timestamp) -> Option<i64> {
    moment(time)?.and_utc().timestamp_nanos_opt()
}

/// A value of column `name` that isn't a valid `what`.
fn invalid(name: &str, what: &str) -> OdbcSourceError {
    OdbcSourceError::Invalid(format!(
        "column {name} holds a value that is not a valid {what}"
    ))
}

/// Fails when a value of `length` passes `cap`, naming the setting: SQLLoad never cuts a value.
fn over_cap(
    name: &str,
    (length, exact): (usize, bool),
    cap: Option<usize>,
    setting: &str,
) -> Result<(), OdbcSourceError> {
    let at_least = match exact {
        true => "",
        false => "at least ",
    };
    match cap {
        Some(cap) if length > cap => Err(OdbcSourceError::Invalid(format!(
            "column {name} holds a value of {at_least}{length} bytes, longer than {setting} = {cap}; a value longer than its cap fails the read rather than being cut, so raise {setting} on the source, or remove it"
        ))),
        _ => Ok(()),
    }
}

/// A row's text or binary value read under a cap.
trait CappedValue {
    /// Read column `number`'s value into `buffer` piece by piece with `SQLGetData`, as odbc-api's `get_text` does, but stop with `over_cap`'s error as soon as the value is known to be longer than the cap (in `K`'s elements): once the driver reports the length still to come, or the pieces read pass it. Returns the buffer, and `false` for NULL (the buffer then empty).
    fn read_capped<K: VarKind>(
        &mut self,
        number: u16,
        buffer: Vec<K::Element>,
        capped: (&str, (Option<usize>, &str)),
    ) -> Result<(Vec<K::Element>, bool), OdbcSourceError>;
}

impl CappedValue for CursorRow<'_> {
    fn read_capped<K: VarKind>(
        &mut self,
        number: u16,
        mut buffer: Vec<K::Element>,
        (name, (cap, setting)): (&str, (Option<usize>, &str)),
    ) -> Result<(Vec<K::Element>, bool), OdbcSourceError> {
        let failed = |error: arrow_odbc::odbc_api::Error| {
            OdbcSourceError::Invalid(format!("column {name}: ODBC: {error}"))
        };
        let element = size_of::<K::Element>();
        if buffer.capacity() == 0 {
            buffer.reserve(256);
        }
        buffer.resize(buffer.capacity(), K::ZERO);
        let mut start = 0;
        loop {
            let mut piece = VarCell::<&mut [K::Element], K>::from_buffer(
                &mut buffer[start..],
                Indicator::NoTotal,
            );
            self.get_data(number, &mut piece).map_err(failed)?;
            let (complete, indicator) = (piece.is_complete(), piece.indicator());
            let Some(fetched) = piece.len_in_bytes() else {
                buffer.clear();
                return Ok((buffer, false));
            };
            if complete {
                buffer.truncate(start + fetched / element);
                return Ok((buffer, true));
            }
            let known = match indicator {
                Indicator::Length(length) => (start + length / element, true),
                _ => (start + fetched / element, false),
            };
            over_cap(name, known, cap, setting)?;
            let old = buffer.len();
            let grown = match indicator {
                Indicator::Length(length) => old + (length - fetched) / element,
                _ => old * 2,
            };
            buffer.resize(grown, K::ZERO);
            start = old - K::TERMINATING_ZEROES;
        }
    }
}

/// `buffer`, or an empty one once it has grown past the size kept between values, since each read fills the whole buffer.
fn kept<T>(buffer: Vec<T>) -> Vec<T> {
    match buffer.capacity() > KEPT_VALUE_BUFFER {
        true => Vec::new(),
        false => buffer,
    }
}