1use crate::connection::{RemoteDbType, projections_contains};
2use crate::{
3 Connection, ConnectionOptions, DFResult, Pool, RemoteField, RemoteSchema, RemoteSchemaRef,
4 RemoteType, SqliteType,
5};
6use datafusion::arrow::array::{
7 ArrayBuilder, ArrayRef, BinaryBuilder, Float64Builder, Int32Builder, Int64Builder, NullBuilder,
8 RecordBatch, RecordBatchOptions, StringBuilder, make_builder,
9};
10use datafusion::arrow::datatypes::{DataType, SchemaRef};
11use datafusion::common::{DataFusionError, project_schema};
12use datafusion::execution::SendableRecordBatchStream;
13use datafusion::physical_plan::stream::RecordBatchStreamAdapter;
14use derive_getters::Getters;
15use derive_with::With;
16use itertools::Itertools;
17use log::{debug, error};
18use rusqlite::types::ValueRef;
19use rusqlite::{Column, Row, Rows};
20use std::any::Any;
21use std::collections::HashMap;
22use std::path::PathBuf;
23use std::sync::Arc;
24
25#[derive(Debug, Clone, With, Getters)]
26pub struct SqliteConnectionOptions {
27 pub path: PathBuf,
28 pub stream_chunk_size: usize,
29}
30
31impl SqliteConnectionOptions {
32 pub fn new(path: PathBuf) -> Self {
33 Self {
34 path,
35 stream_chunk_size: 2048,
36 }
37 }
38}
39
40impl From<SqliteConnectionOptions> for ConnectionOptions {
41 fn from(options: SqliteConnectionOptions) -> Self {
42 ConnectionOptions::Sqlite(options)
43 }
44}
45
46#[derive(Debug)]
47pub struct SqlitePool {
48 path: PathBuf,
49}
50
51pub async fn connect_sqlite(options: &SqliteConnectionOptions) -> DFResult<SqlitePool> {
52 let _ = rusqlite::Connection::open(&options.path).map_err(|e| {
53 DataFusionError::Execution(format!("Failed to open sqlite connection: {e:?}"))
54 })?;
55 Ok(SqlitePool {
56 path: options.path.clone(),
57 })
58}
59
60#[async_trait::async_trait]
61impl Pool for SqlitePool {
62 async fn get(&self) -> DFResult<Arc<dyn Connection>> {
63 Ok(Arc::new(SqliteConnection {
64 path: self.path.clone(),
65 }))
66 }
67}
68
69#[derive(Debug)]
70pub struct SqliteConnection {
71 path: PathBuf,
72}
73
74#[async_trait::async_trait]
75impl Connection for SqliteConnection {
76 fn as_any(&self) -> &dyn Any {
77 self
78 }
79
80 async fn infer_schema(&self, sql: &str) -> DFResult<RemoteSchemaRef> {
81 let sql = RemoteDbType::Sqlite.query_limit_1(sql);
82 let conn = rusqlite::Connection::open(&self.path).map_err(|e| {
83 DataFusionError::Execution(format!("Failed to open sqlite connection: {e:?}"))
84 })?;
85 let mut stmt = conn.prepare(&sql).map_err(|e| {
86 DataFusionError::Execution(format!("Failed to prepare sqlite statement: {e:?}"))
87 })?;
88 let columns: Vec<OwnedColumn> =
89 stmt.columns().iter().map(sqlite_col_to_owned_col).collect();
90 let rows = stmt.query([]).map_err(|e| {
91 DataFusionError::Execution(format!("Failed to query sqlite statement: {e:?}"))
92 })?;
93
94 let remote_schema = Arc::new(build_remote_schema(columns.as_slice(), rows)?);
95 Ok(remote_schema)
96 }
97
98 async fn query(
99 &self,
100 conn_options: &ConnectionOptions,
101 sql: &str,
102 table_schema: SchemaRef,
103 projection: Option<&Vec<usize>>,
104 unparsed_filters: &[String],
105 limit: Option<usize>,
106 ) -> DFResult<SendableRecordBatchStream> {
107 let projected_schema = project_schema(&table_schema, projection)?;
108 let sql = RemoteDbType::Sqlite.rewrite_query(sql, unparsed_filters, limit);
109 debug!("[remote-table] executing sqlite query: {sql}");
110
111 let (tx, mut rx) = tokio::sync::mpsc::channel::<DFResult<RecordBatch>>(1);
112 let conn = rusqlite::Connection::open(&self.path).map_err(|e| {
113 DataFusionError::Execution(format!("Failed to open sqlite connection: {e:?}"))
114 })?;
115
116 let projection = projection.cloned();
117 let chunk_size = conn_options.stream_chunk_size();
118
119 spawn_background_task(tx, conn, sql, table_schema, projection, chunk_size);
120
121 let stream = async_stream::stream! {
122 while let Some(batch) = rx.recv().await {
123 yield batch;
124 }
125 };
126 Ok(Box::pin(RecordBatchStreamAdapter::new(
127 projected_schema,
128 stream,
129 )))
130 }
131}
132
133#[derive(Debug)]
134struct OwnedColumn {
135 name: String,
136 decl_type: Option<String>,
137}
138
139fn sqlite_col_to_owned_col(sqlite_col: &Column) -> OwnedColumn {
140 OwnedColumn {
141 name: sqlite_col.name().to_string(),
142 decl_type: sqlite_col.decl_type().map(|x| x.to_string()),
143 }
144}
145
146fn decl_type_to_remote_type(decl_type: &str) -> DFResult<SqliteType> {
147 if ["tinyint", "smallint", "int", "integer", "bigint"].contains(&decl_type) {
148 return Ok(SqliteType::Integer);
149 }
150 if ["real", "float", "double"].contains(&decl_type) {
151 return Ok(SqliteType::Real);
152 }
153 if decl_type.starts_with("real") {
154 return Ok(SqliteType::Real);
155 }
156 if ["text", "varchar", "char", "string"].contains(&decl_type) {
157 return Ok(SqliteType::Text);
158 }
159 if decl_type.starts_with("char")
160 || decl_type.starts_with("varchar")
161 || decl_type.starts_with("text")
162 {
163 return Ok(SqliteType::Text);
164 }
165 if ["binary", "varbinary", "tinyblob", "blob"].contains(&decl_type) {
166 return Ok(SqliteType::Blob);
167 }
168 if decl_type.starts_with("binary") || decl_type.starts_with("varbinary") {
169 return Ok(SqliteType::Blob);
170 }
171 Err(DataFusionError::NotImplemented(format!(
172 "Unsupported sqlite decl type: {decl_type}",
173 )))
174}
175
176fn build_remote_schema(columns: &[OwnedColumn], mut rows: Rows) -> DFResult<RemoteSchema> {
177 let mut remote_field_map = HashMap::with_capacity(columns.len());
178 let mut unknown_cols = vec![];
179 for (col_idx, col) in columns.iter().enumerate() {
180 if let Some(decl_type) = &col.decl_type {
181 let remote_type =
182 RemoteType::Sqlite(decl_type_to_remote_type(&decl_type.to_ascii_lowercase())?);
183 remote_field_map.insert(col_idx, RemoteField::new(&col.name, remote_type, true));
184 } else {
185 unknown_cols.push(col_idx);
187 }
188 }
189
190 if !unknown_cols.is_empty() {
191 while let Some(row) = rows.next().map_err(|e| {
192 DataFusionError::Execution(format!("Failed to get next row from sqlite: {e:?}"))
193 })? {
194 let mut to_be_removed = vec![];
195 for col_idx in unknown_cols.iter() {
196 let value_ref = row.get_ref(*col_idx).map_err(|e| {
197 DataFusionError::Execution(format!(
198 "Failed to get value ref for column {col_idx}: {e:?}"
199 ))
200 })?;
201 match value_ref {
202 ValueRef::Null => {}
203 ValueRef::Integer(_) => {
204 remote_field_map.insert(
205 *col_idx,
206 RemoteField::new(
207 columns[*col_idx].name.clone(),
208 RemoteType::Sqlite(SqliteType::Integer),
209 true,
210 ),
211 );
212 to_be_removed.push(*col_idx);
213 }
214 ValueRef::Real(_) => {
215 remote_field_map.insert(
216 *col_idx,
217 RemoteField::new(
218 columns[*col_idx].name.clone(),
219 RemoteType::Sqlite(SqliteType::Real),
220 true,
221 ),
222 );
223 to_be_removed.push(*col_idx);
224 }
225 ValueRef::Text(_) => {
226 remote_field_map.insert(
227 *col_idx,
228 RemoteField::new(
229 columns[*col_idx].name.clone(),
230 RemoteType::Sqlite(SqliteType::Text),
231 true,
232 ),
233 );
234 to_be_removed.push(*col_idx);
235 }
236 ValueRef::Blob(_) => {
237 remote_field_map.insert(
238 *col_idx,
239 RemoteField::new(
240 columns[*col_idx].name.clone(),
241 RemoteType::Sqlite(SqliteType::Blob),
242 true,
243 ),
244 );
245 to_be_removed.push(*col_idx);
246 }
247 }
248 }
249 for col_idx in to_be_removed.iter() {
250 unknown_cols.retain(|&x| x != *col_idx);
251 }
252 if unknown_cols.is_empty() {
253 break;
254 }
255 }
256 }
257
258 if !unknown_cols.is_empty() {
259 return Err(DataFusionError::NotImplemented(format!(
260 "Failed to infer sqlite decl type for columns: {unknown_cols:?}"
261 )));
262 }
263 let remote_fields = remote_field_map
264 .into_iter()
265 .sorted_by_key(|entry| entry.0)
266 .map(|entry| entry.1)
267 .collect::<Vec<_>>();
268 Ok(RemoteSchema::new(remote_fields))
269}
270
271fn spawn_background_task(
272 tx: tokio::sync::mpsc::Sender<DFResult<RecordBatch>>,
273 conn: rusqlite::Connection,
274 sql: String,
275 table_schema: SchemaRef,
276 projection: Option<Vec<usize>>,
277 chunk_size: usize,
278) {
279 std::thread::spawn(move || {
280 let runtime = match tokio::runtime::Builder::new_current_thread().build() {
281 Ok(runtime) => runtime,
282 Err(e) => {
283 error!("Failed to create tokio runtime to run sqlite query: {e:?}");
284 return;
285 }
286 };
287 let local_set = tokio::task::LocalSet::new();
288 local_set.block_on(&runtime, async move {
289 let mut stmt = match conn.prepare(&sql) {
290 Ok(stmt) => stmt,
291 Err(e) => {
292 let _ = tx
293 .send(Err(DataFusionError::Execution(format!(
294 "Failed to prepare sqlite statement: {e:?}"
295 ))))
296 .await;
297 return;
298 }
299 };
300 let columns: Vec<OwnedColumn> =
301 stmt.columns().iter().map(sqlite_col_to_owned_col).collect();
302 let mut rows = match stmt.query([]) {
303 Ok(rows) => rows,
304 Err(e) => {
305 let _ = tx
306 .send(Err(DataFusionError::Execution(format!(
307 "Failed to query sqlite statement: {e:?}"
308 ))))
309 .await;
310 return;
311 }
312 };
313
314 loop {
315 let (batch, is_empty) = match rows_to_batch(
316 &mut rows,
317 &table_schema,
318 &columns,
319 projection.as_ref(),
320 chunk_size,
321 ) {
322 Ok((batch, is_empty)) => (batch, is_empty),
323 Err(e) => {
324 let _ = tx
325 .send(Err(DataFusionError::Execution(format!(
326 "Failed to convert rows to batch: {e:?}"
327 ))))
328 .await;
329 return;
330 }
331 };
332 if is_empty {
333 break;
334 }
335 if tx.send(Ok(batch)).await.is_err() {
336 return;
337 }
338 }
339 });
340 });
341}
342
343fn rows_to_batch(
344 rows: &mut Rows,
345 table_schema: &SchemaRef,
346 columns: &[OwnedColumn],
347 projection: Option<&Vec<usize>>,
348 chunk_size: usize,
349) -> DFResult<(RecordBatch, bool)> {
350 let projected_schema = project_schema(table_schema, projection)?;
351 let mut array_builders = vec![];
352 for field in table_schema.fields() {
353 let builder = make_builder(field.data_type(), 1000);
354 array_builders.push(builder);
355 }
356
357 let mut is_empty = true;
358 let mut row_count = 0;
359 while let Some(row) = rows.next().map_err(|e| {
360 DataFusionError::Execution(format!("Failed to get next row from sqlite: {e:?}"))
361 })? {
362 is_empty = false;
363 row_count += 1;
364 append_rows_to_array_builders(
365 row,
366 table_schema,
367 columns,
368 projection,
369 array_builders.as_mut_slice(),
370 )?;
371 if row_count >= chunk_size {
372 break;
373 }
374 }
375
376 let projected_columns = array_builders
377 .into_iter()
378 .enumerate()
379 .filter(|(idx, _)| projections_contains(projection, *idx))
380 .map(|(_, mut builder)| builder.finish())
381 .collect::<Vec<ArrayRef>>();
382 let options = RecordBatchOptions::new().with_row_count(Some(row_count));
383 Ok((
384 RecordBatch::try_new_with_options(projected_schema, projected_columns, &options)?,
385 is_empty,
386 ))
387}
388
389macro_rules! handle_primitive_type {
390 ($builder:expr, $field:expr, $col:expr, $builder_ty:ty, $value_ty:ty, $row:expr, $index:expr) => {{
391 let builder = $builder
392 .as_any_mut()
393 .downcast_mut::<$builder_ty>()
394 .unwrap_or_else(|| {
395 panic!(
396 "Failed to downcast builder to {} for {:?} and {:?}",
397 stringify!($builder_ty),
398 $field,
399 $col
400 )
401 });
402
403 let v: Option<$value_ty> = $row.get($index).map_err(|e| {
404 DataFusionError::Execution(format!(
405 "Failed to get optional {} value for {:?} and {:?}: {e:?}",
406 stringify!($value_ty),
407 $field,
408 $col
409 ))
410 })?;
411
412 match v {
413 Some(v) => builder.append_value(v),
414 None => builder.append_null(),
415 }
416 }};
417}
418
419fn append_rows_to_array_builders(
420 row: &Row,
421 table_schema: &SchemaRef,
422 columns: &[OwnedColumn],
423 projection: Option<&Vec<usize>>,
424 array_builders: &mut [Box<dyn ArrayBuilder>],
425) -> DFResult<()> {
426 for (idx, field) in table_schema.fields.iter().enumerate() {
427 if !projections_contains(projection, idx) {
428 continue;
429 }
430 let builder = &mut array_builders[idx];
431 let col = columns.get(idx);
432 match field.data_type() {
433 DataType::Null => {
434 let builder = builder
435 .as_any_mut()
436 .downcast_mut::<NullBuilder>()
437 .expect("Failed to downcast builder to NullBuilder");
438 builder.append_null();
439 }
440 DataType::Int32 => {
441 handle_primitive_type!(builder, field, col, Int32Builder, i32, row, idx);
442 }
443 DataType::Int64 => {
444 handle_primitive_type!(builder, field, col, Int64Builder, i64, row, idx);
445 }
446 DataType::Float64 => {
447 handle_primitive_type!(builder, field, col, Float64Builder, f64, row, idx);
448 }
449 DataType::Utf8 => {
450 handle_primitive_type!(builder, field, col, StringBuilder, String, row, idx);
451 }
452 DataType::Binary => {
453 handle_primitive_type!(builder, field, col, BinaryBuilder, Vec<u8>, row, idx);
454 }
455 _ => {
456 return Err(DataFusionError::NotImplemented(format!(
457 "Unsupported data type {:?} for col: {:?}",
458 field.data_type(),
459 col
460 )));
461 }
462 }
463 }
464 Ok(())
465}