fix: ensure streaming parquet datasets deal with limits (#11977)

crates/polars-io/src/parquet/read_impl.rs

@@ -552,6 +552,14 @@ impl BatchedParquetReader {
         })
     }
 
+    pub fn limit_reached(&self) -> bool {
+        self.limit == 0
+    }
+
+    pub fn schema(&self) -> &SchemaRef {
+        &self.schema
+    }
+
     pub async fn next_batches(&mut self, n: usize) -> PolarsResult<Option<Vec<DataFrame>>> {
         if self.limit == 0 && self.has_returned {
             return Ok(None);

crates/polars-pipe/src/executors/sources/parquet.rs

@@ -2,7 +2,7 @@ use std::ops::Deref;
 use std::path::PathBuf;
 use std::sync::Arc;
 
-use polars_core::error::PolarsResult;
+use polars_core::error::*;
 use polars_core::utils::arrow::io::parquet::read::FileMetaData;
 use polars_core::POOL;
 use polars_io::cloud::CloudOptions;
@@ -22,6 +22,7 @@ use crate::pipeline::determine_chunk_size;
 pub struct ParquetSource {
     batched_reader: Option<BatchedParquetReader>,
     n_threads: usize,
+    processed_paths: usize,
     chunk_index: IdxSize,
     paths: std::slice::Iter<'static, PathBuf>,
     _paths_lifetime: Arc<[PathBuf]>,
@@ -69,6 +70,12 @@ impl ParquetSource {
             eprintln!("STREAMING CHUNK SIZE: {chunk_size} rows")
         }
 
+        let reader_schema = if self.processed_paths == 0 {
+            Some(self.file_info.reader_schema.clone())
+        } else {
+            None
+        };
+
         let batched_reader = if is_cloud_url(path) {
             #[cfg(not(feature = "async"))]
             {
@@ -83,7 +90,7 @@ impl ParquetSource {
                     ParquetAsyncReader::from_uri(
                         &uri,
                         self.cloud_options.as_ref(),
-                        Some(self.file_info.reader_schema.clone()),
+                        reader_schema,
                         self.metadata.clone(),
                     )
                     .await?
@@ -105,7 +112,7 @@ impl ParquetSource {
             let file = std::fs::File::open(path).unwrap();
 
             ParquetReader::new(file)
-                .with_schema(Some(self.file_info.reader_schema.clone()))
+                .with_schema(reader_schema)
                 .with_n_rows(file_options.n_rows)
                 .with_row_count(file_options.row_count)
                 .with_projection(projection)
@@ -118,7 +125,11 @@ impl ParquetSource {
                 )
                 .batched(chunk_size)?
         };
+        if self.processed_paths >= 1 {
+            polars_ensure!(batched_reader.schema().as_ref() == self.file_info.reader_schema.as_ref(), ComputeError: "schema of all files in a single scan_parquet must be equal");
+        }
         self.batched_reader = Some(batched_reader);
+        self.processed_paths += 1;
         Ok(())
     }
 
@@ -145,6 +156,7 @@ impl ParquetSource {
             batched_reader: None,
             n_threads,
             chunk_index: 0,
+            processed_paths: 0,
             options,
             file_options,
             paths: iter,
@@ -175,14 +187,23 @@ impl Source for ParquetSource {
         )?;
         Ok(match batches {
             None => {
+                if self.batched_reader.as_ref().unwrap().limit_reached() {
+                    return Ok(SourceResult::Finished);
+                }
                 // reset the reader
                 self.batched_reader = None;
+                self.metadata = None;
                 return self.get_batches(_context);
             },
             Some(batches) => SourceResult::GotMoreData(
                 batches
                     .into_iter()
                     .map(|data| {
+                        // Keep the row limit updated so the next reader will have a correct limit.
+                        if let Some(n_rows) = &mut self.file_options.n_rows {
+                            *n_rows = n_rows.saturating_sub(data.height())
+                        }
+
                         let chunk_index = self.chunk_index;
                         self.chunk_index += 1;
                         DataChunk { chunk_index, data }