16421522964

Committed 21 Jul 2025 03:40PM UTC coverage: 81.54% (+0.02%) from 81.523%

Build # 16421522964

Build Type

push

github

Committed by

web-flow

Commit Message

Improve work stealing loop (#3946)

Signed-off-by: Nicholas Gates <nick@nickgates.com>

Run Details

18 of 18 new or added lines in 1 file covered. (100.0%)

110 existing lines in 5 files now uncovered.

42099 of 51630 relevant lines covered (81.54%)

170770.88 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

80.77

/vortex-datafusion/src/persistent/format.rs

// SPDX-License-Identifier: Apache-2.0
// SPDX-FileCopyrightText: Copyright the Vortex contributors

use std::any::Any;
use std::fmt::{Debug, Formatter};
use std::sync::Arc;

use async_trait::async_trait;
use datafusion::arrow::datatypes::{Schema, SchemaRef};
use datafusion::catalog::Session;
use datafusion::common::parsers::CompressionTypeVariant;
use datafusion::common::runtime::SpawnedTask;
use datafusion::common::stats::Precision;
use datafusion::common::{
    ColumnStatistics, DataFusionError, GetExt, Result as DFResult, Statistics,
    config_datafusion_err, not_impl_err,
};
use datafusion::datasource::file_format::file_compression_type::FileCompressionType;
use datafusion::datasource::file_format::{FileFormat, FileFormatFactory};
use datafusion::datasource::physical_plan::{
    FileScanConfig, FileScanConfigBuilder, FileSinkConfig, FileSource,
};
use datafusion::datasource::sink::DataSinkExec;
use datafusion::datasource::source::DataSourceExec;
use datafusion::logical_expr::dml::InsertOp;
use datafusion::physical_expr::LexRequirement;
use datafusion::physical_plan::ExecutionPlan;
use futures::{FutureExt, StreamExt as _, TryStreamExt as _, stream};
use itertools::Itertools;
use object_store::{ObjectMeta, ObjectStore};
use vortex::dtype::DType;
use vortex::dtype::arrow::FromArrowType;
use vortex::error::{VortexExpect, VortexResult, vortex_err};
use vortex::file::VORTEX_FILE_EXTENSION;
use vortex::metrics::VortexMetrics;
use vortex::session::VortexSession;
use vortex::stats;
use vortex::stats::{Stat, StatsProviderExt, StatsSet};

use super::cache::VortexFileCache;
use super::sink::VortexSink;
use super::source::VortexSource;
use crate::PrecisionExt as _;
use crate::convert::TryToDataFusion;

/// Vortex implementation of a DataFusion [`FileFormat`].
pub struct VortexFormat {
    session: Arc<VortexSession>,
    file_cache: VortexFileCache,
    opts: VortexFormatOptions,
}

impl Debug for VortexFormat {
    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("VortexFormat")
            .field("opts", &self.opts)
            .finish()
    }
}

/// Options to configure the [`VortexFormat`].
#[derive(Debug)]
pub struct VortexFormatOptions {
    /// The size of the in-memory [`vortex::file::Footer`] cache.
    pub footer_cache_size_mb: usize,
    /// The size of the in-memory segment cache.
    pub segment_cache_size_mb: usize,
}

impl Default for VortexFormatOptions {
    fn default() -> Self {
        Self {
            footer_cache_size_mb: 64,
            segment_cache_size_mb: 0,
        }
    }
}

/// Minimal factory to create [`VortexFormat`] instances.
#[derive(Default, Debug)]
pub struct VortexFormatFactory {
    session: Arc<VortexSession>,
}

impl GetExt for VortexFormatFactory {
    fn get_ext(&self) -> String {
        VORTEX_FILE_EXTENSION.to_string()
    }
}

impl FileFormatFactory for VortexFormatFactory {
    #[allow(clippy::disallowed_types)]
    fn create(
        &self,
        _state: &dyn Session,
        format_options: &std::collections::HashMap<String, String>,
    ) -> DFResult<Arc<dyn FileFormat>> {
        if !format_options.is_empty() {
            return Err(config_datafusion_err!(
                "Vortex tables don't support any options"
            ));
        }

        Ok(Arc::new(VortexFormat::new(self.session.clone())))
    }

    fn default(&self) -> Arc<dyn FileFormat> {
        Arc::new(VortexFormat::default())
    }

    fn as_any(&self) -> &dyn Any {
        self
    }
}

impl Default for VortexFormat {
    fn default() -> Self {
        Self::new(Arc::new(VortexSession::default()))
    }
}

impl VortexFormat {
    /// Create a new instance of the [`VortexFormat`].
    pub fn new(session: Arc<VortexSession>) -> Self {
        let opts = VortexFormatOptions::default();
        Self {
            session: session.clone(),
            file_cache: VortexFileCache::new(
                opts.footer_cache_size_mb,
                opts.segment_cache_size_mb,
                session,
            ),
            opts,
        }
    }

    /// Return the format specific configuration
    pub fn options(&self) -> &VortexFormatOptions {
        &self.opts
    }
}

#[async_trait]
impl FileFormat for VortexFormat {
    fn as_any(&self) -> &dyn Any {
        self
    }

    fn get_ext(&self) -> String {
        VORTEX_FILE_EXTENSION.to_string()
    }

    fn get_ext_with_compression(
        &self,
        file_compression_type: &FileCompressionType,
    ) -> DFResult<String> {
        match file_compression_type.get_variant() {
            CompressionTypeVariant::UNCOMPRESSED => Ok(self.get_ext()),
            _ => Err(DataFusionError::Internal(
                "Vortex does not support file level compression.".into(),
            )),
        }
    }

    async fn infer_schema(
        &self,
        state: &dyn Session,
        store: &Arc<dyn ObjectStore>,
        objects: &[ObjectMeta],
    ) -> DFResult<SchemaRef> {
        let mut file_schemas = stream::iter(objects.iter().cloned())
            .map(|o| {
                let store = store.clone();
                let cache = self.file_cache.clone();
                SpawnedTask::spawn(async move {
                    let vxf = cache.try_get(&o, store).await?;
                    let inferred_schema = vxf.dtype().to_arrow_schema()?;
                    VortexResult::Ok((o.location, inferred_schema))
                })
                .map(|f| f.vortex_expect("Failed to spawn infer_schema"))
            })
            .buffer_unordered(state.config_options().execution.meta_fetch_concurrency)
            .try_collect::<Vec<_>>()
            .await
            .map_err(|e| DataFusionError::Execution(format!("Failed to infer schema: {e}")))?;

        // Get consistent order of schemas for `Schema::try_merge`, as some filesystems don't have deterministic listing orders
        file_schemas.sort_by(|(l1, _), (l2, _)| l1.cmp(l2));
        let file_schemas = file_schemas.into_iter().map(|(_, schema)| schema);

        Ok(Arc::new(Schema::try_merge(file_schemas)?))
    }

    #[cfg_attr(feature = "tracing", tracing::instrument(skip_all, fields(location = object.location.as_ref()
    )))]
    async fn infer_stats(
        &self,
        _state: &dyn Session,
        store: &Arc<dyn ObjectStore>,
        table_schema: SchemaRef,
        object: &ObjectMeta,
    ) -> DFResult<Statistics> {
        let object = object.clone();
        let store = store.clone();
        let cache = self.file_cache.clone();
        SpawnedTask::spawn(async move {
            let vxf = cache.try_get(&object, store.clone()).await.map_err(|e| {
                DataFusionError::Execution(format!(
                    "Failed to open Vortex file {}: {e}",
                    object.location
                ))
            })?;

            let struct_dtype = vxf
                .dtype()
                .as_struct()
                .vortex_expect("dtype is not a struct");

            // Evaluate the statistics for each column that we are able to return to DataFusion.
            let Some(file_stats) = vxf.file_stats() else {
                // If the file has no column stats, the best we can do is return a row count.
                return Ok(Statistics {
                    num_rows: Precision::Exact(
                        usize::try_from(vxf.row_count())
                            .map_err(|_| vortex_err!("Row count overflow"))
                            .vortex_expect("Row count overflow"),
                    ),
                    total_byte_size: Precision::Absent,
                    column_statistics: vec![ColumnStatistics::default(); struct_dtype.nfields()],
                });
            };

            let stats = table_schema
                .fields()
                .iter()
                .map(|field| struct_dtype.find(field.name()))
                .map(|idx| match idx {
                    None => StatsSet::default(),
                    Some(id) => file_stats[id].clone(),
                })
                .collect_vec();

            let total_byte_size = stats
                .iter()
                .map(|stats_set| {
                    stats_set
                        .get_as::<usize>(Stat::UncompressedSizeInBytes)
                        .unwrap_or_else(|| stats::Precision::inexact(0_usize))
                })
                .fold(stats::Precision::exact(0_usize), |acc, stats_set| {
                    acc.zip(stats_set).map(|(acc, stats_set)| acc + stats_set)
                });

            // Sum up the total byte size across all the columns.
            let total_byte_size = total_byte_size.to_df();

            let column_statistics = stats
                .into_iter()
                .zip(table_schema.fields().iter())
                .map(|(stats_set, field)| {
                    let null_count = stats_set.get_as::<usize>(Stat::NullCount);
                    let min = stats_set
                        .get_scalar(Stat::Min, &DType::from_arrow(field.as_ref()))
                        .and_then(|n| n.map(|n| n.try_to_df().ok()).transpose());

                    let max = stats_set
                        .get_scalar(Stat::Max, &DType::from_arrow(field.as_ref()))
                        .and_then(|n| n.map(|n| n.try_to_df().ok()).transpose());

                    ColumnStatistics {
                        null_count: null_count.to_df(),
                        max_value: max.to_df(),
                        min_value: min.to_df(),
                        sum_value: Precision::Absent,
                        distinct_count: stats_set
                            .get_as::<bool>(Stat::IsConstant)
                            .and_then(|is_constant| {
                                is_constant.as_exact().map(|_| Precision::Exact(1))
                            })
                            .unwrap_or(Precision::Absent),
                    }
                })
                .collect::<Vec<_>>();

            Ok(Statistics {
                num_rows: Precision::Exact(
                    usize::try_from(vxf.row_count())
                        .map_err(|_| vortex_err!("Row count overflow"))
                        .vortex_expect("Row count overflow"),
                ),
                total_byte_size,
                column_statistics,
            })
        })
        .await
        .vortex_expect("Failed to spawn infer_stats")
    }

    async fn create_physical_plan(
        &self,
        _state: &dyn Session,
        file_scan_config: FileScanConfig,
    ) -> DFResult<Arc<dyn ExecutionPlan>> {
        if file_scan_config
            .file_groups
            .iter()
            .flat_map(|fg| fg.files())
            .any(|f| f.range.is_some())
        {
            return not_impl_err!("File level partitioning isn't implemented yet for Vortex");
        }

        if !file_scan_config.table_partition_cols.is_empty() {
            return not_impl_err!("Hive style partitioning isn't implemented yet for Vortex");
        }

        if !file_scan_config.output_ordering.is_empty() {
            return not_impl_err!("Vortex doesn't support output ordering");
        }

        let source = VortexSource::new(self.file_cache.clone(), self.session.metrics().clone());
        Ok(DataSourceExec::from_data_source(
            FileScanConfigBuilder::from(file_scan_config)
                .with_source(Arc::new(source))
                .build(),
        ))
    }

    async fn create_writer_physical_plan(
        &self,
        input: Arc<dyn ExecutionPlan>,
        _state: &dyn Session,
        conf: FileSinkConfig,
        order_requirements: Option<LexRequirement>,
    ) -> DFResult<Arc<dyn ExecutionPlan>> {
        if conf.insert_op != InsertOp::Append {
            return not_impl_err!("Overwrites are not implemented yet for Vortex");
        }

        if !conf.table_partition_cols.is_empty() {
            return not_impl_err!("Hive style partitioning isn't implemented yet for Vortex");
        }

        let schema = conf.output_schema().clone();
        let sink = Arc::new(VortexSink::new(conf, schema));

        Ok(Arc::new(DataSinkExec::new(input, sink, order_requirements)) as _)
    }

    fn file_source(&self) -> Arc<dyn FileSource> {
        Arc::new(VortexSource::new(
            self.file_cache.clone(),
            VortexMetrics::default(),
        ))
    }
}

#[cfg(test)]
mod tests {
    use datafusion::execution::SessionStateBuilder;
    use datafusion::prelude::SessionContext;
    use tempfile::TempDir;

    use super::*;
    use crate::persistent::register_vortex_format_factory;

    #[tokio::test]
    async fn create_table() {
        let dir = TempDir::new().unwrap();

        let factory: VortexFormatFactory = Default::default();
        let mut session_state_builder = SessionStateBuilder::new().with_default_features();
        register_vortex_format_factory(factory, &mut session_state_builder);
        let session = SessionContext::new_with_state(session_state_builder.build());

        let df = session
            .sql(&format!(
                "CREATE EXTERNAL TABLE my_tbl \
                (c1 VARCHAR NOT NULL, c2 INT NOT NULL) \
                STORED AS vortex LOCATION '{}'",
                dir.path().to_str().unwrap()
            ))
            .await
            .unwrap();

        assert_eq!(df.count().await.unwrap(), 0);
    }

    #[tokio::test]
    #[should_panic]
    async fn fail_table_config() {
        let dir = TempDir::new().unwrap();

        let factory: VortexFormatFactory = Default::default();
        let mut session_state_builder = SessionStateBuilder::new().with_default_features();
        register_vortex_format_factory(factory, &mut session_state_builder);
        let session = SessionContext::new_with_state(session_state_builder.build());

        session
            .sql(&format!(
                "CREATE EXTERNAL TABLE my_tbl \
                (c1 VARCHAR NOT NULL, c2 INT NOT NULL) \
                STORED AS vortex LOCATION '{}' \
                OPTIONS( some_key 'value' );",
                dir.path().to_str().unwrap()
            ))
            .await
            .unwrap()
            .collect()
            .await
            .unwrap();
    }
}

1	// SPDX-License-Identifier: Apache-2.0
2	// SPDX-FileCopyrightText: Copyright the Vortex contributors
3
4	use std::any::Any;
5	use std::fmt::{Debug, Formatter};
6	use std::sync::Arc;
7
8	use async_trait::async_trait;
9	use datafusion::arrow::datatypes::{Schema, SchemaRef};
10	use datafusion::catalog::Session;
11	use datafusion::common::parsers::CompressionTypeVariant;
12	use datafusion::common::runtime::SpawnedTask;
13	use datafusion::common::stats::Precision;
14	use datafusion::common::{
15	ColumnStatistics, DataFusionError, GetExt, Result as DFResult, Statistics,
16	config_datafusion_err, not_impl_err,
17	};
18	use datafusion::datasource::file_format::file_compression_type::FileCompressionType;
19	use datafusion::datasource::file_format::{FileFormat, FileFormatFactory};
20	use datafusion::datasource::physical_plan::{
21	FileScanConfig, FileScanConfigBuilder, FileSinkConfig, FileSource,
22	};
23	use datafusion::datasource::sink::DataSinkExec;
24	use datafusion::datasource::source::DataSourceExec;
25	use datafusion::logical_expr::dml::InsertOp;
26	use datafusion::physical_expr::LexRequirement;
27	use datafusion::physical_plan::ExecutionPlan;
28	use futures::{FutureExt, StreamExt as _, TryStreamExt as _, stream};
29	use itertools::Itertools;
30	use object_store::{ObjectMeta, ObjectStore};
31	use vortex::dtype::DType;
32	use vortex::dtype::arrow::FromArrowType;
33	use vortex::error::{VortexExpect, VortexResult, vortex_err};
34	use vortex::file::VORTEX_FILE_EXTENSION;
35	use vortex::metrics::VortexMetrics;
36	use vortex::session::VortexSession;
37	use vortex::stats;
38	use vortex::stats::{Stat, StatsProviderExt, StatsSet};
39
40	use super::cache::VortexFileCache;
41	use super::sink::VortexSink;
42	use super::source::VortexSource;
43	use crate::PrecisionExt as _;
44	use crate::convert::TryToDataFusion;
45
46	/// Vortex implementation of a DataFusion [`FileFormat`].
47	pub struct VortexFormat {
48	session: Arc<VortexSession>,
49	file_cache: VortexFileCache,
50	opts: VortexFormatOptions,
51	}
52
53	impl Debug for VortexFormat {
54	fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {	×
55	f.debug_struct("VortexFormat")	×
56	.field("opts", &self.opts)	×
57	.finish()	×
UNCOV 58	}	×
59	}
60
61	/// Options to configure the [`VortexFormat`].
62	#[derive(Debug)]
63	pub struct VortexFormatOptions {
64	/// The size of the in-memory [`vortex::file::Footer`] cache.
65	pub footer_cache_size_mb: usize,
66	/// The size of the in-memory segment cache.
67	pub segment_cache_size_mb: usize,
68	}
69
70	impl Default for VortexFormatOptions {
71	fn default() -> Self {	20✔
72	Self {	20✔
73	footer_cache_size_mb: 64,	20✔
74	segment_cache_size_mb: 0,	20✔
75	}	20✔
76	}	20✔
77	}
78
79	/// Minimal factory to create [`VortexFormat`] instances.
80	#[derive(Default, Debug)]
81	pub struct VortexFormatFactory {
82	session: Arc<VortexSession>,
83	}
84
85	impl GetExt for VortexFormatFactory {
86	fn get_ext(&self) -> String {	18✔
87	VORTEX_FILE_EXTENSION.to_string()	18✔
88	}	18✔
89	}
90
91	impl FileFormatFactory for VortexFormatFactory {
92	#[allow(clippy::disallowed_types)]
93	fn create(	3✔
94	&self,	3✔
95	_state: &dyn Session,	3✔
96	format_options: &std::collections::HashMap<String, String>,	3✔
97	) -> DFResult<Arc<dyn FileFormat>> {	3✔
98	if !format_options.is_empty() {	3✔
99	return Err(config_datafusion_err!(	1✔
100	"Vortex tables don't support any options"	1✔
101	));	1✔
102	}	2✔
103
104	Ok(Arc::new(VortexFormat::new(self.session.clone())))	2✔
105	}	3✔
106
107	fn default(&self) -> Arc<dyn FileFormat> {	×
108	Arc::new(VortexFormat::default())	×
UNCOV 109	}	×
110
111	fn as_any(&self) -> &dyn Any {	×
112	self	×
UNCOV 113	}	×
114	}
115
116	impl Default for VortexFormat {
117	fn default() -> Self {	18✔
118	Self::new(Arc::new(VortexSession::default()))	18✔
119	}	18✔
120	}
121
122	impl VortexFormat {
123	/// Create a new instance of the [`VortexFormat`].
124	pub fn new(session: Arc<VortexSession>) -> Self {	20✔
125	let opts = VortexFormatOptions::default();	20✔
126	Self {	20✔
127	session: session.clone(),	20✔
128	file_cache: VortexFileCache::new(	20✔
129	opts.footer_cache_size_mb,	20✔
130	opts.segment_cache_size_mb,	20✔
131	session,	20✔
132	),	20✔
133	opts,	20✔
134	}	20✔
135	}	20✔
136
137	/// Return the format specific configuration
138	pub fn options(&self) -> &VortexFormatOptions {	×
139	&self.opts	×
UNCOV 140	}	×
141	}
142
143	#[async_trait]
144	impl FileFormat for VortexFormat {
145	fn as_any(&self) -> &dyn Any {	×
146	self	×
UNCOV 147	}	×
148
149	fn get_ext(&self) -> String {	22✔
150	VORTEX_FILE_EXTENSION.to_string()	22✔
151	}	22✔
152
153	fn get_ext_with_compression(	×
154	&self,	×
155	file_compression_type: &FileCompressionType,	×
156	) -> DFResult<String> {	×
157	match file_compression_type.get_variant() {	×
158	CompressionTypeVariant::UNCOMPRESSED => Ok(self.get_ext()),	×
159	_ => Err(DataFusionError::Internal(	×
160	"Vortex does not support file level compression.".into(),	×
UNCOV 161	)),	×
162	}
UNCOV 163	}	×
164
165	async fn infer_schema(
166	&self,
167	state: &dyn Session,
168	store: &Arc<dyn ObjectStore>,
169	objects: &[ObjectMeta],
170	) -> DFResult<SchemaRef> {	4✔
171	let mut file_schemas = stream::iter(objects.iter().cloned())	2✔
172	.map(\|o\| {	2✔
173	let store = store.clone();	2✔
174	let cache = self.file_cache.clone();	2✔
175	SpawnedTask::spawn(async move {	2✔
176	let vxf = cache.try_get(&o, store).await?;	2✔
177	let inferred_schema = vxf.dtype().to_arrow_schema()?;	2✔
178	VortexResult::Ok((o.location, inferred_schema))	2✔
179	})	2✔
180	.map(\|f\| f.vortex_expect("Failed to spawn infer_schema"))	2✔
181	})	2✔
182	.buffer_unordered(state.config_options().execution.meta_fetch_concurrency)	2✔
183	.try_collect::<Vec<_>>()	2✔
184	.await	2✔
185	.map_err(\|e\| DataFusionError::Execution(format!("Failed to infer schema: {e}")))?;	2✔
186
187	// Get consistent order of schemas for `Schema::try_merge`, as some filesystems don't have deterministic listing orders
188	file_schemas.sort_by(\|(l1, _), (l2, _)\| l1.cmp(l2));	2✔
189	let file_schemas = file_schemas.into_iter().map(\|(_, schema)\| schema);	2✔
190
191	Ok(Arc::new(Schema::try_merge(file_schemas)?))	2✔
192	}	4✔
193
194	#[cfg_attr(feature = "tracing", tracing::instrument(skip_all, fields(location = object.location.as_ref()
195	)))]
196	async fn infer_stats(
197	&self,
198	_state: &dyn Session,
199	store: &Arc<dyn ObjectStore>,
200	table_schema: SchemaRef,
201	object: &ObjectMeta,
202	) -> DFResult<Statistics> {	52✔
203	let object = object.clone();	28✔
204	let store = store.clone();	28✔
205	let cache = self.file_cache.clone();	28✔
206	SpawnedTask::spawn(async move {	28✔
207	let vxf = cache.try_get(&object, store.clone()).await.map_err(\|e\| {	28✔
208	DataFusionError::Execution(format!(	×
209	"Failed to open Vortex file {}: {e}",	×
210	object.location	×
211	))	×
UNCOV 212	})?;	×
213
214	let struct_dtype = vxf	28✔
215	.dtype()	28✔
216	.as_struct()	28✔
217	.vortex_expect("dtype is not a struct");	28✔
218
219	// Evaluate the statistics for each column that we are able to return to DataFusion.
220	let Some(file_stats) = vxf.file_stats() else {	28✔
221	// If the file has no column stats, the best we can do is return a row count.
222	return Ok(Statistics {
223	num_rows: Precision::Exact(
224	usize::try_from(vxf.row_count())	×
225	.map_err(\|_\| vortex_err!("Row count overflow"))	×
UNCOV 226	.vortex_expect("Row count overflow"),	×
227	),
228	total_byte_size: Precision::Absent,	×
UNCOV 229	column_statistics: vec![ColumnStatistics::default(); struct_dtype.nfields()],	×
230	});
231	};
232
233	let stats = table_schema	28✔
234	.fields()	28✔
235	.iter()	28✔
236	.map(\|field\| struct_dtype.find(field.name()))	168✔
237	.map(\|idx\| match idx {	168✔
UNCOV 238	None => StatsSet::default(),	×
239	Some(id) => file_stats[id].clone(),	168✔
240	})	168✔
241	.collect_vec();	28✔
242
243	let total_byte_size = stats	28✔
244	.iter()	28✔
245	.map(\|stats_set\| {	168✔
246	stats_set	168✔
247	.get_as::<usize>(Stat::UncompressedSizeInBytes)	168✔
248	.unwrap_or_else(\|\| stats::Precision::inexact(0_usize))	168✔
249	})	168✔
250	.fold(stats::Precision::exact(0_usize), \|acc, stats_set\| {	168✔
251	acc.zip(stats_set).map(\|(acc, stats_set)\| acc + stats_set)	168✔
252	});	168✔
253
254	// Sum up the total byte size across all the columns.
255	let total_byte_size = total_byte_size.to_df();	28✔
256
257	let column_statistics = stats	28✔
258	.into_iter()	28✔
259	.zip(table_schema.fields().iter())	28✔
260	.map(\|(stats_set, field)\| {	168✔
261	let null_count = stats_set.get_as::<usize>(Stat::NullCount);	168✔
262	let min = stats_set	168✔
263	.get_scalar(Stat::Min, &DType::from_arrow(field.as_ref()))	168✔
264	.and_then(\|n\| n.map(\|n\| n.try_to_df().ok()).transpose());	168✔
265
266	let max = stats_set	168✔
267	.get_scalar(Stat::Max, &DType::from_arrow(field.as_ref()))	168✔
268	.and_then(\|n\| n.map(\|n\| n.try_to_df().ok()).transpose());	168✔
269
270	ColumnStatistics {
271	null_count: null_count.to_df(),	168✔
272	max_value: max.to_df(),	168✔
273	min_value: min.to_df(),	168✔
274	sum_value: Precision::Absent,	168✔
275	distinct_count: stats_set	168✔
276	.get_as::<bool>(Stat::IsConstant)	168✔
277	.and_then(\|is_constant\| {	168✔
278	is_constant.as_exact().map(\|_\| Precision::Exact(1))	×
UNCOV 279	})	×
280	.unwrap_or(Precision::Absent),	168✔
281	}
282	})	168✔
283	.collect::<Vec<_>>();	28✔
284
285	Ok(Statistics {
286	num_rows: Precision::Exact(
287	usize::try_from(vxf.row_count())	28✔
288	.map_err(\|_\| vortex_err!("Row count overflow"))	28✔
289	.vortex_expect("Row count overflow"),	28✔
290	),
291	total_byte_size,	28✔
292	column_statistics,	28✔
293	})
294	})	28✔
295	.await	28✔
296	.vortex_expect("Failed to spawn infer_stats")	28✔
297	}	52✔
298
299	async fn create_physical_plan(
300	&self,
301	_state: &dyn Session,
302	file_scan_config: FileScanConfig,
303	) -> DFResult<Arc<dyn ExecutionPlan>> {	360✔
304	if file_scan_config	180✔
305	.file_groups	180✔
306	.iter()	180✔
307	.flat_map(\|fg\| fg.files())	182✔
308	.any(\|f\| f.range.is_some())	182✔
309	{
UNCOV 310	return not_impl_err!("File level partitioning isn't implemented yet for Vortex");	×
311	}	180✔
312
313	if !file_scan_config.table_partition_cols.is_empty() {	180✔
UNCOV 314	return not_impl_err!("Hive style partitioning isn't implemented yet for Vortex");	×
315	}	180✔
316
317	if !file_scan_config.output_ordering.is_empty() {	180✔
UNCOV 318	return not_impl_err!("Vortex doesn't support output ordering");	×
319	}	180✔
320
321	let source = VortexSource::new(self.file_cache.clone(), self.session.metrics().clone());	180✔
322	Ok(DataSourceExec::from_data_source(	180✔
323	FileScanConfigBuilder::from(file_scan_config)	180✔
324	.with_source(Arc::new(source))	180✔
325	.build(),	180✔
326	))	180✔
327	}	360✔
328
329	async fn create_writer_physical_plan(
330	&self,
331	input: Arc<dyn ExecutionPlan>,
332	_state: &dyn Session,
333	conf: FileSinkConfig,
334	order_requirements: Option<LexRequirement>,
335	) -> DFResult<Arc<dyn ExecutionPlan>> {	4✔
336	if conf.insert_op != InsertOp::Append {	2✔
UNCOV 337	return not_impl_err!("Overwrites are not implemented yet for Vortex");	×
338	}	2✔
339
340	if !conf.table_partition_cols.is_empty() {	2✔
UNCOV 341	return not_impl_err!("Hive style partitioning isn't implemented yet for Vortex");	×
342	}	2✔
343
344	let schema = conf.output_schema().clone();	2✔
345	let sink = Arc::new(VortexSink::new(conf, schema));	2✔
346
347	Ok(Arc::new(DataSinkExec::new(input, sink, order_requirements)) as _)	2✔
348	}	4✔
349
350	fn file_source(&self) -> Arc<dyn FileSource> {	180✔
351	Arc::new(VortexSource::new(	180✔
352	self.file_cache.clone(),	180✔
353	VortexMetrics::default(),	180✔
354	))	180✔
355	}	180✔
356	}
357
358	#[cfg(test)]
359	mod tests {
360	use datafusion::execution::SessionStateBuilder;
361	use datafusion::prelude::SessionContext;
362	use tempfile::TempDir;
363
364	use super::*;
365	use crate::persistent::register_vortex_format_factory;
366
367	#[tokio::test]
368	async fn create_table() {	1✔
369	let dir = TempDir::new().unwrap();	1✔
370
371	let factory: VortexFormatFactory = Default::default();	1✔
372	let mut session_state_builder = SessionStateBuilder::new().with_default_features();	1✔
373	register_vortex_format_factory(factory, &mut session_state_builder);	1✔
374	let session = SessionContext::new_with_state(session_state_builder.build());	1✔
375
376	let df = session	1✔
377	.sql(&format!(	1✔
378	"CREATE EXTERNAL TABLE my_tbl \	1✔
379	(c1 VARCHAR NOT NULL, c2 INT NOT NULL) \	1✔
380	STORED AS vortex LOCATION '{}'",	1✔
381	dir.path().to_str().unwrap()	1✔
382	))	1✔
383	.await	1✔
384	.unwrap();	1✔
385
386	assert_eq!(df.count().await.unwrap(), 0);	1✔
387	}	1✔
388
389	#[tokio::test]
390	#[should_panic]
391	async fn fail_table_config() {	1✔
392	let dir = TempDir::new().unwrap();	1✔
393
394	let factory: VortexFormatFactory = Default::default();	1✔
395	let mut session_state_builder = SessionStateBuilder::new().with_default_features();	1✔
396	register_vortex_format_factory(factory, &mut session_state_builder);	1✔
397	let session = SessionContext::new_with_state(session_state_builder.build());	1✔
398
399	session	1✔
400	.sql(&format!(	1✔
401	"CREATE EXTERNAL TABLE my_tbl \	1✔
402	(c1 VARCHAR NOT NULL, c2 INT NOT NULL) \	1✔
403	STORED AS vortex LOCATION '{}' \	1✔
404	OPTIONS( some_key 'value' );",	1✔
405	dir.path().to_str().unwrap()	1✔
406	))	1✔
407	.await	1✔
408	.unwrap()	1✔
409	.collect()	1✔
410	.await	1✔
411	.unwrap();	1✔
412	}	1✔
413	}

vortex-data / vortex / 16421522964

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous