Skip to content

Commit

Permalink
Fix handling of nested leaf columns in parallel parquet writer (#8923)
Browse files Browse the repository at this point in the history 
* fix handling of nested columns

* lint

* add suggested tests
  • Loading branch information
@devinjdangelo
devinjdangelo authored Jan 22, 2024
1 parent 2b218be commit 38d5f75
Show file tree
Hide file tree
Showing 6 changed files with 71 additions and 20 deletions.
2 changes: 1 addition & 1 deletion datafusion/common/src/config.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -408,7 +408,7 @@ config_namespace! {
/// parquet files by serializing them in parallel. Each column
/// in each row group in each output file are serialized in parallel
/// leveraging a maximum possible core count of n_files*n_row_groups*n_columns.
pub allow_single_file_parallelism: bool, default = false
pub allow_single_file_parallelism: bool, default = true

/// By default parallel parquet writer is tuned for minimum
/// memory usage in a streaming execution plan. You may see
Expand Down
21 changes: 11 additions & 10 deletions datafusion/core/src/datasource/file_format/parquet.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -885,24 +885,25 @@ async fn send_arrays_to_col_writers(
rb: &RecordBatch,
schema: Arc<Schema>,
) -> Result<()> {
for (tx, array, field) in col_array_channels
.iter()
.zip(rb.columns())
.zip(schema.fields())
.map(|((a, b), c)| (a, b, c))
{
// Each leaf column has its own channel, increment next_channel for each leaf column sent.
let mut next_channel = 0;
for (array, field) in rb.columns().iter().zip(schema.fields()) {
for c in compute_leaves(field, array)? {
tx.send(c).await.map_err(|_| {
DataFusionError::Internal("Unable to send array to writer!".into())
})?;
col_array_channels[next_channel]
.send(c)
.await
.map_err(|_| {
DataFusionError::Internal("Unable to send array to writer!".into())
})?;
next_channel += 1;
}
}

Ok(())
}

/// Spawns a tokio task which joins the parallel column writer tasks,
/// and finalizes the row group.
/// and finalizes the row group
fn spawn_rg_join_and_finalize_task(
column_writer_handles: Vec<JoinHandle<Result<ArrowColumnWriter>>>,
rg_rows: usize,
Expand Down
54 changes: 52 additions & 2 deletions datafusion/sqllogictest/test_files/copy.slt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -64,6 +64,24 @@ select * from validate_parquet;
1 Foo
2 Bar

query ?
copy (values (struct(timestamp '2021-01-01 01:00:01', 1)), (struct(timestamp '2022-01-01 01:00:01', 2)),
(struct(timestamp '2023-01-03 01:00:01', 3)), (struct(timestamp '2024-01-01 01:00:01', 4)))
to 'test_files/scratch/copy/table_nested2' (format parquet, single_file_output false);
----
4

statement ok
CREATE EXTERNAL TABLE validate_parquet_nested2 STORED AS PARQUET LOCATION 'test_files/scratch/copy/table_nested2/';

query ?
select * from validate_parquet_nested2;
----
{c0: 2021-01-01T01:00:01, c1: 1}
{c0: 2022-01-01T01:00:01, c1: 2}
{c0: 2023-01-03T01:00:01, c1: 3}
{c0: 2024-01-01T01:00:01, c1: 4}

query ??
COPY
(values (struct ('foo', (struct ('foo', make_array(struct('a',1), struct('b',2))))), make_array(timestamp '2023-01-01 01:00:01',timestamp '2023-01-01 01:00:01')),
Expand All @@ -72,16 +90,48 @@ to 'test_files/scratch/copy/table_nested' (format parquet, single_file_output fa
----
2

# validate multiple parquet file output
statement ok
CREATE EXTERNAL TABLE validate_parquet_nested STORED AS PARQUET LOCATION 'test_files/scratch/copy/table_nested/';
CREATE EXTERNAL TABLE validate_parquet_nested STORED AS PARQUET
LOCATION 'test_files/scratch/copy/table_nested/';

query ??
select * from validate_parquet_nested;
----
{c0: foo, c1: {c0: foo, c1: [{c0: a, c1: 1}, {c0: b, c1: 2}]}} [2023-01-01T01:00:01, 2023-01-01T01:00:01]
{c0: bar, c1: {c0: foo, c1: [{c0: aa, c1: 10}, {c0: bb, c1: 20}]}} [2024-01-01T01:00:01, 2024-01-01T01:00:01]

query ?
copy (values ([struct('foo', 1), struct('bar', 2)]))
to 'test_files/scratch/copy/array_of_struct/'
(format parquet, single_file_output false);
----
1

statement ok
CREATE EXTERNAL TABLE validate_array_of_struct
STORED AS PARQUET LOCATION 'test_files/scratch/copy/array_of_struct/';

query ?
select * from validate_array_of_struct;
----
[{c0: foo, c1: 1}, {c0: bar, c1: 2}]

query ?
copy (values (struct('foo', [1,2,3], struct('bar', [2,3,4]))))
to 'test_files/scratch/copy/struct_with_array/'
(format parquet, single_file_output false);
----
1

statement ok
CREATE EXTERNAL TABLE validate_struct_with_array
STORED AS PARQUET LOCATION 'test_files/scratch/copy/struct_with_array/';

query ?
select * from validate_struct_with_array;
----
{c0: foo, c1: [1, 2, 3], c2: {c0: bar, c1: [2, 3, 4]}}


# Copy parquet with all supported statment overrides
query IT
Expand Down
4 changes: 2 additions & 2 deletions datafusion/sqllogictest/test_files/information_schema.slt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -155,7 +155,7 @@ datafusion.execution.listing_table_ignore_subdirectory true
datafusion.execution.max_buffered_batches_per_output_file 2
datafusion.execution.meta_fetch_concurrency 32
datafusion.execution.minimum_parallel_output_files 4
datafusion.execution.parquet.allow_single_file_parallelism false
datafusion.execution.parquet.allow_single_file_parallelism true
datafusion.execution.parquet.bloom_filter_enabled false
datafusion.execution.parquet.bloom_filter_fpp NULL
datafusion.execution.parquet.bloom_filter_ndv NULL
Expand Down Expand Up @@ -232,7 +232,7 @@ datafusion.execution.listing_table_ignore_subdirectory true Should sub directori
datafusion.execution.max_buffered_batches_per_output_file 2 This is the maximum number of RecordBatches buffered for each output file being worked. Higher values can potentially give faster write performance at the cost of higher peak memory consumption
datafusion.execution.meta_fetch_concurrency 32 Number of files to read in parallel when inferring schema and statistics
datafusion.execution.minimum_parallel_output_files 4 Guarantees a minimum level of output files running in parallel. RecordBatches will be distributed in round robin fashion to each parallel writer. Each writer is closed and a new file opened once soft_max_rows_per_output_file is reached.
datafusion.execution.parquet.allow_single_file_parallelism false Controls whether DataFusion will attempt to speed up writing parquet files by serializing them in parallel. Each column in each row group in each output file are serialized in parallel leveraging a maximum possible core count of n_files*n_row_groups*n_columns.
datafusion.execution.parquet.allow_single_file_parallelism true Controls whether DataFusion will attempt to speed up writing parquet files by serializing them in parallel. Each column in each row group in each output file are serialized in parallel leveraging a maximum possible core count of n_files*n_row_groups*n_columns.
datafusion.execution.parquet.bloom_filter_enabled false Sets if bloom filter is enabled for any column
datafusion.execution.parquet.bloom_filter_fpp NULL Sets bloom filter false positive probability. If NULL, uses default parquet writer setting
datafusion.execution.parquet.bloom_filter_ndv NULL Sets bloom filter number of distinct values. If NULL, uses default parquet writer setting
Expand Down
8 changes: 4 additions & 4 deletions datafusion/sqllogictest/test_files/repartition_scan.slt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -61,7 +61,7 @@ Filter: parquet_table.column1 != Int32(42)
physical_plan
CoalesceBatchesExec: target_batch_size=8192
--FilterExec: column1@0 != 42
----ParquetExec: file_groups={4 groups: [[WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:0..153], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:153..306], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:306..459], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:459..610]]}, projection=[column1], predicate=column1@0 != 42, pruning_predicate=column1_min@0 != 42 OR 42 != column1_max@1, required_guarantees=[column1 not in (42)]
----ParquetExec: file_groups={4 groups: [[WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:0..104], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:104..208], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:208..312], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:312..414]]}, projection=[column1], predicate=column1@0 != 42, pruning_predicate=column1_min@0 != 42 OR 42 != column1_max@1, required_guarantees=[column1 not in (42)]

# disable round robin repartitioning
statement ok
Expand All @@ -77,7 +77,7 @@ Filter: parquet_table.column1 != Int32(42)
physical_plan
CoalesceBatchesExec: target_batch_size=8192
--FilterExec: column1@0 != 42
----ParquetExec: file_groups={4 groups: [[WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:0..153], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:153..306], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:306..459], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:459..610]]}, projection=[column1], predicate=column1@0 != 42, pruning_predicate=column1_min@0 != 42 OR 42 != column1_max@1, required_guarantees=[column1 not in (42)]
----ParquetExec: file_groups={4 groups: [[WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:0..104], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:104..208], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:208..312], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:312..414]]}, projection=[column1], predicate=column1@0 != 42, pruning_predicate=column1_min@0 != 42 OR 42 != column1_max@1, required_guarantees=[column1 not in (42)]

# enable round robin repartitioning again
statement ok
Expand All @@ -102,7 +102,7 @@ SortPreservingMergeExec: [column1@0 ASC NULLS LAST]
--SortExec: expr=[column1@0 ASC NULLS LAST]
----CoalesceBatchesExec: target_batch_size=8192
------FilterExec: column1@0 != 42
--------ParquetExec: file_groups={4 groups: [[WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/1.parquet:0..303], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/1.parquet:303..601, WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:0..5], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:5..308], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:308..610]]}, projection=[column1], predicate=column1@0 != 42, pruning_predicate=column1_min@0 != 42 OR 42 != column1_max@1, required_guarantees=[column1 not in (42)]
--------ParquetExec: file_groups={4 groups: [[WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/1.parquet:0..205], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/1.parquet:205..405, WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:0..5], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:5..210], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:210..414]]}, projection=[column1], predicate=column1@0 != 42, pruning_predicate=column1_min@0 != 42 OR 42 != column1_max@1, required_guarantees=[column1 not in (42)]


## Read the files as though they are ordered
Expand Down Expand Up @@ -138,7 +138,7 @@ physical_plan
SortPreservingMergeExec: [column1@0 ASC NULLS LAST]
--CoalesceBatchesExec: target_batch_size=8192
----FilterExec: column1@0 != 42
------ParquetExec: file_groups={4 groups: [[WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/1.parquet:0..300], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:0..305], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:305..610], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/1.parquet:300..601]]}, projection=[column1], output_ordering=[column1@0 ASC NULLS LAST], predicate=column1@0 != 42, pruning_predicate=column1_min@0 != 42 OR 42 != column1_max@1, required_guarantees=[column1 not in (42)]
------ParquetExec: file_groups={4 groups: [[WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/1.parquet:0..202], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:0..207], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/2.parquet:207..414], [WORKSPACE_ROOT/datafusion/sqllogictest/test_files/scratch/repartition_scan/parquet_table/1.parquet:202..405]]}, projection=[column1], output_ordering=[column1@0 ASC NULLS LAST], predicate=column1@0 != 42, pruning_predicate=column1_min@0 != 42 OR 42 != column1_max@1, required_guarantees=[column1 not in (42)]

# Cleanup
statement ok
Expand Down
2 changes: 1 addition & 1 deletion docs/source/user-guide/configs.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -71,7 +71,7 @@ Environment variables are read during `SessionConfig` initialisation so they mus
| datafusion.execution.parquet.bloom_filter_enabled | false | Sets if bloom filter is enabled for any column |
| datafusion.execution.parquet.bloom_filter_fpp | NULL | Sets bloom filter false positive probability. If NULL, uses default parquet writer setting |
| datafusion.execution.parquet.bloom_filter_ndv | NULL | Sets bloom filter number of distinct values. If NULL, uses default parquet writer setting |
| datafusion.execution.parquet.allow_single_file_parallelism | false | Controls whether DataFusion will attempt to speed up writing parquet files by serializing them in parallel. Each column in each row group in each output file are serialized in parallel leveraging a maximum possible core count of n_files*n_row_groups*n_columns. |
| datafusion.execution.parquet.allow_single_file_parallelism | true | Controls whether DataFusion will attempt to speed up writing parquet files by serializing them in parallel. Each column in each row group in each output file are serialized in parallel leveraging a maximum possible core count of n_files*n_row_groups*n_columns. |
| datafusion.execution.parquet.maximum_parallel_row_group_writers | 1 | By default parallel parquet writer is tuned for minimum memory usage in a streaming execution plan. You may see a performance benefit when writing large parquet files by increasing maximum_parallel_row_group_writers and maximum_buffered_record_batches_per_stream if your system has idle cores and can tolerate additional memory usage. Boosting these values is likely worthwhile when writing out already in-memory data, such as from a cached data frame. |
| datafusion.execution.parquet.maximum_buffered_record_batches_per_stream | 2 | By default parallel parquet writer is tuned for minimum memory usage in a streaming execution plan. You may see a performance benefit when writing large parquet files by increasing maximum_parallel_row_group_writers and maximum_buffered_record_batches_per_stream if your system has idle cores and can tolerate additional memory usage. Boosting these values is likely worthwhile when writing out already in-memory data, such as from a cached data frame. |
| datafusion.execution.aggregate.scalar_update_factor | 10 | Specifies the threshold for using `ScalarValue`s to update accumulators during high-cardinality aggregations for each input batch. The aggregation is considered high-cardinality if the number of affected groups is greater than or equal to `batch_size / scalar_update_factor`. In such cases, `ScalarValue`s are utilized for updating accumulators, rather than the default batch-slice approach. This can lead to performance improvements. By adjusting the `scalar_update_factor`, you can balance the trade-off between more efficient accumulator updates and the number of groups affected. |
Expand Down

0 comments on commit 38d5f75

Please sign in to comment.