[Draft v2] Another Multi group by optimization #10976
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,43 +1,3 @@ | ||
| SELECT "UserID", "SearchPhrase", COUNT(*) FROM hits GROUP BY "UserID", "SearchPhrase" ORDER BY COUNT(*) DESC LIMIT 10; | ||
| SELECT "UserID", "SearchPhrase", COUNT(*) FROM hits GROUP BY "UserID", "SearchPhrase" LIMIT 10; | ||
| SELECT "UserID", concat("SearchPhrase", repeat('hello', 100)) as s, COUNT(*) FROM hits GROUP BY "UserID", s LIMIT 10; |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -15,17 +15,23 @@ | |
| // specific language governing permissions and limitations | ||
| // under the License. | ||
|
|
||
| use std::sync::Arc; | ||
|
|
||
| use crate::aggregates::group_values::GroupValues; | ||
| use ahash::RandomState; | ||
| use arrow::array::{AsArray, StringBuilder}; | ||
| use arrow::compute::cast; | ||
| use arrow::datatypes::UInt64Type; | ||
| use arrow::record_batch::RecordBatch; | ||
| use arrow::row::{RowConverter, Rows, SortField}; | ||
| use arrow_array::{Array, ArrayRef, UInt64Array}; | ||
| use arrow_schema::{DataType, SchemaRef}; | ||
| use datafusion_common::hash_utils::create_hashes; | ||
| use datafusion_common::{DataFusionError, Result}; | ||
| use datafusion_execution::memory_pool::proxy::{RawTableAllocExt, VecAllocExt}; | ||
| use datafusion_expr::EmitTo; | ||
| use datafusion_physical_expr::binary_map::OutputType; | ||
| use datafusion_physical_expr_common::binary_map::ArrowBytesMap; | ||
| use hashbrown::raw::RawTable; | ||
|
|
||
| /// A [`GroupValues`] making use of [`Rows`] | ||
|
|
@@ -67,6 +73,10 @@ pub struct GroupValuesRows { | |
|
|
||
| /// Random state for creating hashes | ||
| random_state: RandomState, | ||
|
|
||
| // variable length column map | ||
| var_map: ArrowBytesMap<i32, usize>, | ||
| num_groups: usize, | ||
| } | ||
|
|
||
| impl GroupValuesRows { | ||
|
|
@@ -75,7 +85,13 @@ impl GroupValuesRows { | |
| schema | ||
| .fields() | ||
| .iter() | ||
| .map(|f| { | ||
| if f.data_type() == &DataType::Utf8 { | ||
| SortField::new(DataType::UInt64) | ||
| } else { | ||
| SortField::new(f.data_type().clone()) | ||
| } | ||
| }) | ||
| .collect(), | ||
| )?; | ||
|
|
||
|
|
@@ -94,16 +110,41 @@ impl GroupValuesRows { | |
| hashes_buffer: Default::default(), | ||
| rows_buffer, | ||
| random_state: Default::default(), | ||
| var_map: ArrowBytesMap::new(OutputType::Utf8), | ||
| num_groups: 0, | ||
| }) | ||
| } | ||
| } | ||
|
|
||
| impl GroupValues for GroupValuesRows { | ||
| fn intern(&mut self, cols: &[ArrayRef], groups: &mut Vec<usize>) -> Result<()> { | ||
| let cols_var = &cols[1]; | ||
| groups.clear(); | ||
| self.var_map.insert_if_new( | ||
| cols_var, | ||
| // called for each new group | ||
| |_value| { | ||
| // assign new group index on each insert | ||
| let group_idx = self.num_groups; | ||
| self.num_groups += 1; | ||
| group_idx | ||
| }, | ||
| // called for each group | ||
| |group_idx| { | ||
| groups.push(group_idx); | ||
| }, | ||
| ); | ||
|
|
||
| let u64_vec: Vec<u64> = groups.iter().map(|&x| x as u64).collect(); | ||
| let arr = Arc::new(UInt64Array::from(u64_vec)) as ArrayRef; | ||
|
|
||
| let mut cols = cols[0..1].to_vec(); | ||
| cols.push(arr); | ||
|
|
||
| // Convert the group keys into the row format | ||
| let group_rows = &mut self.rows_buffer; | ||
| group_rows.clear(); | ||
| self.row_converter.append(group_rows, &cols)?; | ||
| let n_rows = group_rows.num_rows(); | ||
|
|
||
| let mut group_values = match self.group_values.take() { | ||
|
|
@@ -118,7 +159,7 @@ impl GroupValues for GroupValuesRows { | |
| let batch_hashes = &mut self.hashes_buffer; | ||
| batch_hashes.clear(); | ||
| batch_hashes.resize(n_rows, 0); | ||
| create_hashes(&cols, &self.random_state, batch_hashes)?; | ||
|
|
||
| for (row, &hash) in batch_hashes.iter().enumerate() { | ||
| let entry = self.map.get_mut(hash, |(_hash, group_idx)| { | ||
|
|
@@ -180,15 +221,39 @@ impl GroupValues for GroupValuesRows { | |
| .take() | ||
| .expect("Can not emit from empty rows"); | ||
|
|
||
| let map_contents = self.var_map.take().into_state(); | ||
| let map_contents = map_contents.as_string::<i32>(); | ||
|
|
||
| let mut output = match emit_to { | ||
| EmitTo::All => { | ||
| let mut output = self.row_converter.convert_rows(&group_values)?; | ||
|
|
||
| // Index Array: [0, 1, 1, 0] | ||
| // Data Array: ['a', 'c'] | ||
|
There was a problem hiding this comment. This code copies all the strings again, which likely takes significant time One way to make this faster is likey to special case the "has no nulls" path (so we can avoid The only other ways I can think to make this faster is to avoid copying the strings all together. The only way to do so I can figure are:
🤔 Maybe we can try to hack in the ability to have the HashAggregateExec return If so then we can figure out how to thread that ability through the engine. |
||
| // Result Array: ['a', 'c', 'c', 'a'] | ||
|
|
||
| let mut string_build = StringBuilder::new(); | ||
|
|
||
| let arr = output[1].as_primitive::<UInt64Type>(); | ||
| for v in arr.iter() { | ||
| if let Some(index) = v { | ||
| let value = map_contents.value(index as usize); | ||
| string_build.append_value(value) | ||
| } else { | ||
| string_build.append_null(); | ||
| } | ||
| } | ||
|
|
||
| let output_str = string_build.finish(); | ||
|
|
||
| output[1] = Arc::new(output_str); | ||
|
|
||
| group_values.clear(); | ||
| output | ||
| } | ||
| EmitTo::First(n) => { | ||
| let groups_rows = group_values.iter().take(n); | ||
| let mut output = self.row_converter.convert_rows(groups_rows)?; | ||
| // Clear out first n group keys by copying them to a new Rows. | ||
| // TODO file some ticket in arrow-rs to make this more efficent? | ||
| let mut new_group_values = self.row_converter.empty_rows(0, 0); | ||
|
|
@@ -209,6 +274,25 @@ impl GroupValues for GroupValuesRows { | |
| } | ||
| } | ||
| } | ||
|
|
||
| let mut string_build = StringBuilder::new(); | ||
|
|
||
| let arr = output[1].as_primitive::<UInt64Type>(); | ||
| for v in arr.iter() { | ||
| if let Some(index) = v { | ||
| let value = map_contents.value(index as usize); | ||
| string_build.append_value(value) | ||
| } else { | ||
| string_build.append_null(); | ||
| } | ||
| } | ||
|
|
||
| let output_str = string_build.finish(); | ||
|
|
||
| output[1] = Arc::new(output_str); | ||
|
|
||
| // self.num_groups -= map_contents.len(); | ||
| // output.push(map_contents); | ||
| output | ||
| } | ||
| }; | ||
|
|
||
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,29 @@ | ||
| statement ok | ||
| create table t(a int, b varchar, c int) as values (1, 'a', 3), (2, 'abcabcabcabcabc', 1), (1, 'c', 2), (1, 'a', 4); | ||
|
|
||
| query ITI | ||
| select a, concat(b, 'abcdabcdabcd') as b_v, count(*) from t group by a, b_v order by count(*) desc; | ||
| ---- | ||
| 1 a 2 | ||
| 2 c 1 | ||
| 1 c 1 | ||
|
|
||
| query TT | ||
| explain select a, b, count(*) from t group by a, b order by count(*) desc; | ||
| ---- | ||
| logical_plan | ||
| 01)Sort: COUNT(*) DESC NULLS FIRST | ||
| 02)--Aggregate: groupBy=[[t.a, t.b]], aggr=[[COUNT(Int64(1)) AS COUNT(*)]] | ||
| 03)----TableScan: t projection=[a, b] | ||
| physical_plan | ||
| 01)SortPreservingMergeExec: [COUNT(*)@2 DESC] | ||
| 02)--SortExec: expr=[COUNT(*)@2 DESC], preserve_partitioning=[true] | ||
| 03)----AggregateExec: mode=FinalPartitioned, gby=[a@0 as a, b@1 as b], aggr=[COUNT(*)] | ||
| 04)------CoalesceBatchesExec: target_batch_size=8192 | ||
| 05)--------RepartitionExec: partitioning=Hash([a@0, b@1], 4), input_partitions=4 | ||
| 06)----------RepartitionExec: partitioning=RoundRobinBatch(4), input_partitions=1 | ||
| 07)------------AggregateExec: mode=Partial, gby=[a@0 as a, b@1 as b], aggr=[COUNT(*)] | ||
| 08)--------------MemoryExec: partitions=1, partition_sizes=[1] | ||
|
|
||
| statement ok | ||
| drop table t; |
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It might be faster to write into
u64_vecdirectly rather than write to groups and then copy over