Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat(query): push rank limit into aggregate partial node #16466

Merged
merged 7 commits into from
Sep 19, 2024
Merged
Show file tree
Hide file tree
Changes from 4 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions src/query/expression/src/kernels/sort.rs
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,12 @@ pub enum LimitType {
}

impl LimitType {
pub fn from_limit_rows(limit: Option<usize>) -> Self {
match limit {
Some(limit) => LimitType::LimitRows(limit),
None => LimitType::None,
}
}
pub fn limit_rows(&self, rows: usize) -> usize {
match self {
LimitType::LimitRows(limit) => *limit,
Expand Down
10 changes: 7 additions & 3 deletions src/query/expression/src/kernels/sort_compare.rs
Original file line number Diff line number Diff line change
Expand Up @@ -98,7 +98,7 @@ macro_rules! do_sorter {
impl SortCompare {
pub fn new(ordering_descs: Vec<SortColumnDescription>, rows: usize, limit: LimitType) -> Self {
let equality_index =
if ordering_descs.len() == 1 && matches!(limit, LimitType::LimitRank(_)) {
if ordering_descs.len() == 1 && !matches!(limit, LimitType::LimitRank(_)) {
vec![]
} else {
vec![1; rows as _]
Expand All @@ -114,6 +114,11 @@ impl SortCompare {
}
}

fn need_update_equality_index(&self) -> bool {
self.current_column_index != self.ordering_descs.len() - 1
|| matches!(self.limit, LimitType::LimitRank(_))
}

pub fn increment_column_index(&mut self) {
self.current_column_index += 1;
}
Expand Down Expand Up @@ -196,8 +201,7 @@ impl SortCompare {
} else {
let mut current = 1;
let len = self.rows;
let need_update_equality_index =
self.current_column_index != self.ordering_descs.len() - 1;
let need_update_equality_index = self.need_update_equality_index();

while current < len {
// Find the start of the next range of equal elements
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@ use std::sync::Arc;

use databend_common_exception::Result;
use databend_common_expression::DataBlock;
use databend_common_expression::LimitType;
use databend_common_expression::SortColumnDescription;
use databend_common_pipeline_core::processors::InputPort;
use databend_common_pipeline_core::processors::OutputPort;
Expand All @@ -25,13 +26,13 @@ use crate::processors::transforms::Transform;
use crate::processors::transforms::Transformer;

pub struct TransformSortPartial {
limit: Option<usize>,
limit: LimitType,
sort_columns_descriptions: Arc<Vec<SortColumnDescription>>,
}

impl TransformSortPartial {
pub fn new(
limit: Option<usize>,
limit: LimitType,
sort_columns_descriptions: Arc<Vec<SortColumnDescription>>,
) -> Self {
Self {
Expand All @@ -43,7 +44,7 @@ impl TransformSortPartial {
pub fn try_create(
input: Arc<InputPort>,
output: Arc<OutputPort>,
limit: Option<usize>,
limit: LimitType,
sort_columns_descriptions: Arc<Vec<SortColumnDescription>>,
) -> Result<Box<dyn Processor>> {
Ok(Transformer::create(input, output, TransformSortPartial {
Expand All @@ -58,6 +59,6 @@ impl Transform for TransformSortPartial {
const NAME: &'static str = "SortPartialTransform";

fn transform(&mut self, block: DataBlock) -> Result<DataBlock> {
DataBlock::sort(&block, &self.sort_columns_descriptions, self.limit)
DataBlock::sort_with_type(&block, &self.sort_columns_descriptions, self.limit)
}
}
Original file line number Diff line number Diff line change
Expand Up @@ -298,6 +298,7 @@ fn remove_exchange(plan: PhysicalPlan) -> PhysicalPlan {
input: Box::new(traverse(*plan.input)),
group_by: plan.group_by,
agg_funcs: plan.agg_funcs,
rank_limit: plan.rank_limit,
enable_experimental_aggregate_hashtable: plan
.enable_experimental_aggregate_hashtable,
group_by_display: plan.group_by_display,
Expand All @@ -310,7 +311,6 @@ fn remove_exchange(plan: PhysicalPlan) -> PhysicalPlan {
group_by: plan.group_by,
agg_funcs: plan.agg_funcs,
before_group_by_schema: plan.before_group_by_schema,
limit: plan.limit,
group_by_display: plan.group_by_display,
stat_info: plan.stat_info,
}),
Expand Down
31 changes: 26 additions & 5 deletions src/query/service/src/pipelines/builders/builder_aggregate.rs
Original file line number Diff line number Diff line change
Expand Up @@ -23,10 +23,13 @@ use databend_common_expression::DataBlock;
use databend_common_expression::DataSchemaRef;
use databend_common_expression::HashMethodKind;
use databend_common_expression::HashTableConfig;
use databend_common_expression::LimitType;
use databend_common_expression::SortColumnDescription;
use databend_common_functions::aggregates::AggregateFunctionFactory;
use databend_common_pipeline_core::processors::ProcessorPtr;
use databend_common_pipeline_core::query_spill_prefix;
use databend_common_pipeline_transforms::processors::TransformPipelineHelper;
use databend_common_pipeline_transforms::processors::TransformSortPartial;
use databend_common_sql::executor::physical_plans::AggregateExpand;
use databend_common_sql::executor::physical_plans::AggregateFinal;
use databend_common_sql::executor::physical_plans::AggregateFunctionDesc;
Expand Down Expand Up @@ -111,7 +114,6 @@ impl PipelineBuilder {
enable_experimental_aggregate_hashtable,
self.is_exchange_neighbor,
max_block_size as usize,
None,
max_spill_io_requests as usize,
)?;

Expand All @@ -125,7 +127,7 @@ impl PipelineBuilder {

let group_cols = &params.group_columns;
let schema_before_group_by = params.input_schema.clone();
let sample_block = DataBlock::empty_with_schema(schema_before_group_by);
let sample_block = DataBlock::empty_with_schema(schema_before_group_by.clone());
let method = DataBlock::choose_hash_method(&sample_block, group_cols, efficiently_memory)?;

// Need a global atomic to read the max current radix bits hint
Expand All @@ -136,6 +138,28 @@ impl PipelineBuilder {
.cluster_with_partial(true, self.ctx.get_cluster().nodes.len())
};

// For rank limit, we can filter data using sort with rank before partial
if let Some(rank_limit) = &aggregate.rank_limit {
let sort_desc = rank_limit
.0
.iter()
.map(|desc| {
let offset = schema_before_group_by.index_of(&desc.order_by.to_string())?;
Ok(SortColumnDescription {
offset,
asc: desc.asc,
nulls_first: desc.nulls_first,
is_nullable: schema_before_group_by.field(offset).is_nullable(), // This information is not needed here.
})
})
.collect::<Result<Vec<_>>>()?;
let sort_desc = Arc::new(sort_desc);

self.main_pipeline.add_transformer(|| {
TransformSortPartial::new(LimitType::LimitRank(rank_limit.1), sort_desc.clone())
});
}

self.main_pipeline.add_transform(|input, output| {
Ok(ProcessorPtr::create(
match params.aggregate_functions.is_empty() {
Expand Down Expand Up @@ -225,7 +249,6 @@ impl PipelineBuilder {
enable_experimental_aggregate_hashtable,
self.is_exchange_neighbor,
max_block_size as usize,
aggregate.limit,
max_spill_io_requests as usize,
)?;

Expand Down Expand Up @@ -292,7 +315,6 @@ impl PipelineBuilder {
enable_experimental_aggregate_hashtable: bool,
cluster_aggregator: bool,
max_block_size: usize,
limit: Option<usize>,
max_spill_io_requests: usize,
) -> Result<Arc<AggregatorParams>> {
let mut agg_args = Vec::with_capacity(agg_funcs.len());
Expand Down Expand Up @@ -335,7 +357,6 @@ impl PipelineBuilder {
enable_experimental_aggregate_hashtable,
cluster_aggregator,
max_block_size,
limit,
max_spill_io_requests,
)?;

Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,7 @@ use std::sync::Arc;
use databend_common_catalog::catalog::CatalogManager;
use databend_common_exception::Result;
use databend_common_expression::DataSchema;
use databend_common_expression::LimitType;
use databend_common_expression::SortColumnDescription;
use databend_common_pipeline_core::processors::ProcessorPtr;
use databend_common_pipeline_core::DynTransformBuilder;
Expand Down Expand Up @@ -230,7 +231,7 @@ impl PipelineBuilder {
Ok(ProcessorPtr::create(TransformSortPartial::try_create(
transform_input_port,
transform_output_port,
None,
LimitType::None,
sort_desc.clone(),
)?))
},
Expand Down
8 changes: 7 additions & 1 deletion src/query/service/src/pipelines/builders/builder_sort.rs
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@ use std::sync::Arc;

use databend_common_exception::Result;
use databend_common_expression::DataSchemaRef;
use databend_common_expression::LimitType;
use databend_common_expression::SortColumnDescription;
use databend_common_pipeline_core::processors::ProcessorPtr;
use databend_common_pipeline_core::query_spill_prefix;
Expand Down Expand Up @@ -197,7 +198,12 @@ impl SortPipelineBuilder {

pub fn build_full_sort_pipeline(self, pipeline: &mut Pipeline) -> Result<()> {
// Partial sort
pipeline.add_transformer(|| TransformSortPartial::new(self.limit, self.sort_desc.clone()));
pipeline.add_transformer(|| {
TransformSortPartial::new(
LimitType::from_limit_rows(self.limit),
self.sort_desc.clone(),
)
});

self.build_merge_sort_pipeline(pipeline, false)
}
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -44,8 +44,6 @@ pub struct AggregatorParams {
pub enable_experimental_aggregate_hashtable: bool,
pub cluster_aggregator: bool,
pub max_block_size: usize,
// Limit is push down to AggregatorTransform
pub limit: Option<usize>,
pub max_spill_io_requests: usize,
}

Expand All @@ -59,7 +57,6 @@ impl AggregatorParams {
enable_experimental_aggregate_hashtable: bool,
cluster_aggregator: bool,
max_block_size: usize,
limit: Option<usize>,
max_spill_io_requests: usize,
) -> Result<Arc<AggregatorParams>> {
let mut states_offsets: Vec<usize> = Vec::with_capacity(agg_funcs.len());
Expand All @@ -80,7 +77,6 @@ impl AggregatorParams {
enable_experimental_aggregate_hashtable,
cluster_aggregator,
max_block_size,
limit,
max_spill_io_requests,
}))
}
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,6 @@ pub struct TransformFinalAggregate<Method: HashMethodBounds> {
method: Method,
params: Arc<AggregatorParams>,
flush_state: PayloadFlushState,
reach_limit: bool,
}

impl<Method: HashMethodBounds> TransformFinalAggregate<Method> {
Expand All @@ -63,7 +62,6 @@ impl<Method: HashMethodBounds> TransformFinalAggregate<Method> {
method,
params,
flush_state: PayloadFlushState::default(),
reach_limit: false,
},
))
}
Expand Down Expand Up @@ -124,23 +122,11 @@ impl<Method: HashMethodBounds> TransformFinalAggregate<Method> {
let mut blocks = vec![];
self.flush_state.clear();

let mut rows = 0;
loop {
if ht.merge_result(&mut self.flush_state)? {
let mut cols = self.flush_state.take_aggregate_results();
cols.extend_from_slice(&self.flush_state.take_group_columns());
rows += cols[0].len();
blocks.push(DataBlock::new_from_columns(cols));

if rows >= self.params.limit.unwrap_or(usize::MAX) {
log::info!(
"reach limit optimization in flush agg hashtable, current {}, total {}",
rows,
ht.len(),
);
self.reach_limit = true;
break;
}
} else {
break;
}
Expand All @@ -162,10 +148,6 @@ where Method: HashMethodBounds
const NAME: &'static str = "TransformFinalAggregate";

fn transform(&mut self, meta: AggregateMeta<Method, usize>) -> Result<Vec<DataBlock>> {
if self.reach_limit {
return Ok(vec![self.params.empty_result_block()]);
}

if self.params.enable_experimental_aggregate_hashtable {
return Ok(vec![self.transform_agg_hashtable(meta)?]);
}
Expand Down Expand Up @@ -196,32 +178,15 @@ where Method: HashMethodBounds
let (len, _) = keys_iter.size_hint();
let mut places = Vec::with_capacity(len);

let mut current_len = hash_cell.hashtable.len();
unsafe {
for key in keys_iter {
if self.reach_limit {
let entry = hash_cell.hashtable.entry(key);
if let Some(entry) = entry {
let place = Into::<StateAddr>::into(*entry.get());
places.push(place);
}
continue;
}

match hash_cell.hashtable.insert_and_entry(key) {
Ok(mut entry) => {
let place =
self.params.alloc_layout(&mut hash_cell.arena);
places.push(place);

*entry.get_mut() = place.addr();

if let Some(limit) = self.params.limit {
current_len += 1;
if current_len >= limit {
self.reach_limit = true;
}
}
}
Err(entry) => {
let place = Into::<StateAddr>::into(*entry.get());
Expand Down
Loading
Loading