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

Add ParquetMetadata::memory_size size estimation #5965

Merged
merged 3 commits into from
Jul 2, 2024
Merged
Show file tree
Hide file tree
Changes from 1 commit
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
18 changes: 18 additions & 0 deletions parquet/src/data_type.rs
Original file line number Diff line number Diff line change
Expand Up @@ -654,6 +654,13 @@ pub(crate) mod private {

/// Return the value as an mutable Any to allow for downcasts without transmutation
fn as_mut_any(&mut self) -> &mut dyn std::any::Any;

/// Returns the number of bytes of memory this instance uses on the heap.
alamb marked this conversation as resolved.
Show resolved Hide resolved
///
/// Defaults to none (0)
fn heap_size(&self) -> usize {
0
}
}

impl ParquetValueType for bool {
Expand Down Expand Up @@ -968,6 +975,13 @@ pub(crate) mod private {
fn as_mut_any(&mut self) -> &mut dyn std::any::Any {
self
}

fn heap_size(&self) -> usize {
// note: this is an estimate, not exact, so just return the size
// of the actual data used, don't try to handle the fact that it may
// be shared.
self.data.as_ref().map(|data| data.len()).unwrap_or(0)
}
}

impl ParquetValueType for super::FixedLenByteArray {
Expand Down Expand Up @@ -1054,6 +1068,10 @@ pub(crate) mod private {
fn as_mut_any(&mut self) -> &mut dyn std::any::Any {
self
}

fn heap_size(&self) -> usize {
self.0.heap_size()
}
}
}

Expand Down
209 changes: 209 additions & 0 deletions parquet/src/file/metadata/memory.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,209 @@
// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements. See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership. The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License. You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, either express or implied. See the License for the
// specific language governing permissions and limitations
// under the License.

//! Memory calculations for [`ParquetMetadata::memory_size`]
//!
//! [`ParquetMetadata::memory_size`]: crate::file::metadata::ParquetMetaData::memory_size
use crate::basic::{ColumnOrder, Compression, Encoding, PageType};
use crate::data_type::private::ParquetValueType;
use crate::file::metadata::{ColumnChunkMetaData, FileMetaData, KeyValue, RowGroupMetaData};
use crate::file::page_encoding_stats::PageEncodingStats;
use crate::file::page_index::index::{Index, NativeIndex, PageIndex};
use crate::file::statistics::{Statistics, ValueStatistics};
use crate::format::{BoundaryOrder, PageLocation, SortingColumn};
use std::sync::Arc;

/// Trait for calculating the size of various containers
Copy link
Contributor Author

@alamb alamb Jun 26, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I chose to add as many memory estimation calculations as possible in their own module rather than sprinkling it next to the structure definitions. I put it next to the structures when the internal fields are private.

I can put the code next to the definitions if people think that would be cleaner / less likely to be forgotten if new fields are added in the future

pub(crate) trait HeapSize {
/// Return the size of any bytes allocated on the heap by this object,
/// including heap memory in those structures
///
/// Note that the size of the type itself is not included in the result --
/// instead, that size is added by the caller (e.g. container).
fn heap_size(&self) -> usize;
}

impl<T: HeapSize> HeapSize for Vec<T> {
fn heap_size(&self) -> usize {
let item_size = std::mem::size_of::<T>();
// account for the contents of the Vec
(self.capacity() * item_size) +
// add any heap allocations by contents
self.iter().map(|t| t.heap_size()).sum::<usize>()
}
}

impl<T: HeapSize> HeapSize for Arc<T> {
fn heap_size(&self) -> usize {
self.as_ref().heap_size()
}
}

impl<T: HeapSize> HeapSize for Option<T> {
fn heap_size(&self) -> usize {
self.as_ref().map(|inner| inner.heap_size()).unwrap_or(0)
}
}

impl HeapSize for String {
fn heap_size(&self) -> usize {
self.capacity()
}
}

impl HeapSize for FileMetaData {
fn heap_size(&self) -> usize {
self.created_by.heap_size()
+ self.key_value_metadata.heap_size()
+ self.schema_descr.heap_size()
+ self.column_orders.heap_size()
}
}

impl HeapSize for KeyValue {
fn heap_size(&self) -> usize {
self.key.heap_size() + self.value.heap_size()
}
}

impl HeapSize for RowGroupMetaData {
fn heap_size(&self) -> usize {
// don't count schema_descr here because it is already
// counted in FileMetaData
self.columns.heap_size() + self.sorting_columns.heap_size()
}
}

impl HeapSize for ColumnChunkMetaData {
fn heap_size(&self) -> usize {
// don't count column_descr here because it is already counted in
// FileMetaData
self.encodings.heap_size()
+ self.file_path.heap_size()
+ self.compression.heap_size()
+ self.statistics.heap_size()
+ self.encoding_stats.heap_size()
}
}

impl HeapSize for Encoding {
fn heap_size(&self) -> usize {
0 // no heap allocations
}
}

impl HeapSize for PageEncodingStats {
fn heap_size(&self) -> usize {
self.page_type.heap_size() + self.encoding.heap_size()
}
}

impl HeapSize for SortingColumn {
fn heap_size(&self) -> usize {
0 // no heap allocations
}
}
impl HeapSize for Compression {
fn heap_size(&self) -> usize {
0 // no heap allocations
}
}

impl HeapSize for PageType {
fn heap_size(&self) -> usize {
0 // no heap allocations
}
}
impl HeapSize for Statistics {
fn heap_size(&self) -> usize {
match self {
Statistics::Boolean(value_statistics) => value_statistics.heap_size(),
Statistics::Int32(value_statistics) => value_statistics.heap_size(),
Statistics::Int64(value_statistics) => value_statistics.heap_size(),
Statistics::Int96(value_statistics) => value_statistics.heap_size(),
Statistics::Float(value_statistics) => value_statistics.heap_size(),
Statistics::Double(value_statistics) => value_statistics.heap_size(),
Statistics::ByteArray(value_statistics) => value_statistics.heap_size(),
Statistics::FixedLenByteArray(value_statistics) => value_statistics.heap_size(),
}
}
}

impl HeapSize for Index {
fn heap_size(&self) -> usize {
match self {
Index::NONE => 0,
Index::BOOLEAN(native_index) => native_index.heap_size(),
Index::INT32(native_index) => native_index.heap_size(),
Index::INT64(native_index) => native_index.heap_size(),
Index::INT96(native_index) => native_index.heap_size(),
Index::FLOAT(native_index) => native_index.heap_size(),
Index::DOUBLE(native_index) => native_index.heap_size(),
Index::BYTE_ARRAY(native_index) => native_index.heap_size(),
Index::FIXED_LEN_BYTE_ARRAY(native_index) => native_index.heap_size(),
}
}
}

impl<T: ParquetValueType> HeapSize for NativeIndex<T> {
fn heap_size(&self) -> usize {
self.indexes.heap_size() + self.boundary_order.heap_size()
}
}

impl<T: ParquetValueType> HeapSize for PageIndex<T> {
fn heap_size(&self) -> usize {
self.min.heap_size() + self.max.heap_size() + self.null_count.heap_size()
}
}

impl<T: ParquetValueType> HeapSize for ValueStatistics<T> {
fn heap_size(&self) -> usize {
self.min().heap_size() + self.max().heap_size()
}
}

// Note this impl gets most primitive types like bool, i32, etc
impl<T: ParquetValueType> HeapSize for T {
fn heap_size(&self) -> usize {
self.heap_size()
}
}

impl HeapSize for usize {
fn heap_size(&self) -> usize {
0 // no heap allocations
}
}

impl HeapSize for BoundaryOrder {
fn heap_size(&self) -> usize {
0 // no heap allocations
}
}

impl HeapSize for PageLocation {
fn heap_size(&self) -> usize {
0 // no heap allocations
}
}

impl HeapSize for ColumnOrder {
fn heap_size(&self) -> usize {
0 // no heap allocations in ColumnOrder
}
}
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,8 @@
//! * [`ColumnChunkMetaData`]: Metadata for each column chunk (primitive leaf)
//! within a Row Group including encoding and compression information,
//! number of values, statistics, etc.
mod memory;

use std::ops::Range;
use std::sync::Arc;

Expand All @@ -39,6 +41,7 @@ use crate::format::{

use crate::basic::{ColumnOrder, Compression, Encoding, Type};
use crate::errors::{ParquetError, Result};
pub(crate) use crate::file::metadata::memory::HeapSize;
use crate::file::page_encoding_stats::{self, PageEncodingStats};
use crate::file::page_index::index::Index;
use crate::file::statistics::{self, Statistics};
Expand Down Expand Up @@ -176,6 +179,28 @@ impl ParquetMetaData {
self.offset_index.as_ref()
}

/// Estimate of the bytes allocated to store `ParquetMetadata`
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here is the (only) new API introduced in this PR

///
/// # Notes:
///
/// 1. Includes size of self
///
/// 2. Includes heap memory for sub fields such as [`FileMetaData`] and
/// [`RowGroupMetaData`].
///
/// 3. Includes memory from shared pointers (e.g. [`SchemaDescPtr`]). This
/// means `memory_size` will over estimate the memory size if such pointers
/// are shared.
///
/// 4. Does not include any allocator overheads
pub fn memory_size(&self) -> usize {
std::mem::size_of::<Self>()
+ self.file_metadata.heap_size()
+ self.row_groups.heap_size()
+ self.column_index.heap_size()
+ self.offset_index.heap_size()
}

/// Override the column index
#[cfg(feature = "arrow")]
pub(crate) fn set_column_index(&mut self, index: Option<ParquetColumnIndex>) {
Expand Down Expand Up @@ -1034,7 +1059,8 @@ impl OffsetIndexBuilder {
#[cfg(test)]
mod tests {
use super::*;
use crate::basic::PageType;
use crate::basic::{PageType, SortOrder};
use crate::file::page_index::index::NativeIndex;

#[test]
fn test_row_group_metadata_thrift_conversion() {
Expand Down Expand Up @@ -1227,6 +1253,68 @@ mod tests {
assert_eq!(compressed_size_res, compressed_size_exp);
}

#[test]
fn test_memory_size() {
let schema_descr = get_test_schema_descr();

let columns = schema_descr
.columns()
.iter()
.map(|column_descr| ColumnChunkMetaData::builder(column_descr.clone()).build())
.collect::<Result<Vec<_>>>()
.unwrap();
let row_group_meta = RowGroupMetaData::builder(schema_descr.clone())
.set_num_rows(1000)
.set_column_metadata(columns)
.build()
.unwrap();
let row_group_meta = vec![row_group_meta];

let version = 2;
let num_rows = 1000;
let created_by = Some(String::from("test harness"));
let key_value_metadata = Some(vec![KeyValue::new(
String::from("Foo"),
Some(String::from("bar")),
)]);
let column_orders = Some(vec![
ColumnOrder::UNDEFINED,
ColumnOrder::TYPE_DEFINED_ORDER(SortOrder::UNSIGNED),
]);
let file_metadata = FileMetaData::new(
version,
num_rows,
created_by,
key_value_metadata,
schema_descr,
column_orders,
);
let parquet_meta = ParquetMetaData::new(file_metadata.clone(), row_group_meta.clone());
let base_expected_size = 1320;
assert_eq!(parquet_meta.memory_size(), base_expected_size);

let mut column_index = ColumnIndexBuilder::new();
column_index.append(false, vec![1u8], vec![2u8, 3u8], 4);
let column_index = column_index.build_to_thrift();
let native_index = NativeIndex::<bool>::try_new(column_index).unwrap();

// Now, add in OffsetIndex
let parquet_meta = ParquetMetaData::new_with_page_index(
file_metadata,
row_group_meta,
Some(vec![vec![Index::BOOLEAN(native_index)]]),
Some(vec![vec![
vec![PageLocation::new(1, 2, 3)],
vec![PageLocation::new(1, 2, 3)],
]]),
);

let bigger_expected_size = 2304;
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this shows there is non trivial overhead with storing these structures -- 2K already for a 2 column with a single row group

// more set fields means more memory usage
assert!(bigger_expected_size > base_expected_size);
assert_eq!(parquet_meta.memory_size(), bigger_expected_size);
}

/// Returns sample schema descriptor so we can create column metadata.
fn get_test_schema_descr() -> SchemaDescPtr {
let schema = SchemaType::group_type_builder("schema")
Expand Down
Loading
Loading