Traverse an index in the leaf page block list order and send records to adapter. More...

#include <row0pread-adapter.h>

Classes
struct	MySQL_row
	MySQL row meta data. More...

struct	Thread_ctx
	Callback thread context for each of the spawned threads. More...

Public Types
using	Load_fn = handler::Load_cbk

using	End_fn = handler::Load_end_cbk

using	Init_fn = handler::Load_init_cbk

Public Member Functions
	Parallel_reader_adapter (size_t max_threads, ulint rowlen, bool use_reserved_threads=false)
	Constructor. More...

size_t	max_threads () const

	~Parallel_reader_adapter ()=default
	Destructor. More...

dberr_t	add_scan (trx_t *trx, const Parallel_reader::Config &config, Parallel_reader::F &&f)
	Add scan context. More...

dberr_t	run (void **thread_contexts, Init_fn init_fn, Load_fn load_fn, End_fn end_fn)
	Run the parallel scan. More...

dberr_t	process_rows (const Parallel_reader::Ctx *reader_ctx)
	Convert the record in InnoDB format to MySQL format and send them. More...

void	set (row_prebuilt_t *prebuilt)
	Set up the query processing state cache. More...

Private Member Functions
dberr_t	init (Parallel_reader::Thread_ctx reader_thread_ctx, row_prebuilt_t prebuilt)
	Each parallel reader thread's init function. More...

dberr_t	end (Parallel_reader::Thread_ctx *reader_thread_ctx)
	Each parallel reader thread's end function. More...

dberr_t	send_batch (Parallel_reader::Thread_ctx *reader_thread_ctx, size_t partition_id, uint64_t n_recs)
	Send a batch of records. More...

size_t	pending (Thread_ctx *ctx) const
	Get the number of rows buffered but not sent. More...

bool	is_buffer_full (Thread_ctx *ctx) const
	Check if the buffer is full. More...

Private Attributes
void **	m_thread_ctxs {}
	Adapter context for each of the spawned threads. More...

Init_fn	m_init_fn {}
	Callback called by each parallel load thread at the beginning of the parallel load for the scan. More...

Load_fn	m_load_fn {}
	Callback called by each parallel load thread when processing of rows is required for the scan. More...

End_fn	m_end_fn {}
	Callback called by each parallel load thread when processing of rows has ended for the scan. More...

uint64_t	m_batch_size {}
	Number of records to be sent across to the caller in a batch. More...

MySQL_row	m_mysql_row {}
	Row meta data per scan context. More...

row_prebuilt_t *	m_prebuilt {}
	Prebuilt to use for conversion to MySQL row format. More...

Parallel_reader	m_parallel_reader
	Parallel reader to use. More...

Static Private Attributes
static constexpr size_t	ADAPTER_SEND_BUFFER_SIZE = 2 * 1024 * 1024
	Size of the buffer used to store InnoDB records and sent to the adapter. More...

Detailed Description

Traverse an index in the leaf page block list order and send records to adapter.

Member Typedef Documentation

◆ End_fn

using Parallel_reader_adapter::End_fn = handler::Load_end_cbk

◆ Init_fn

using Parallel_reader_adapter::Init_fn = handler::Load_init_cbk

◆ Load_fn

using Parallel_reader_adapter::Load_fn = handler::Load_cbk

Constructor & Destructor Documentation

◆ Parallel_reader_adapter()

Parallel_reader_adapter::Parallel_reader_adapter	(	size_t	max_threads,
		ulint	rowlen,
		bool	use_reserved_threads = `false`
	)

Constructor.

Parameters

[in]	max_threads	Maximum threads to use for all scan contexts.
[in]	rowlen	Row length.
[in]	use_reserved_threads	Whether to use reserved parallel-read slots.

◆ ~Parallel_reader_adapter()

Parallel_reader_adapter::~Parallel_reader_adapter ( )

default

Destructor.

Member Function Documentation

◆ add_scan()

dberr_t Parallel_reader_adapter::add_scan	(	trx_t *	trx,
		const Parallel_reader::Config &	config,
		Parallel_reader::F &&	f
	)

Add scan context.

Parameters

[in]	trx	Transaction used for parallel read.
[in]	config	(Cluster) Index scan configuration.
[in]	f	Callback function.

Returns: error.

◆ end()

dberr_t Parallel_reader_adapter::end ( Parallel_reader::Thread_ctx * reader_thread_ctx )

private

Each parallel reader thread's end function.

Parameters

[in] reader_thread_ctx context info related to the current thread

Returns: DB_SUCCESS or error code.

◆ init()

dberr_t Parallel_reader_adapter::init	(	Parallel_reader::Thread_ctx *	reader_thread_ctx,
		row_prebuilt_t *	prebuilt
	)

private

Each parallel reader thread's init function.

Parameters

[in]	reader_thread_ctx	context info related to the current thread
[in]	prebuilt	prebuilt cache

Returns: DB_SUCCESS or error code.

There are data members in row_prebuilt_t that cannot be accessed in multi-threaded mode e.g., blob_heap.

row_prebuilt_t is designed for single threaded access and to share it among threads is not recommended unless "you know what you are doing". This is very fragile code as it stands.

To solve the blob heap issue in prebuilt we request parallel reader thread to use blob heap per thread and we pass this blob heap to the InnoDB to MySQL row format conversion function.

◆ is_buffer_full()

bool Parallel_reader_adapter::is_buffer_full ( Thread_ctx * ctx ) const

inlineprivate

Check if the buffer is full.

Parameters

[in] ctx adapter related thread context information.

Returns: true if the buffer is full.

◆ max_threads()

size_t Parallel_reader_adapter::max_threads ( ) const

inline

Returns: the number of global parallel-read slots reserved.

◆ pending()

size_t Parallel_reader_adapter::pending ( Thread_ctx * ctx ) const

inlineprivate

Get the number of rows buffered but not sent.

Parameters

[in] ctx adapter related thread context information.

Returns: number of buffered items.

◆ process_rows()

dberr_t Parallel_reader_adapter::process_rows ( const Parallel_reader::Ctx * reader_ctx )

Convert the record in InnoDB format to MySQL format and send them.

Parameters

[in] reader_ctx Parallel read context.

Returns: error code

◆ run()

dberr_t Parallel_reader_adapter::run	(	void **	thread_contexts,
		Init_fn	init_fn,
		Load_fn	load_fn,
		End_fn	end_fn
	)

Run the parallel scan.

Parameters

[in]	thread_contexts	Context for each of the spawned threads
[in]	init_fn	Callback called by each parallel load thread at the beginning of the parallel load.
[in]	load_fn	Callback called by each parallel load thread when processing of rows is required.
[in]	end_fn	Callback called by each parallel load thread when processing of rows has ended.

Returns: DB_SUCCESS or error code.

◆ send_batch()

dberr_t Parallel_reader_adapter::send_batch	(	Parallel_reader::Thread_ctx *	reader_thread_ctx,
		size_t	partition_id,
		uint64_t	n_recs
	)

private

Send a batch of records.

Parameters

[in]	reader_thread_ctx	reader threads related thread context info
[in]	partition_id	partition ID of the index the record belongs to
[in]	n_recs	Number of records to send.

Returns: DB_SUCCESS or error code.

◆ set()

void Parallel_reader_adapter::set ( row_prebuilt_t * prebuilt )

Set up the query processing state cache.

Parameters

[in] prebuilt The prebuilt cache for the query.

Member Data Documentation

◆ ADAPTER_SEND_BUFFER_SIZE

constexpr size_t Parallel_reader_adapter::ADAPTER_SEND_BUFFER_SIZE = 2 * 1024 * 1024

staticconstexprprivate

Size of the buffer used to store InnoDB records and sent to the adapter.

◆ m_batch_size

uint64_t Parallel_reader_adapter::m_batch_size {}

private

Number of records to be sent across to the caller in a batch.

◆ m_end_fn

End_fn Parallel_reader_adapter::m_end_fn {}

private

Callback called by each parallel load thread when processing of rows has ended for the scan.

◆ m_init_fn

Init_fn Parallel_reader_adapter::m_init_fn {}

private

Callback called by each parallel load thread at the beginning of the parallel load for the scan.

◆ m_load_fn

Load_fn Parallel_reader_adapter::m_load_fn {}

private

Callback called by each parallel load thread when processing of rows is required for the scan.

◆ m_mysql_row

MySQL_row Parallel_reader_adapter::m_mysql_row {}

private

Row meta data per scan context.

◆ m_parallel_reader

Parallel_reader Parallel_reader_adapter::m_parallel_reader

private

Parallel reader to use.

◆ m_prebuilt

row_prebuilt_t* Parallel_reader_adapter::m_prebuilt {}

private

Prebuilt to use for conversion to MySQL row format.

NOTE: We are sharing this because we don't convert BLOBs yet. There are data members in row_prebuilt_t that cannot be accessed in multi-threaded mode e.g., blob_heap.

row_prebuilt_t is designed for single threaded access and to share it among threads is not recommended unless "you know what you are doing". This is very fragile code as it stands.

To solve the blob heap issue in prebuilt we use per thread m_blob_heaps. Pass the blob heap to the InnoDB to MySQL row format conversion function.

◆ m_thread_ctxs

void** Parallel_reader_adapter::m_thread_ctxs {}

private

Adapter context for each of the spawned threads.

We don't know the type of the context it's passed to us as a void *.

The documentation for this class was generated from the following files:

storage/innobase/include/row0pread-adapter.h
storage/innobase/row/row0pread-adapter.cc

Classes

Public Types

Public Member Functions

Private Member Functions

Private Attributes

Static Private Attributes

Detailed Description

Member Typedef Documentation

◆ End_fn

◆ Init_fn

◆ Load_fn

Constructor & Destructor Documentation

◆ Parallel_reader_adapter()

◆ ~Parallel_reader_adapter()

Member Function Documentation

◆ add_scan()

◆ end()

◆ init()

◆ is_buffer_full()

◆ max_threads()

◆ pending()

◆ process_rows()

◆ run()

◆ send_batch()

◆ set()

Member Data Documentation

◆ ADAPTER_SEND_BUFFER_SIZE

◆ m_batch_size

◆ m_end_fn

◆ m_init_fn

◆ m_load_fn

◆ m_mysql_row

◆ m_parallel_reader

◆ m_prebuilt

◆ m_thread_ctxs