Recovery. More...

#include "ha_prototypes.h"
#include <my_aes.h>
#include <sys/types.h>
#include <ostream>
#include <array>
#include <iomanip>
#include <map>
#include <new>
#include <optional>
#include <span>
#include <string>
#include <vector>
#include "arch0arch.h"
#include "btr0btr.h"
#include "btr0cur.h"
#include "buf0buf.h"
#include "buf0flu.h"
#include "clone0api.h"
#include "dict0dd.h"
#include "fil0fil.h"
#include "fil0pages_persistence_interface.h"
#include "ibuf0ibuf.h"
#include "log0chkp.h"
#include "log0encryption.h"
#include "log0files_io.h"
#include "log0log.h"
#include "log0pre_8_0_30.h"
#include "log0recv.h"
#include "log0test.h"
#include "mem0mem.h"
#include "mtr0log.h"
#include "mtr0mtr.h"
#include "os0thread-create.h"
#include "page0cur.h"
#include "page0zip.h"
#include "trx0rec.h"
#include "trx0undo.h"
#include "ut0new.h"
#include "my_dbug.h"
#include "buf0rea.h"
#include "ddl0ddl.h"
#include "log0handler.h"
#include "srv0srv.h"
#include "srv0start.h"
#include "trx0purge.h"

Macros
#define	RECV_DATA_BLOCK_SIZE (MEM_MAX_ALLOC_IN_BUF - sizeof(recv_data_t))
	Log records are stored in the hash table in chunks at most of this size; this must be less than UNIV_PAGE_SIZE as it is stored in the buffer pool. More...

Functions
static bool	recv_writer_is_active ()

static void	recv_init_crash_recovery ()
	Initialize crash recovery environment. More...

lsn_t	recv_calc_lsn_on_data_add (lsn_t lsn, os_offset_t len)
	Calculates the new value for lsn when more data is added to the log. More...

void	recv_sys_create ()
	Creates the recovery system. More...

bool	recv_sys_resize_buf ()
	Resize the recovery parsing buffer up to log_buffer_size. More...

static void	recv_sys_finish ()
	Free up recovery data structures. More...

void	recv_sys_close ()
	Release recovery system mutexes. More...

void	recv_sys_var_init ()
	Reset the state of the recovery system variables. More...

static size_t	recv_heap_used ()
	Get the number of bytes used by all the heaps. More...

static void	recv_report_corrupt_log (std::span< const byte > buffer, size_t pos, lsn_t lsn)
	Prints diagnostic info of corrupt log. More...

void	recv_sys_init ()
	Inits the recovery system for a recovery operation. More...

static void	recv_sys_empty_hash ()
	Empties the hash table when it has been fully processed. More...

static recv_sys_t::Space *	recv_get_page_map (space_id_t space_id, bool create)
	Get the page map for a tablespace. More...

static recv_addr_t *	recv_get_rec (const page_id_t &page_id)
	Gets the list of log records for a <space, page>. More...

static void	recv_writer_thread ()
	recv_writer thread tasked with flushing dirty pages from the buffer pools. More...

void	recv_sys_free ()
	Frees the recovery system. More...

static void	one_less_page_to_recover ()

static void	recv_read_in_area (const page_id_t &requested_page_id, const page_size_t &page_size)
	Reads in pages which have hashed log records, from an area around a given page number. More...

static void	recv_apply_log_rec (recv_addr_t *recv_addr)
	Apply the log records to a page. More...

static void	recv_apply_hashed_log_recs ()
	Empties the hash table of stored log records, applying them to appropriate pages. More...

static void	recv_add_to_hash_table (mlog_id_t type, space_id_t space_id, page_no_t page_no, std::span< const byte > body, lsn_t start_lsn, lsn_t end_lsn)
	Adds a new log record to the hash table of log records. More...

static void	recv_data_copy_to_buf (byte buf, recv_t recv)
	Copies the log record body from recv to buf. More...

bool	recv_page_is_brand_new (buf_block_t *block)
	Returns true if the page is brand new (the next log record is init_file_page or no records to apply). More...

static void	apply_tablespace_side_effects (const ib::redo::Record_handle &record_handle, const ib::redo::Page_handle &page_handle)
	Apply tablespace metadata side-effects after page record application. More...

void	recv_recover_page_func (bool just_read_in, buf_block_t *block)
	Applies the hashed log records to the page, if the page lsn is less than the lsn of a log record. More...

void	recv_track_changes_of_recovered_lsn ()

static bool	recv_process_space_record (const ib::redo::Record_view &rec, lsn_t start_lsn)
	Process a table space record. More...

static bool	recv_process_page_record (const ib::redo::Record_view &rec, lsn_t start_lsn, lsn_t end_lsn)
	Process a page record. More...

static bool	process_mtr (const ib::redo::Mtr_view &mtr, lsn_t start_lsn, lsn_t end_lsn)
	Process an mtr. More...

static std::optional< ib::redo::Mtr_view >	parse_buffer (std::span< const byte > buffer)
	Parses the buffer starting from the beginning. More...

static bool	process_mtr_and_possibly_apply_records (ib::redo::Mtr_view mtr)
	mtr buffer is processed if the start_lsn is more than checkpoint_lsn. More...

dberr_t	recv_parse_and_apply_log_recs (size_t mem_limit)
	Parse log records from a buffer and stores them to a hash table. More...

static void	recv_reset_buffer ()
	Moves the parsing buffer data left to the buffer start. More...

dberr_t	recv_recovery_begin (lsn_t checkpoint_lsn)
	Scans log from a stream and stores new log data to the parsing buffer. More...

dberr_t	recv_recovery_from_checkpoint_start (lsn_t flush_lsn)
	Recovers all tablespaces from the redo log. More...

MetadataRecover *	recv_recovery_from_checkpoint_finish (bool aborting)
	Complete the recovery from the latest checkpoint. More...

const char *	get_mlog_string (mlog_id_t type)
	Return string name of the redo log record type. More...

Variables
static const size_t	RECV_READ_AHEAD_AREA = 32
	Read-ahead area in applying log records to file pages. More...

recv_sys_t *	recv_sys = nullptr
	The recovery system. More...

volatile bool	recv_recovery_on
	true when applying redo log records during crash recovery; false otherwise. More...

static const size_t	PARSING_BUF_MINIMUM_SIZE = 2048

PSI_memory_key	mem_log_recv_space_hash_key

bool	recv_needed_recovery
	true when recv_init_crash_recovery() has been called. More...

bool	recv_lsn_checks_on
	true if buf_page_is_corrupted() should check if the log sequence number (FIL_PAGE_LSN) is in the future. More...

static ulint	recv_scan_print_counter
	The following counter is used to decide when to print info on log scan. More...

size_t	recv_n_frames_for_pages_per_pool_instance
	This many blocks must be left in each Buffer Pool instance to be managed by the LRU when we scan the log and store the scanned log records in a hashmap allocated in the Buffer Pool in frames of non-LRU managed blocks. More...

static lsn_t	recv_max_page_lsn
	The maximum lsn we see for a page during the recovery process. More...

Detailed Description

Recovery.

Created 9/20/1997 Heikki Tuuri

Macro Definition Documentation

◆ RECV_DATA_BLOCK_SIZE

#define RECV_DATA_BLOCK_SIZE (MEM_MAX_ALLOC_IN_BUF - sizeof(recv_data_t))

Log records are stored in the hash table in chunks at most of this size; this must be less than UNIV_PAGE_SIZE as it is stored in the buffer pool.

Function Documentation

◆ apply_tablespace_side_effects()

static void apply_tablespace_side_effects	(	const ib::redo::Record_handle &	record_handle,
		const ib::redo::Page_handle &	page_handle
	)

static

Apply tablespace metadata side-effects after page record application.

This function handles tablespace-level side-effects that occur when applying certain redo log records.

Note: This function should be called immediately after redo_applier->apply()

Parameters

record_handle	Handle containing the redo log record type and body
page_handle	Handle containing the page buffer and metadata after the page modifications have been applied.

◆ get_mlog_string()

const char * get_mlog_string ( mlog_id_t type )

Return string name of the redo log record type.

Parameters

[in] type record log record enum

Returns: string name of record log record

◆ one_less_page_to_recover()

static void one_less_page_to_recover ( )

static

◆ parse_buffer()

static std::optional< ib::redo::Mtr_view > parse_buffer ( std::span< const byte > buffer )

static

Parses the buffer starting from the beginning.

If the buffer content causes parse error, it sets the found_corrupt_log flag to true. The function assumes recovered_lsn is the start_lsn of the mtr.

Parameters

[in] buffer The buffer to parse

Returns: an mtr with parsed log records if parsing was successful, otherwise an empty container that indicates end of the processing. End of the processing may happen either due to a parsing error or, buffer does not start with a full mtr

◆ process_mtr()

static bool process_mtr	(	const ib::redo::Mtr_view &	mtr,
		lsn_t	start_lsn,
		lsn_t	end_lsn
	)

static

Process an mtr.

Parameters

[in]	mtr	mtr to process
[in]	start_lsn	mtr start lsn
[in]	end_lsn	mtr end lsn

Returns: true on success, false otherwise.

◆ process_mtr_and_possibly_apply_records()

static bool process_mtr_and_possibly_apply_records ( ib::redo::Mtr_view mtr )

static

mtr buffer is processed if the start_lsn is more than checkpoint_lsn.

The function assumes recovered_lsn is the start_lsn of the mtr. Processing involves the following two steps :

Optionally applying all earlier records from the hashmap : We follow the order that changes to the space has to happen before to the page because a change which relates to whole space also implicitly applies to each of its pages. As a consequence, changes to the page which have lsn smaller than the change to the space are applied before the change to the space.
Adding page records to hashmap and, additionally executing some actions like tablespace file manipulations, updating tablespace flags, dealing with encryption etc.

Parameters

[in] mtr mtr to process

Returns: true if end of processing

◆ recv_add_to_hash_table()

static void recv_add_to_hash_table	(	mlog_id_t	type,
		space_id_t	space_id,
		page_no_t	page_no,
		std::span< const byte >	body,
		lsn_t	start_lsn,
		lsn_t	end_lsn
	)

static

Adds a new log record to the hash table of log records.

Parameters

[in]	type	log record type
[in]	space_id	Tablespace id
[in]	page_no	page number
[in]	body	log record body
[in]	start_lsn	start lsn of the mtr
[in]	end_lsn	end lsn of the mtr

◆ recv_apply_hashed_log_recs()

static void recv_apply_hashed_log_recs ( )

static

Empties the hash table of stored log records, applying them to appropriate pages.

◆ recv_apply_log_rec()

static void recv_apply_log_rec ( recv_addr_t * recv_addr )

static

Apply the log records to a page.

Parameters

[in,out] recv_addr Redo log records to apply

◆ recv_calc_lsn_on_data_add()

lsn_t recv_calc_lsn_on_data_add	(	lsn_t	lsn,
		os_offset_t	len
	)

Calculates the new value for lsn when more data is added to the log.

Parameters

[in]	lsn	Old LSN
[in]	len	This many bytes of data is added, log block headers not included

Returns: LSN after data addition

◆ recv_data_copy_to_buf()

static void recv_data_copy_to_buf	(	byte *	buf,
		recv_t *	recv
	)

static

Copies the log record body from recv to buf.

Parameters

[in]	buf	Buffer of length at least recv->len
[in]	recv	Log record

◆ recv_get_page_map()

static recv_sys_t::Space * recv_get_page_map	(	space_id_t	space_id,
		bool	create
	)

static

Get the page map for a tablespace.

It will create one if one isn't found.

Parameters

[in]	space_id	Tablespace ID for which page map required.
[in]	create	false if lookup only

Returns: the space data or null if not found

◆ recv_get_rec()

static recv_addr_t * recv_get_rec ( const page_id_t & page_id )

static

Gets the list of log records for a <space, page>.

Parameters

[in] page_id The <Tablespace ID,Page number> pair

Returns: the redo log entries or nullptr if not found

◆ recv_heap_used()

static size_t recv_heap_used ( )

static

Get the number of bytes used by all the heaps.

Returns: number of bytes used

◆ recv_init_crash_recovery()

static void recv_init_crash_recovery ( )

static

Initialize crash recovery environment.

Can be called iff recv_needed_recovery == false.

◆ recv_page_is_brand_new()

bool recv_page_is_brand_new ( buf_block_t * block )

Returns true if the page is brand new (the next log record is init_file_page or no records to apply).

Parameters

[in] block buffer block

Returns: true if brand new

◆ recv_parse_and_apply_log_recs()

dberr_t recv_parse_and_apply_log_recs ( size_t mem_limit )

Parse log records from a buffer and stores them to a hash table.

Some (, none or all) of the oldest redo log records from the hash table may be applied as a side effect, when the hash table overflows or an mtr requires strict ordering.

Parameters

mem_limit hash table threshold. max() value would skip applying the log records

Returns: DB_SUCCESS if all goes well

◆ recv_process_page_record()

static bool recv_process_page_record	(	const ib::redo::Record_view &	rec,
		lsn_t	start_lsn,
		lsn_t	end_lsn
	)

static

Process a page record.

Parameters

[in]	rec	record to process
[in]	start_lsn	start lsn of mtr containing the record
[in]	end_lsn	end lsn of mtr containing the record

Returns: true on success, false otherwise.

◆ recv_process_space_record()

static bool recv_process_space_record	(	const ib::redo::Record_view &	rec,
		lsn_t	start_lsn
	)

static

Process a table space record.

Parameters

[in]	rec	record to process
[in]	start_lsn	start lsn of mtr containing the record

Returns: true on success, false otherwise.

◆ recv_read_in_area()

static void recv_read_in_area	(	const page_id_t &	requested_page_id,
		const page_size_t &	page_size
	)

static

Reads in pages which have hashed log records, from an area around a given page number.

Parameters

[in]	requested_page_id	The page which has to be read in anyway, so we have an opportunity to read pages nearby.
[in]	page_size	Size of pages in this page's space

◆ recv_recover_page_func()

void recv_recover_page_func	(	bool	just_read_in,
		buf_block_t *	block
	)

Applies the hashed log records to the page, if the page lsn is less than the lsn of a log record.

This can be called when a buffer page has just been read in, or also for a page already in the buffer pool.

Parameters

[in]	just_read_in	true if the IO handler calls this for a freshly read page
[in,out]	block	buffer block

◆ recv_recovery_begin()

dberr_t recv_recovery_begin ( lsn_t checkpoint_lsn )

Scans log from a stream and stores new log data to the parsing buffer.

Parses and hashes the log records if new data found. Each time the hash table gets full (see max_mem), it will apply all the records from the hashmap to the pages ("a batch"), clearing the hashmap, and continuing this way till the end of stream.

Parameters

[in] checkpoint_lsn log sequence number found in checkpoint header. May be inexact (in a middle of an mtr which we can ignore, as it is already applied to tablespace files) until which all redo log has been scanned

◆ recv_recovery_from_checkpoint_finish()

MetadataRecover * recv_recovery_from_checkpoint_finish ( bool aborting )

Complete the recovery from the latest checkpoint.

Parameters

[in] aborting true if the server has to abort due to an error

Returns: recovered persistent metadata or nullptr if aborting

◆ recv_recovery_from_checkpoint_start()

dberr_t recv_recovery_from_checkpoint_start ( lsn_t flush_lsn )

Recovers all tablespaces from the redo log.

See also: recv_recovery_from_checkpoint_finish

Parameters

[in] flush_lsn lsn stored at offset FIL_PAGE_FILE_FLUSH_LSN in the system tablespace header

Returns: error code or DB_SUCCESS

◆ recv_report_corrupt_log()

static void recv_report_corrupt_log	(	std::span< const byte >	buffer,
		size_t	pos,
		lsn_t	lsn
	)

static

Prints diagnostic info of corrupt log.

Parameters

[in]	buffer	buffer containing corrupted data
[in]	pos	position in buffer where the error occurred
[in]	lsn	last processed lsn before the error occurred

◆ recv_reset_buffer()

static void recv_reset_buffer ( )

static

Moves the parsing buffer data left to the buffer start.

◆ recv_sys_close()

void recv_sys_close ( )

Release recovery system mutexes.

◆ recv_sys_create()

void recv_sys_create ( )

Creates the recovery system.

◆ recv_sys_empty_hash()

static void recv_sys_empty_hash ( )

static

Empties the hash table when it has been fully processed.

◆ recv_sys_finish()

static void recv_sys_finish ( )

static

Free up recovery data structures.

◆ recv_sys_free()

void recv_sys_free ( )

Frees the recovery system.

◆ recv_sys_init()

void recv_sys_init ( )

Inits the recovery system for a recovery operation.

◆ recv_sys_resize_buf()

bool recv_sys_resize_buf ( )

Resize the recovery parsing buffer up to log_buffer_size.

◆ recv_sys_var_init()

void recv_sys_var_init ( )

Reset the state of the recovery system variables.

◆ recv_track_changes_of_recovered_lsn()

void recv_track_changes_of_recovered_lsn ( )

◆ recv_writer_is_active()

static bool recv_writer_is_active ( )

static

◆ recv_writer_thread()

static void recv_writer_thread ( )

static

recv_writer thread tasked with flushing dirty pages from the buffer pools.

Variable Documentation

◆ mem_log_recv_space_hash_key

PSI_memory_key mem_log_recv_space_hash_key

◆ PARSING_BUF_MINIMUM_SIZE

const size_t PARSING_BUF_MINIMUM_SIZE = 2048

static

◆ recv_lsn_checks_on

bool recv_lsn_checks_on

true if buf_page_is_corrupted() should check if the log sequence number (FIL_PAGE_LSN) is in the future.

Initially false, and set by recv_recovery_from_checkpoint_start().

◆ recv_max_page_lsn

lsn_t recv_max_page_lsn

static

The maximum lsn we see for a page during the recovery process.

If this is bigger than the lsn we are able to scan up to, that is an indication that the recovery failed and the database may be corrupt.

◆ recv_n_frames_for_pages_per_pool_instance

size_t recv_n_frames_for_pages_per_pool_instance

This many blocks must be left in each Buffer Pool instance to be managed by the LRU when we scan the log and store the scanned log records in a hashmap allocated in the Buffer Pool in frames of non-LRU managed blocks.

We will use these free blocks to read in pages when we start applying the log records to the database.

◆ recv_needed_recovery

bool recv_needed_recovery

true when recv_init_crash_recovery() has been called.

◆ RECV_READ_AHEAD_AREA

const size_t RECV_READ_AHEAD_AREA = 32

static

Read-ahead area in applying log records to file pages.

◆ recv_recovery_on

volatile bool recv_recovery_on

true when applying redo log records during crash recovery; false otherwise.

Note that this is false while a background thread is rolling back incomplete transactions.

◆ recv_scan_print_counter

ulint recv_scan_print_counter

static

The following counter is used to decide when to print info on log scan.

◆ recv_sys

recv_sys_t* recv_sys = nullptr

The recovery system.

Macros

Functions

Variables

Detailed Description

Macro Definition Documentation

◆ RECV_DATA_BLOCK_SIZE

Function Documentation

◆ apply_tablespace_side_effects()

◆ get_mlog_string()

◆ one_less_page_to_recover()

◆ parse_buffer()

◆ process_mtr()

◆ process_mtr_and_possibly_apply_records()

◆ recv_add_to_hash_table()

◆ recv_apply_hashed_log_recs()

◆ recv_apply_log_rec()

◆ recv_calc_lsn_on_data_add()

◆ recv_data_copy_to_buf()

◆ recv_get_page_map()

◆ recv_get_rec()

◆ recv_heap_used()

◆ recv_init_crash_recovery()

◆ recv_page_is_brand_new()

◆ recv_parse_and_apply_log_recs()

◆ recv_process_page_record()

◆ recv_process_space_record()

◆ recv_read_in_area()

◆ recv_recover_page_func()

◆ recv_recovery_begin()

◆ recv_recovery_from_checkpoint_finish()

◆ recv_recovery_from_checkpoint_start()

◆ recv_report_corrupt_log()

◆ recv_reset_buffer()

◆ recv_sys_close()

◆ recv_sys_create()

◆ recv_sys_empty_hash()

◆ recv_sys_finish()

◆ recv_sys_free()

◆ recv_sys_init()

◆ recv_sys_resize_buf()

◆ recv_sys_var_init()

◆ recv_track_changes_of_recovered_lsn()

◆ recv_writer_is_active()

◆ recv_writer_thread()

Variable Documentation

◆ mem_log_recv_space_hash_key

◆ PARSING_BUF_MINIMUM_SIZE

◆ recv_lsn_checks_on

◆ recv_max_page_lsn

◆ recv_n_frames_for_pages_per_pool_instance

◆ recv_needed_recovery

◆ RECV_READ_AHEAD_AREA

◆ recv_recovery_on

◆ recv_scan_print_counter

◆ recv_sys