海山数据库(He3DB)源码详解：海山Mysql 崩溃恢复（3）-日志应用
崩溃恢复之日志应用
recv_recovery_from_checkpoint_start_for_se(flushed_lsn)为崩溃恢复的入口函数，其从innobase_start_or_create_for_mysql()调用，传入参数flushed_lsn是读取存在系统表空间的fil_write_flushed_lsn获得的，该函数在日志解析、日志应用等线程创建之前，进行崩溃恢复操作。
MySQL redo日志崩溃恢复整体流程如图所示，其中，日志应用的流程主要集中在右边部分，recv_parse_or_apply_log_rec_body用于解析或应用日志记录体，recv_recover_page_func用于恢复页面函数，这是应用日志记录到数据库页面的关键步骤。
图中还显示了日志解析和日志应用流程之间的交互，特别是在解析日志记录后，如何将它们应用到数据库页面以完成恢复过程。本文将针对日志应用过程中涉及到的几个函数源码进行详细解析。
1、recv_apply_hashed_log_recs()函数流程
void
recv_apply_hashed_log_recs(
/*=======================*/
ibool allow_ibuf)
{
recv_addr_t* recv_addr;
ulint i;
ibool has_printed = FALSE;
mtr_t mtr;
loop:
mutex_enter(&(recv_sys->mutex));
if (recv_sys->apply_batch_on) {
mutex_exit(&(recv_sys->mutex));
os_thread_sleep(500000);
goto loop;
}
ut_ad(!allow_ibuf == log_mutex_own());
if (!allow_ibuf) {
recv_no_ibuf_operations = true;
}
recv_sys->apply_log_recs = TRUE;
recv_sys->apply_batch_on = TRUE;
for (i = 0; i < hash_get_n_cells(recv_sys->addr_hash); i++) {
for (recv_addr = static_cast<recv_addr_t*>(
HASH_GET_FIRST(recv_sys->addr_hash, i));
recv_addr != 0;
recv_addr = static_cast<recv_addr_t*>(
HASH_GET_NEXT(addr_hash, recv_addr))) {
if (srv_is_tablespace_truncated(recv_addr->space)) {
ut_a(recv_sys->n_addrs);
recv_addr->state = RECV_DISCARDED;
recv_sys->n_addrs--;
continue;
}
if (recv_addr->state == RECV_DISCARDED) {
ut_a(recv_sys->n_addrs);
recv_sys->n_addrs--;
continue;
}
const page_id_t page_id(recv_addr->space,
recv_addr->page_no);
bool found;
const page_size_t& page_size
= fil_space_get_page_size(recv_addr->space,
&found);
ut_ad(found);
if (recv_addr->state == RECV_NOT_PROCESSED) {
if (!has_printed) {
ib::info() << "Starting an apply batch"
" of log records"
" to the database...";
fputs("InnoDB: Progress in percent: ",
stderr);
has_printed = TRUE;
}
mutex_exit(&(recv_sys->mutex));
if (buf_page_peek(page_id)) {
buf_block_t* block;
mtr_start(&mtr);
block = buf_page_get(
page_id, page_size,
RW_X_LATCH, &mtr);
buf_block_dbg_add_level(
block, SYNC_NO_ORDER_CHECK);
recv_recover_page(FALSE, block);
mtr_commit(&mtr);
} else {
recv_read_in_area(page_id);
}
mutex_enter(&(recv_sys->mutex));
}
}
if (has_printed
&& (i * 100) / hash_get_n_cells(recv_sys->addr_hash)
!= ((i + 1) * 100)
/ hash_get_n_cells(recv_sys->addr_hash)) {
fprintf(stderr, "%lu ", (ulong)
((i * 100)
/ hash_get_n_cells(recv_sys->addr_hash)));
}
}
/* Wait until all the pages have been processed */
while (recv_sys->n_addrs != 0) {
mutex_exit(&(recv_sys->mutex));
os_thread_sleep(500000);
mutex_enter(&(recv_sys->mutex));
}
if (has_printed) {
fprintf(stderr, "\n");
}
if (!allow_ibuf) {
/* Flush all the file pages to disk and invalidate them in
the buffer pool */
ut_d(recv_no_log_write = true);
mutex_exit(&(recv_sys->mutex));
log_mutex_exit();
/* Stop the recv_writer thread from issuing any LRU
flush batches. */
mutex_enter(&recv_sys->writer_mutex);
/* Wait for any currently run batch to end. */
buf_flush_wait_LRU_batch_end();
os_event_reset(recv_sys->flush_end);
recv_sys->flush_type = BUF_FLUSH_LIST;
os_event_set(recv_sys->flush_start);
os_event_wait(recv_sys->flush_end);
buf_pool_invalidate();
/* Allow batches from recv_writer thread. */
mutex_exit(&recv_sys->writer_mutex);
log_mutex_enter();
mutex_enter(&(recv_sys->mutex));
ut_d(recv_no_log_write = false);
recv_no_ibuf_operations = false;
}
recv_sys->apply_log_recs = FALSE;
recv_sys->apply_batch_on = FALSE;
recv_sys_empty_hash();
if (has_printed) {
ib::info() << "Apply batch completed";
}
mutex_exit(&(recv_sys->mutex));
}
recv_apply_hashed_log_recs()函数的作用是将存储的日志记录从哈希表中清空，并将它们应用到相应的页面中。下面是对该函数的详细解析。
1、设置正在进行应用状态
recv_sys->apply_log_recs = TRUE;
recv_sys->apply_batch_on = TRUE;
2、取出哈希节点：
redo日志最终存储在recv_sys->addr_hash哈希表结构如下所示：
- 因此想要取出哈希节点需要进行2次for循环：
- 第1次for循环，是对哈希桶进行遍历。
- 第2次for循环，对哈希链表进行变量，此时每个节点代表每个将进行应用的数据页
for (i = 0; i < hash_get_n_cells(recv_sys->addr_hash); i++) {
for (recv_addr = static_cast<recv_addr_t*>(
HASH_GET_FIRST(recv_sys->addr_hash, i));
recv_addr != 0;
recv_addr = static_cast<recv_addr_t*>(
HASH_GET_NEXT(addr_hash, recv_addr)))
3、如果表已经truncate：
- 检查当前表是否
truncate，修改truncate表的哈希节点状态为RECV_DISCARDED，并且减少n_addrs。
if (srv_is_tablespace_truncated(recv_addr->space)) {
/* Avoid applying REDO log for the tablespace
that is schedule for TRUNCATE. */
ut_a(recv_sys->n_addrs);
recv_addr->state = RECV_DISCARDED;
recv_sys->n_addrs--;
continue;
}
4、缓冲池中存在当前数据页，调用recv_recover_page宏。
recv_recover_page宏将调用recv_recover_page_func()进行应用操作。- 当前if条件下的流程是1个mtr。
5、如果不在缓冲池中，调用recv_read_in_area()读取数据页到缓冲池。
recv_read_in_area()函数后续采用异步读取的方式，由io_handler_thread完成读取，并在读取完成后判断是否在崩溃恢复过程中，在崩溃恢复过程中随后进行应用操作。
if (buf_page_peek(page_id)) {
buf_block_t* block;
mtr_start(&mtr);
block = buf_page_get(
page_id, page_size,
RW_X_LATCH, &mtr);
buf_block_dbg_add_level(
block, SYNC_NO_ORDER_CHECK);
recv_recover_page(FALSE, block);
mtr_commit(&mtr);
} else {
recv_read_in_area(page_id);
}
6、等待所有数据页应用完成。
while (recv_sys->n_addrs != 0) {
mutex_exit(&(recv_sys->mutex));
os_thread_sleep(500000);
mutex_enter(&(recv_sys->mutex));
}
7、修改全局状态并且情况recv_sys的哈希表。
recv_sys_empty_hash()在清理了recv_sys旧的哈希表后，为recv_sys分配了新的哈希表。
recv_sys->apply_log_recs = FALSE;
recv_sys->apply_batch_on = FALSE;
recv_sys_empty_hash();
该函数的大致流程如下图所示：
2、recv_recover_page_func()函数流程
void
recv_recover_page_func(
buf_block_t* block) /*!< in/out: buffer block */
{
page_t* page;
page_zip_des_t* page_zip;
recv_addr_t* recv_addr;
recv_t* recv;
byte* buf;
lsn_t start_lsn;
lsn_t end_lsn;
lsn_t page_lsn;
lsn_t page_newest_lsn;
ibool modification_to_page;
mtr_t mtr;
mutex_enter(&(recv_sys->mutex));
if (recv_sys->apply_log_recs == FALSE) {
mutex_exit(&(recv_sys->mutex));
return;
}
recv_addr = recv_get_fil_addr_struct(block->page.id.space(),
block->page.id.page_no());
if ((recv_addr == NULL)
|| (recv_addr->state == RECV_BEING_PROCESSED)
|| (recv_addr->state == RECV_PROCESSED)) {
ut_ad(recv_addr == NULL || recv_needed_recovery);
mutex_exit(&(recv_sys->mutex));
return;
}
recv_addr->state = RECV_BEING_PROCESSED;
mutex_exit(&(recv_sys->mutex));
mtr_start(&mtr);
mtr_set_log_mode(&mtr, MTR_LOG_NONE);
page = block->frame;
page_zip = buf_block_get_page_zip(block);
/* Read the newest modification lsn from the page */
page_lsn = mach_read_from_8(page + FIL_PAGE_LSN);
modification_to_page = FALSE;
start_lsn = end_lsn = 0;
recv = UT_LIST_GET_FIRST(recv_addr->rec_list);
while (recv) {
end_lsn = recv->end_lsn;
ut_ad(end_lsn
<= UT_LIST_GET_FIRST(log_sys->log_groups)->scanned_lsn);
if (recv->len > RECV_DATA_BLOCK_SIZE) {
/* We have to copy the record body to a separate
buffer */
buf = static_cast<byte*>(ut_malloc_nokey(recv->len));
recv_data_copy_to_buf(buf, recv);
} else {
buf = ((byte*)(recv->data)) + sizeof(recv_data_t);
}
if (recv->type == MLOG_INIT_FILE_PAGE) {
page_lsn = page_newest_lsn;
memset(FIL_PAGE_LSN + page, 0, 8);
memset(UNIV_PAGE_SIZE - FIL_PAGE_END_LSN_OLD_CHKSUM
+ page, 0, 8);
if (page_zip) {
memset(FIL_PAGE_LSN + page_zip->data, 0, 8);
}
}
bool skip_recv = false;
if (srv_was_tablespace_truncated(fil_space_get(recv_addr->space))) {
lsn_t init_lsn =
truncate_t::get_truncated_tablespace_init_lsn(
recv_addr->space);
skip_recv = (recv->start_lsn < init_lsn);
}
if (recv->start_lsn >= page_lsn
&& !srv_is_tablespace_truncated(recv_addr->space)
&& !skip_recv) {
lsn_t end_lsn;
if (!modification_to_page) {
modification_to_page = TRUE;
start_lsn = recv->start_lsn;
}
DBUG_PRINT("ib_log",
("apply " LSN_PF ":"
" %s len " ULINTPF " page %u:%u",
recv->start_lsn,
get_mlog_string(recv->type), recv->len,
recv_addr->space,
recv_addr->page_no));
recv_parse_or_apply_log_rec_body(
recv->type, buf, buf + recv->len,
recv_addr->space, recv_addr->page_no,
block, &mtr);
end_lsn = recv->start_lsn + recv->len;
mach_write_to_8(FIL_PAGE_LSN + page, end_lsn);
mach_write_to_8(UNIV_PAGE_SIZE
- FIL_PAGE_END_LSN_OLD_CHKSUM
+ page, end_lsn);
if (page_zip) {
mach_write_to_8(FIL_PAGE_LSN
+ page_zip->data, end_lsn);
}
}
if (recv->len > RECV_DATA_BLOCK_SIZE) {
ut_free(buf);
}
recv = UT_LIST_GET_NEXT(rec_list, recv);
}
mtr.discard_modifications();
mtr_commit(&mtr);
mutex_enter(&(recv_sys->mutex));
if (recv_max_page_lsn < page_lsn) {
recv_max_page_lsn = page_lsn;
}
recv_addr->state = RECV_PROCESSED;
ut_a(recv_sys->n_addrs);
recv_sys->n_addrs--;
mutex_exit(&(recv_sys->mutex));
}
- 在
recv_recover_page_func()函数中，如果页面的日志序列号（LSN）小于日志记录的LSN，则将哈希日志记录应用到该页面。这可以在缓冲区页面刚刚被读入时调用，也可以对已经在缓冲区池中的页面调用。下面是对该函数的详细解析。
1、根据表空间号和页号取出对应日志链表。
recv_addr = recv_get_fil_addr_struct(block->page.id.space(),
block->page.id.page_no());
if ((recv_addr == NULL)
|| (recv_addr->state == RECV_BEING_PROCESSED)
|| (recv_addr->state == RECV_PROCESSED)) {
ut_ad(recv_addr == NULL || recv_needed_recovery);
mutex_exit(&(recv_sys->mutex));
return;
}
2、修改日志链表状态为RECV_BEING_PROCESSED。
recv_addr->state = RECV_BEING_PROCESSED;
3、从数据页中读取最后被修改的lsn值，如果是在热备份状态下，从缓冲池控制块读取lsn。
page_lsn = mach_read_from_8(page + FIL_PAGE_LSN);
#ifndef UNIV_HOTBACKUP
page_newest_lsn = buf_page_get_newest_modification(&block->page);
if (page_newest_lsn) {
page_lsn = page_newest_lsn;
}
#else
page_newest_lsn = 0;
#endif
4、从日志链表读取日志，并且对每个日志进行循环。
recv = UT_LIST_GET_FIRST(recv_addr->rec_list);
while (recv) {
...
}
5、如果日志长度过长，采用动态分配空间将redo日志拷贝到这块内存中，否则直接使用recv中的数据。
if (recv->len > RECV_DATA_BLOCK_SIZE) {
buf = static_cast<byte*>(ut_malloc_nokey(recv->len));
recv_data_copy_to_buf(buf, recv);
} else {
buf = ((byte*)(recv->data)) + sizeof(recv_data_t);
}
6、如果当前表truncate，跳过truncate操作之前的redo日志
bool skip_recv = false;
if (srv_was_tablespace_truncated(fil_space_get(recv_addr->space))) {
lsn_t init_lsn =
truncate_t::get_truncated_tablespace_init_lsn(
recv_addr->space);
skip_recv = (recv->start_lsn < init_lsn);
}
7、如果日志的lsn>页的lsn，调用recv_parse_or_apply_log_rec_body()进行日志应用，并在应用完成后，更新数据页中的lsn字段。
if (!modification_to_page) {
modification_to_page = TRUE;
start_lsn = recv->start_lsn;
}
DBUG_PRINT("ib_log",
("apply " LSN_PF ":"
" %s len " ULINTPF " page %u:%u",
recv->start_lsn,
get_mlog_string(recv->type), recv->len,
recv_addr->space,
recv_addr->page_no));
recv_parse_or_apply_log_rec_body(
recv->type, buf, buf + recv->len,
recv_addr->space, recv_addr->page_no,
block, &mtr);
end_lsn = recv->start_lsn + recv->len;
mach_write_to_8(FIL_PAGE_LSN + page, end_lsn);
mach_write_to_8(UNIV_PAGE_SIZE
- FIL_PAGE_END_LSN_OLD_CHKSUM
+ page, end_lsn);
if (page_zip) {
mach_write_to_8(FIL_PAGE_LSN
+ page_zip->data, end_lsn);
}
8、如果当前redo日志使用的动态分配的内存，释放这部分内存，并且调用UT_LIST_GET_NEXT宏，获取日志链表的下一条日志，之后进入下一轮循环。
if (recv->len > RECV_DATA_BLOCK_SIZE) {
ut_free(buf);
}
recv = UT_LIST_GET_NEXT(rec_list, recv);
9、调用buf_flush_recv_note_modificatio()函数，更新数据页缓冲池控制块的newest_modification，并且在flush链表中重新排序
10、应用流程接收修改日志链表状态为RECV_PROCESSED并且减少n_addrs。
recv_addr->state = RECV_PROCESSED;
ut_a(recv_sys->n_addrs);
recv_sys->n_addrs--;
mutex_exit(&(recv_sys->mutex));
该函数的大致流程如下图所示：
3、recv_parse_or_apply_log_rec_body()函数流程
static
byte*
recv_parse_or_apply_log_rec_body(
mlog_id_t type,
byte* ptr,
byte* end_ptr,
ulint space_id,
ulint page_no,
buf_block_t* block,
mtr_t* mtr)
{
ut_ad(!block == !mtr);
switch (type) {
/* 根据日志类型，调用对应函数 */
...
default:
break;
}
dict_index_t* index = NULL;
page_t* page;
page_zip_des_t* page_zip;
#ifdef UNIV_DEBUG
ulint page_type;
#endif /* UNIV_DEBUG */
if (block) {
/* Applying a page log record. */
page = block->frame;
page_zip = buf_block_get_page_zip(block);
ut_d(page_type = fil_page_get_type(page));
} else {
/* Parsing a page log record. */
page = NULL;
page_zip = NULL;
ut_d(page_type = FIL_PAGE_TYPE_ALLOCATED);
}
const byte* old_ptr = ptr;
switch (type) {
/* 根据日志类型，调用对应函数 */
...
break;
}
ptr = mlog_parse_nbytes(type, ptr, end_ptr, page, page_zip);
if (ptr != NULL && page != NULL
&& page_no == 0 && type == MLOG_4BYTES) {
ulint offs = mach_read_from_2(old_ptr);
switch (offs) {
fil_space_t* space;
ulint val;
default:
break;
/* 根据日志类型，调用对应函数 */
...
}
}
}
break;
case MLOG_REC_INSERT: case MLOG_COMP_REC_INSERT:
...
...
if (index) {
dict_table_t* table = index->table;
dict_mem_index_free(index);
dict_mem_table_free(table);
}
return(ptr);
}
recv_parse_or_apply_log_rec_body()函数的主要作用是根据redo日志类型，对redo日志的数据部分进行解析和应用的操作都是在该函数中完成的，该函数根据传入参数blcok是否为NULL，来区分是否进行日志应用操作，如果block == NULL，后续只进行日志解析，不进行日志应用。
1151
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
