Skip to content

M
mariadb
Commit 271e6ae3 authored by Marko Mäkelä's avatar Marko Mäkelä
Browse files
Options

Bug#54914: InnoDB: performance drop with innodb_change_buffering=all 

Reduce ibuf_mutex and ibuf_pessimistic_insert_mutex contention further.

Protect ibuf->empty by the insert buffer root page latch, not ibuf_mutex.

ibuf_tree_root_get(): Assert that ibuf_mutex is owned by the
caller. Assert that the stamped page number is correct. Assert that
ibuf->empty agrees with the root page.

ibuf_size_update(): Do not update ibuf->empty.

ibuf_init_at_db_start(): Update ibuf->empty while holding the root page latch.

ibuf_add_free_page(): Return TRUE/FALSE instead of DB_SUCCESS/DB_STRONG_FAIL.

ibuf_remove_free_page(): Release ibuf_pessimistic_insert_mutex as
early as possible.

ibuf_contract_ext(): Rely on a dirty read of ibuf->empty, unless the
server is being shut down. Never acquire ibuf_mutex. Eliminate n_stored.

ibuf_contract_after_insert(): Never acquire ibuf_mutex. Perform dirty
reads of ibuf->size and ibuf->max_size.

ibuf_insert_low(): Only acquire ibuf_mutex for mode==BTR_MODIFY_TREE.
Perform dirty reads of ibuf->size and ibuf->max_size. Update
ibuf->empty while holding the root page latch.

ibuf_delete_rec(): Update ibuf->empty while holding the root page latch.

ibuf_is_empty(): Release ibuf_mutex earlier.
parent a4c5cf7c
Show whitespace changes
Inline Side-by-side
Showing with 134 additions and 127 deletions
storage/innobase/ibuf/ibuf0ibuf.c
View file @ 271e6ae3
...@@ -55,6 +55,7 @@ Created 7/19/1997 Heikki Tuuri ...@@ -55,6 +55,7 @@ Created 7/19/1997 Heikki Tuuri
#include "lock0lock.h" #include "lock0lock.h"
#include "log0recv.h" #include "log0recv.h"
#include "que0que.h" #include "que0que.h"
#include "srv0start.h" /* srv_shutdown_state */
/* STRUCTURE OF AN INSERT BUFFER RECORD /* STRUCTURE OF AN INSERT BUFFER RECORD
...@@ -395,8 +396,10 @@ ibuf_tree_root_get( ...@@ -395,8 +396,10 @@ ibuf_tree_root_get(
mtr_t* mtr) /*!< in: mtr */ mtr_t* mtr) /*!< in: mtr */
{ {
buf_block_t* block; buf_block_t* block;
page_t* root;
ut_ad(ibuf_inside()); ut_ad(ibuf_inside());
ut_ad(mutex_own(&ibuf_mutex));
mtr_x_lock(dict_index_get_lock(ibuf->index), mtr); mtr_x_lock(dict_index_get_lock(ibuf->index), mtr);
...@@ -405,7 +408,13 @@ ibuf_tree_root_get( ...@@ -405,7 +408,13 @@ ibuf_tree_root_get(
buf_block_dbg_add_level(block, SYNC_TREE_NODE); buf_block_dbg_add_level(block, SYNC_TREE_NODE);
return(buf_block_get_frame(block)); root = buf_block_get_frame(block);
ut_ad(page_get_space_id(root) == IBUF_SPACE_ID);
ut_ad(page_get_page_no(root) == FSP_IBUF_TREE_ROOT_PAGE_NO);
ut_ad(ibuf->empty == (page_get_n_recs(root) == 0));
return(root);
} }
#ifdef UNIV_IBUF_COUNT_DEBUG #ifdef UNIV_IBUF_COUNT_DEBUG
...@@ -482,8 +491,6 @@ ibuf_size_update( ...@@ -482,8 +491,6 @@ ibuf_size_update(
/* the '1 +' is the ibuf header page */ /* the '1 +' is the ibuf header page */
ibuf->size = ibuf->seg_size - (1 + ibuf->free_list_len); ibuf->size = ibuf->seg_size - (1 + ibuf->free_list_len);
ibuf->empty = page_get_n_recs(root) == 0;
} }
/******************************************************************//** /******************************************************************//**
...@@ -554,6 +561,7 @@ ibuf_init_at_db_start(void) ...@@ -554,6 +561,7 @@ ibuf_init_at_db_start(void)
ibuf_size_update(root, &mtr); ibuf_size_update(root, &mtr);
mutex_exit(&ibuf_mutex); mutex_exit(&ibuf_mutex);
ibuf->empty = (page_get_n_recs(root) == 0);
mtr_commit(&mtr); mtr_commit(&mtr);
ibuf_exit(); ibuf_exit();
...@@ -2025,9 +2033,9 @@ ibuf_data_too_much_free(void) ...@@ -2025,9 +2033,9 @@ ibuf_data_too_much_free(void)
/*********************************************************************//** /*********************************************************************//**
Allocates a new page from the ibuf file segment and adds it to the free Allocates a new page from the ibuf file segment and adds it to the free
list. list.
@return DB_SUCCESS, or DB_STRONG_FAIL if no space left */ @return TRUE on success, FALSE if no space left */
static static
ulint ibool
ibuf_add_free_page(void) ibuf_add_free_page(void)
/*====================*/ /*====================*/
{ {
...@@ -2063,10 +2071,10 @@ ibuf_add_free_page(void) ...@@ -2063,10 +2071,10 @@ ibuf_add_free_page(void)
header_page + IBUF_HEADER + IBUF_TREE_SEG_HEADER, 0, FSP_UP, header_page + IBUF_HEADER + IBUF_TREE_SEG_HEADER, 0, FSP_UP,
&mtr); &mtr);
if (page_no == FIL_NULL) { if (UNIV_UNLIKELY(page_no == FIL_NULL)) {
mtr_commit(&mtr); mtr_commit(&mtr);
return(DB_STRONG_FAIL); return(FALSE);
} }
{ {
...@@ -2113,7 +2121,7 @@ ibuf_add_free_page(void) ...@@ -2113,7 +2121,7 @@ ibuf_add_free_page(void)
ibuf_exit(); ibuf_exit();
return(DB_SUCCESS); return(TRUE);
} }
/*********************************************************************//** /*********************************************************************//**
...@@ -2143,20 +2151,17 @@ ibuf_remove_free_page(void) ...@@ -2143,20 +2151,17 @@ ibuf_remove_free_page(void)
header_page = ibuf_header_page_get(&mtr); header_page = ibuf_header_page_get(&mtr);
/* Prevent pessimistic inserts to insert buffer trees for a while */ /* Prevent pessimistic inserts to insert buffer trees for a while */
mutex_enter(&ibuf_pessimistic_insert_mutex);
ibuf_enter(); ibuf_enter();
mutex_enter(&ibuf_pessimistic_insert_mutex);
mutex_enter(&ibuf_mutex); mutex_enter(&ibuf_mutex);
if (!ibuf_data_too_much_free()) { if (!ibuf_data_too_much_free()) {
mutex_exit(&ibuf_mutex); mutex_exit(&ibuf_mutex);
mutex_exit(&ibuf_pessimistic_insert_mutex);
ibuf_exit(); ibuf_exit();
mutex_exit(&ibuf_pessimistic_insert_mutex);
mtr_commit(&mtr); mtr_commit(&mtr);
return; return;
...@@ -2218,11 +2223,11 @@ ibuf_remove_free_page(void) ...@@ -2218,11 +2223,11 @@ ibuf_remove_free_page(void)
flst_remove(root + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST, flst_remove(root + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST,
page + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST_NODE, &mtr); page + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST_NODE, &mtr);
mutex_exit(&ibuf_pessimistic_insert_mutex);
ibuf->seg_size--; ibuf->seg_size--;
ibuf->free_list_len--; ibuf->free_list_len--;
mutex_exit(&ibuf_pessimistic_insert_mutex);
/* Set the bit indicating that this page is no more an ibuf tree page /* Set the bit indicating that this page is no more an ibuf tree page
(level 2 page) */ (level 2 page) */
...@@ -2484,17 +2489,19 @@ ibuf_contract_ext( ...@@ -2484,17 +2489,19 @@ ibuf_contract_ext(
ulint page_nos[IBUF_MAX_N_PAGES_MERGED]; ulint page_nos[IBUF_MAX_N_PAGES_MERGED];
ulint space_ids[IBUF_MAX_N_PAGES_MERGED]; ulint space_ids[IBUF_MAX_N_PAGES_MERGED];
ib_int64_t space_versions[IBUF_MAX_N_PAGES_MERGED]; ib_int64_t space_versions[IBUF_MAX_N_PAGES_MERGED];
ulint n_stored;
ulint sum_sizes; ulint sum_sizes;
mtr_t mtr; mtr_t mtr;
*n_pages = 0; *n_pages = 0;
ut_ad(!ibuf_inside()); ut_ad(!ibuf_inside());
mutex_enter(&ibuf_mutex); /* We perform a dirty read of ibuf->empty, without latching
the insert buffer root page. We trust this dirty read except
when a slow shutdown is being executed. During a slow
shutdown, the insert buffer merge must be completed. */
if (ibuf->empty) { if (UNIV_UNLIKELY(ibuf->empty)
mutex_exit(&ibuf_mutex); && UNIV_LIKELY(!srv_shutdown_state)) {
ibuf_is_empty: ibuf_is_empty:
#if 0 /* TODO */ #if 0 /* TODO */
...@@ -2523,18 +2530,18 @@ ibuf_is_empty: ...@@ -2523,18 +2530,18 @@ ibuf_is_empty:
position within the leaf */ position within the leaf */
btr_pcur_open_at_rnd_pos(ibuf->index, BTR_SEARCH_LEAF, &pcur, &mtr); btr_pcur_open_at_rnd_pos(ibuf->index, BTR_SEARCH_LEAF, &pcur, &mtr);
mutex_exit(&ibuf_mutex);
ut_ad(page_validate(btr_pcur_get_page(&pcur), ibuf->index)); ut_ad(page_validate(btr_pcur_get_page(&pcur), ibuf->index));
if (page_get_n_recs(btr_pcur_get_page(&pcur)) == 0) { if (page_get_n_recs(btr_pcur_get_page(&pcur)) == 0) {
/* When the ibuf tree is emptied completely, the last record /* If a B-tree page is empty, it must be the root page
is removed using an optimistic delete and ibuf_size_update and the whole B-tree must be empty. InnoDB does not
is not called, causing ibuf->empty to remain FALSE. If we do allow empty B-tree pages other than the root. */
not reset it to TRUE here then database shutdown will hang ut_ad(ibuf->empty);
in the loop in ibuf_contract_for_n_pages. */ ut_ad(page_get_space_id(btr_pcur_get_page(&pcur))
== IBUF_SPACE_ID);
ibuf->empty = TRUE; ut_ad(page_get_page_no(btr_pcur_get_page(&pcur))
== FSP_IBUF_TREE_ROOT_PAGE_NO);
ibuf_exit(); ibuf_exit();
...@@ -2546,10 +2553,10 @@ ibuf_is_empty: ...@@ -2546,10 +2553,10 @@ ibuf_is_empty:
sum_sizes = ibuf_get_merge_page_nos(TRUE, btr_pcur_get_rec(&pcur), sum_sizes = ibuf_get_merge_page_nos(TRUE, btr_pcur_get_rec(&pcur),
space_ids, space_versions, space_ids, space_versions,
page_nos, &n_stored); page_nos, n_pages);
#if 0 /* defined UNIV_IBUF_DEBUG */ #if 0 /* defined UNIV_IBUF_DEBUG */
fprintf(stderr, "Ibuf contract sync %lu pages %lu volume %lu\n", fprintf(stderr, "Ibuf contract sync %lu pages %lu volume %lu\n",
sync, n_stored, sum_sizes); sync, *n_pages, sum_sizes);
#endif #endif
ibuf_exit(); ibuf_exit();
...@@ -2557,8 +2564,7 @@ ibuf_is_empty: ...@@ -2557,8 +2564,7 @@ ibuf_is_empty:
btr_pcur_close(&pcur); btr_pcur_close(&pcur);
buf_read_ibuf_merge_pages(sync, space_ids, space_versions, page_nos, buf_read_ibuf_merge_pages(sync, space_ids, space_versions, page_nos,
n_stored); *n_pages);
*n_pages = n_stored;
return(sum_sizes + 1); return(sum_sizes + 1);
} }
...@@ -2628,33 +2634,33 @@ ibuf_contract_after_insert( ...@@ -2628,33 +2634,33 @@ ibuf_contract_after_insert(
ibool sync; ibool sync;
ulint sum_sizes; ulint sum_sizes;
ulint size; ulint size;
ulint max_size;
mutex_enter(&ibuf_mutex);
/* Perform dirty reads of ibuf->size and ibuf->max_size, to
if (ibuf->size < ibuf->max_size + IBUF_CONTRACT_ON_INSERT_NON_SYNC) { reduce ibuf_mutex contention. ibuf->max_size remains constant
mutex_exit(&ibuf_mutex); after ibuf_init_at_db_start(), but ibuf->size should be
protected by ibuf_mutex. Given that ibuf->size fits in a
machine word, this should be OK; at worst we are doing some
excessive ibuf_contract() or occasionally skipping a
ibuf_contract(). */
size = ibuf->size;
max_size = ibuf->max_size;
if (size < max_size + IBUF_CONTRACT_ON_INSERT_NON_SYNC) {
return; return;
} }
sync = FALSE; sync = (size >= max_size + IBUF_CONTRACT_ON_INSERT_SYNC);
if (ibuf->size >= ibuf->max_size + IBUF_CONTRACT_ON_INSERT_SYNC) {
sync = TRUE;
}
mutex_exit(&ibuf_mutex);
/* Contract at least entry_size many bytes */ /* Contract at least entry_size many bytes */
sum_sizes = 0; sum_sizes = 0;
size = 1; size = 1;
while ((size > 0) && (sum_sizes < entry_size)) { do {
size = ibuf_contract(sync); size = ibuf_contract(sync);
sum_sizes += size; sum_sizes += size;
} } while (size > 0 && sum_sizes < entry_size);
} }
/*********************************************************************//** /*********************************************************************//**
...@@ -3272,7 +3278,7 @@ ibuf_set_entry_counter( ...@@ -3272,7 +3278,7 @@ ibuf_set_entry_counter(
/*********************************************************************//** /*********************************************************************//**
Buffer an operation in the insert/delete buffer, instead of doing it Buffer an operation in the insert/delete buffer, instead of doing it
directly to the disk page, if this is possible. directly to the disk page, if this is possible.
@return DB_SUCCESS, DB_FAIL, DB_STRONG_FAIL */ @return DB_SUCCESS, DB_STRONG_FAIL or other error */
static static
ulint ulint
ibuf_insert_low( ibuf_insert_low(
...@@ -3302,6 +3308,7 @@ ibuf_insert_low( ...@@ -3302,6 +3308,7 @@ ibuf_insert_low(
rec_t* ins_rec; rec_t* ins_rec;
ibool old_bit_value; ibool old_bit_value;
page_t* bitmap_page; page_t* bitmap_page;
buf_block_t* block;
page_t* root; page_t* root;
ulint err; ulint err;
ibool do_merge; ibool do_merge;
...@@ -3311,7 +3318,6 @@ ibuf_insert_low( ...@@ -3311,7 +3318,6 @@ ibuf_insert_low(
ulint n_stored; ulint n_stored;
mtr_t mtr; mtr_t mtr;
mtr_t bitmap_mtr; mtr_t bitmap_mtr;
ibool too_big;
ut_a(!dict_index_is_clust(index)); ut_a(!dict_index_is_clust(index));
ut_ad(dtuple_check_typed(entry)); ut_ad(dtuple_check_typed(entry));
...@@ -3323,11 +3329,14 @@ ibuf_insert_low( ...@@ -3323,11 +3329,14 @@ ibuf_insert_low(
do_merge = FALSE; do_merge = FALSE;
mutex_enter(&ibuf_mutex); /* Perform dirty reads of ibuf->size and ibuf->max_size, to
too_big = ibuf->size >= ibuf->max_size + IBUF_CONTRACT_DO_NOT_INSERT; reduce ibuf_mutex contention. ibuf->max_size remains constant
mutex_exit(&ibuf_mutex); after ibuf_init_at_db_start(), but ibuf->size should be
protected by ibuf_mutex. Given that ibuf->size fits in a
if (too_big) { machine word, this should be OK; at worst we are doing some
excessive ibuf_contract() or occasionally skipping a
ibuf_contract(). */
if (ibuf->size >= ibuf->max_size + IBUF_CONTRACT_DO_NOT_INSERT) {
/* Insert buffer is now too big, contract it but do not try /* Insert buffer is now too big, contract it but do not try
to insert */ to insert */
...@@ -3361,10 +3370,8 @@ ibuf_insert_low( ...@@ -3361,10 +3370,8 @@ ibuf_insert_low(
if (mode == BTR_MODIFY_TREE) { if (mode == BTR_MODIFY_TREE) {
for (;;) { for (;;) {
mutex_enter(&ibuf_pessimistic_insert_mutex);
ibuf_enter(); ibuf_enter();
mutex_enter(&ibuf_pessimistic_insert_mutex);
mutex_enter(&ibuf_mutex); mutex_enter(&ibuf_mutex);
if (UNIV_LIKELY(ibuf_data_enough_free_for_insert())) { if (UNIV_LIKELY(ibuf_data_enough_free_for_insert())) {
...@@ -3373,17 +3380,13 @@ ibuf_insert_low( ...@@ -3373,17 +3380,13 @@ ibuf_insert_low(
} }
mutex_exit(&ibuf_mutex); mutex_exit(&ibuf_mutex);
ibuf_exit();
mutex_exit(&ibuf_pessimistic_insert_mutex); mutex_exit(&ibuf_pessimistic_insert_mutex);
ibuf_exit();
err = ibuf_add_free_page(); if (UNIV_UNLIKELY(!ibuf_add_free_page())) {
if (UNIV_UNLIKELY(err == DB_STRONG_FAIL)) {
mem_heap_free(heap); mem_heap_free(heap);
return(err); return(DB_STRONG_FAIL);
} }
} }
} else { } else {
...@@ -3423,9 +3426,14 @@ ibuf_insert_low( ...@@ -3423,9 +3426,14 @@ ibuf_insert_low(
before mtr_commit(&mtr). We must not mtr_commit(&mtr) before mtr_commit(&mtr). We must not mtr_commit(&mtr)
until after the IBUF_OP_DELETE has been buffered. */ until after the IBUF_OP_DELETE has been buffered. */
err = DB_STRONG_FAIL; fail_exit:
if (mode == BTR_MODIFY_TREE) {
mutex_exit(&ibuf_mutex);
mutex_exit(&ibuf_pessimistic_insert_mutex);
}
goto function_exit; err = DB_STRONG_FAIL;
goto func_exit;
} }
/* After this point, the page could still be loaded to the /* After this point, the page could still be loaded to the
...@@ -3471,9 +3479,7 @@ ibuf_insert_low( ...@@ -3471,9 +3479,7 @@ ibuf_insert_low(
space_ids, space_versions, space_ids, space_versions,
page_nos, &n_stored); page_nos, &n_stored);
err = DB_STRONG_FAIL; goto fail_exit;
goto function_exit;
} }
} }
...@@ -3484,11 +3490,9 @@ ibuf_insert_low( ...@@ -3484,11 +3490,9 @@ ibuf_insert_low(
&& !ibuf_set_entry_counter(ibuf_entry, space, page_no, &pcur, && !ibuf_set_entry_counter(ibuf_entry, space, page_no, &pcur,
mode == BTR_MODIFY_PREV, &mtr)) { mode == BTR_MODIFY_PREV, &mtr)) {
bitmap_fail: bitmap_fail:
err = DB_STRONG_FAIL;
mtr_commit(&bitmap_mtr); mtr_commit(&bitmap_mtr);
goto function_exit; goto fail_exit;
} }
/* Set the bitmap bit denoting that the insert buffer contains /* Set the bitmap bit denoting that the insert buffer contains
...@@ -3512,10 +3516,19 @@ bitmap_fail: ...@@ -3512,10 +3516,19 @@ bitmap_fail:
err = btr_cur_optimistic_insert(BTR_NO_LOCKING_FLAG, cursor, err = btr_cur_optimistic_insert(BTR_NO_LOCKING_FLAG, cursor,
ibuf_entry, &ins_rec, ibuf_entry, &ins_rec,
&dummy_big_rec, 0, thr, &mtr); &dummy_big_rec, 0, thr, &mtr);
if (err == DB_SUCCESS && op != IBUF_OP_DELETE) { block = btr_cur_get_block(cursor);
/* Update the page max trx id field */ ut_ad(buf_block_get_space(block) == IBUF_SPACE_ID);
page_update_max_trx_id(btr_cur_get_block(cursor), NULL,
thr_get_trx(thr)->id, &mtr); /* If this is the root page, update ibuf->empty. */
if (UNIV_UNLIKELY(buf_block_get_page_no(block)
== FSP_IBUF_TREE_ROOT_PAGE_NO)) {
const page_t* root = buf_block_get_frame(block);
ut_ad(page_get_space_id(root) == IBUF_SPACE_ID);
ut_ad(page_get_page_no(root)
== FSP_IBUF_TREE_ROOT_PAGE_NO);
ibuf->empty = (page_get_n_recs(root) == 0);
} }
} else { } else {
ut_ad(mode == BTR_MODIFY_TREE); ut_ad(mode == BTR_MODIFY_TREE);
...@@ -3532,16 +3545,22 @@ bitmap_fail: ...@@ -3532,16 +3545,22 @@ bitmap_fail:
cursor, cursor,
ibuf_entry, &ins_rec, ibuf_entry, &ins_rec,
&dummy_big_rec, 0, thr, &mtr); &dummy_big_rec, 0, thr, &mtr);
mutex_exit(&ibuf_pessimistic_insert_mutex);
ibuf_size_update(root, &mtr);
mutex_exit(&ibuf_mutex);
ibuf->empty = (page_get_n_recs(root) == 0);
block = btr_cur_get_block(cursor);
ut_ad(buf_block_get_space(block) == IBUF_SPACE_ID);
}
if (err == DB_SUCCESS && op != IBUF_OP_DELETE) { if (err == DB_SUCCESS && op != IBUF_OP_DELETE) {
/* Update the page max trx id field */ /* Update the page max trx id field */
page_update_max_trx_id(btr_cur_get_block(cursor), NULL, page_update_max_trx_id(block, NULL,
thr_get_trx(thr)->id, &mtr); thr_get_trx(thr)->id, &mtr);
} }
ibuf_size_update(root, &mtr); func_exit:
}
function_exit:
#ifdef UNIV_IBUF_COUNT_DEBUG #ifdef UNIV_IBUF_COUNT_DEBUG
if (err == DB_SUCCESS) { if (err == DB_SUCCESS) {
fprintf(stderr, fprintf(stderr,
...@@ -3553,11 +3572,6 @@ function_exit: ...@@ -3553,11 +3572,6 @@ function_exit:
ibuf_count_get(space, page_no) + 1); ibuf_count_get(space, page_no) + 1);
} }
#endif #endif
if (mode == BTR_MODIFY_TREE) {
mutex_exit(&ibuf_mutex);
mutex_exit(&ibuf_pessimistic_insert_mutex);
}
mtr_commit(&mtr); mtr_commit(&mtr);
btr_pcur_close(&pcur); btr_pcur_close(&pcur);
...@@ -3565,17 +3579,9 @@ function_exit: ...@@ -3565,17 +3579,9 @@ function_exit:
mem_heap_free(heap); mem_heap_free(heap);
if (err == DB_SUCCESS) { if (err == DB_SUCCESS && mode == BTR_MODIFY_TREE) {
mutex_enter(&ibuf_mutex);
ibuf->empty = FALSE;
mutex_exit(&ibuf_mutex);
if (mode == BTR_MODIFY_TREE) {
ibuf_contract_after_insert(entry_size); ibuf_contract_after_insert(entry_size);
} }
}
if (do_merge) { if (do_merge) {
#ifdef UNIV_IBUF_DEBUG #ifdef UNIV_IBUF_DEBUG
...@@ -4081,6 +4087,22 @@ ibuf_delete_rec( ...@@ -4081,6 +4087,22 @@ ibuf_delete_rec(
success = btr_cur_optimistic_delete(btr_pcur_get_btr_cur(pcur), mtr); success = btr_cur_optimistic_delete(btr_pcur_get_btr_cur(pcur), mtr);
if (success) { if (success) {
if (UNIV_UNLIKELY(!page_get_n_recs(btr_pcur_get_page(pcur)))) {
/* If a B-tree page is empty, it must be the root page
and the whole B-tree must be empty. InnoDB does not
allow empty B-tree pages other than the root. */
root = btr_pcur_get_page(pcur);
ut_ad(page_get_space_id(root) == IBUF_SPACE_ID);
ut_ad(page_get_page_no(root)
== FSP_IBUF_TREE_ROOT_PAGE_NO);
/* ibuf->empty is protected by the root page latch.
Before the deletion, it had to be FALSE. */
ut_ad(!ibuf->empty);
ibuf->empty = TRUE;
}
#ifdef UNIV_IBUF_COUNT_DEBUG #ifdef UNIV_IBUF_COUNT_DEBUG
fprintf(stderr, fprintf(stderr,
"Decrementing ibuf count of space %lu page %lu\n" "Decrementing ibuf count of space %lu page %lu\n"
...@@ -4108,6 +4130,7 @@ ibuf_delete_rec( ...@@ -4108,6 +4130,7 @@ ibuf_delete_rec(
if (!ibuf_restore_pos(space, page_no, search_tuple, if (!ibuf_restore_pos(space, page_no, search_tuple,
BTR_MODIFY_TREE, pcur, mtr)) { BTR_MODIFY_TREE, pcur, mtr)) {
mutex_exit(&ibuf_mutex);
goto func_exit; goto func_exit;
} }
...@@ -4121,10 +4144,12 @@ ibuf_delete_rec( ...@@ -4121,10 +4144,12 @@ ibuf_delete_rec(
ibuf_count_set(space, page_no, ibuf_count_get(space, page_no) - 1); ibuf_count_set(space, page_no, ibuf_count_get(space, page_no) - 1);
#endif #endif
ibuf_size_update(root, mtr); ibuf_size_update(root, mtr);
mutex_exit(&ibuf_mutex);
ibuf->empty = (page_get_n_recs(root) == 0);
btr_pcur_commit_specify_mtr(pcur, mtr); btr_pcur_commit_specify_mtr(pcur, mtr);
func_exit: func_exit:
mutex_exit(&ibuf_mutex);
btr_pcur_close(pcur); btr_pcur_close(pcur);
return(TRUE); return(TRUE);
...@@ -4642,37 +4667,18 @@ ibuf_is_empty(void) ...@@ -4642,37 +4667,18 @@ ibuf_is_empty(void)
mtr_t mtr; mtr_t mtr;
ibuf_enter(); ibuf_enter();
mutex_enter(&ibuf_mutex);
mtr_start(&mtr); mtr_start(&mtr);
mutex_enter(&ibuf_mutex);
root = ibuf_tree_root_get(&mtr); root = ibuf_tree_root_get(&mtr);
if (page_get_n_recs(root) == 0) {
is_empty = TRUE;
if (ibuf->empty == FALSE) {
fprintf(stderr,
"InnoDB: Warning: insert buffer tree is empty"
" but the data struct does not\n"
"InnoDB: know it. This condition is legal"
" if the master thread has not yet\n"
"InnoDB: run to completion.\n");
}
} else {
ut_a(ibuf->empty == FALSE);
is_empty = FALSE;
}
mutex_exit(&ibuf_mutex); mutex_exit(&ibuf_mutex);
is_empty = (page_get_n_recs(root) == 0);
mtr_commit(&mtr); mtr_commit(&mtr);
ibuf_exit(); ibuf_exit();
ut_a(is_empty == ibuf->empty);
return(is_empty); return(is_empty);
} }
......
storage/innobase/include/ibuf0ibuf.ic
View file @ 271e6ae3
...@@ -46,11 +46,12 @@ struct ibuf_struct{ ...@@ -46,11 +46,12 @@ struct ibuf_struct{
ulint seg_size; /*!< allocated pages of the file ulint seg_size; /*!< allocated pages of the file
segment containing ibuf header and segment containing ibuf header and
tree */ tree */
ibool empty; /*!< after an insert to the ibuf tree ibool empty; /*!< Protected by the page
is performed, this is set to FALSE, latch of the root page of the
and if a contract operation finds insert buffer tree
the tree empty, this is set to (FSP_IBUF_TREE_ROOT_PAGE_NO). TRUE
TRUE */ if and only if the insert
buffer tree is empty. */
ulint free_list_len; /*!< length of the free list */ ulint free_list_len; /*!< length of the free list */
ulint height; /*!< tree height */ ulint height; /*!< tree height */
dict_index_t* index; /*!< insert buffer index */ dict_index_t* index; /*!< insert buffer index */
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7