1
/******************************************************
6
Created 1/8/1997 Heikki Tuuri
7
*******************************************************/
12
#include "row0undo.ic"
16
#include "mach0data.h"
21
#include "trx0purge.h"
28
#include "row0mysql.h"
31
/* How to undo row operations?
32
(1) For an insert, we have stored a prefix of the clustered index record
33
in the undo log. Using it, we look for the clustered record, and using
34
that we look for the records in the secondary indexes. The insert operation
35
may have been left incomplete, if the database crashed, for example.
36
We may have look at the trx id and roll ptr to make sure the record in the
37
clustered index is really the one for which the undo log record was
38
written. We can use the framework we get from the original insert op.
39
(2) Delete marking: We can use the framework we get from the original
40
delete mark op. We only have to check the trx id.
41
(3) Update: This may be the most complicated. We have to use the framework
42
we get from the original update op.
44
What if the same trx repeatedly deletes and inserts an identical row.
45
Then the row id changes and also roll ptr. What if the row id was not
46
part of the ordering fields in the clustered index? Maybe we have to write
47
it to undo log. Well, maybe not, because if we order the row id and trx id
48
in descending order, then the only undeleted copy is the first in the
49
index. Our searches in row operations always position the cursor before
50
the first record in the result set. But, if there is no key defined for
51
a table, then it would be desirable that row id is in ascending order.
52
So, lets store row id in descending order only if it is not an ordering
53
field in the clustered index.
55
NOTE: Deletes and inserts may lead to situation where there are identical
56
records in a secondary index. Is that a problem in the B-tree? Yes.
57
Also updates can lead to this, unless trx id and roll ptr are included in
59
(1) Fix in clustered indexes: include row id, trx id, and roll ptr
60
in node pointers of B-tree.
61
(2) Fix in secondary indexes: include all fields in node pointers, and
62
if an entry is inserted, check if it is equal to the right neighbor,
63
in which case update the right neighbor: the neighbor must be delete
64
marked, set it unmarked and write the trx id of the current transaction.
66
What if the same trx repeatedly updates the same row, updating a secondary
67
index field or not? Updating a clustered index ordering field?
69
(1) If it does not update the secondary index and not the clustered index
70
ord field. Then the secondary index record stays unchanged, but the
71
trx id in the secondary index record may be smaller than in the clustered
72
index record. This is no problem?
73
(2) If it updates secondary index ord field but not clustered: then in
74
secondary index there are delete marked records, which differ in an
75
ord field. No problem.
76
(3) Updates clustered ord field but not secondary, and secondary index
77
is unique. Then the record in secondary index is just updated at the
81
Problem with duplicate records:
82
Fix 1: Add a trx op no field to all indexes. A problem: if a trx with a
83
bigger trx id has inserted and delete marked a similar row, our trx inserts
84
again a similar row, and a trx with an even bigger id delete marks it. Then
85
the position of the row should change in the index if the trx id affects
86
the alphabetical ordering.
88
Fix 2: If an insert encounters a similar row marked deleted, we turn the
89
insert into an 'update' of the row marked deleted. Then we must write undo
90
info on the update. A problem: what if a purge operation tries to remove
91
the delete marked row?
93
We can think of the database row versions as a linked list which starts
94
from the record in the clustered index, and is linked by roll ptrs
95
through undo logs. The secondary index records are references which tell
96
what kinds of records can be found in this linked list for a record
97
in the clustered index.
99
How to do the purge? A record can be removed from the clustered index
100
if its linked list becomes empty, i.e., the row has been marked deleted
101
and its roll ptr points to the record in the undo log we are going through,
102
doing the purge. Similarly, during a rollback, a record can be removed
103
if the stored roll ptr in the undo log points to a trx already (being) purged,
104
or if the roll ptr is NULL, i.e., it was a fresh insert. */
106
/************************************************************************
107
Creates a row undo node to a query graph. */
110
row_undo_node_create(
111
/*=================*/
112
/* out, own: undo node */
113
trx_t* trx, /* in: transaction */
114
que_thr_t* parent, /* in: parent node, i.e., a thr node */
115
mem_heap_t* heap) /* in: memory heap where created */
119
ut_ad(trx && parent && heap);
121
undo = mem_heap_alloc(heap, sizeof(undo_node_t));
123
undo->common.type = QUE_NODE_UNDO;
124
undo->common.parent = parent;
126
undo->state = UNDO_NODE_FETCH_NEXT;
129
btr_pcur_init(&(undo->pcur));
131
undo->heap = mem_heap_create(256);
136
/***************************************************************
137
Looks for the clustered index record when node has the row reference.
138
The pcur in node is used in the search. If found, stores the row to node,
139
and stores the position of pcur, and detaches it. The pcur must be closed
140
by the caller in any case. */
143
row_undo_search_clust_to_pcur(
144
/*==========================*/
145
/* out: TRUE if found; NOTE the node->pcur
146
must be closed by the caller, regardless of
148
undo_node_t* node) /* in: row undo node */
150
dict_index_t* clust_index;
155
mem_heap_t* heap = NULL;
156
ulint offsets_[REC_OFFS_NORMAL_SIZE];
157
ulint* offsets = offsets_;
158
rec_offs_init(offsets_);
162
clust_index = dict_table_get_first_index(node->table);
164
found = row_search_on_row_ref(&(node->pcur), BTR_MODIFY_LEAF,
165
node->table, node->ref, &mtr);
167
rec = btr_pcur_get_rec(&(node->pcur));
169
offsets = rec_get_offsets(rec, clust_index, offsets,
170
ULINT_UNDEFINED, &heap);
172
if (!found || 0 != ut_dulint_cmp(node->roll_ptr,
173
row_get_rec_roll_ptr(rec, clust_index,
176
/* We must remove the reservation on the undo log record
177
BEFORE releasing the latch on the clustered index page: this
178
is to make sure that some thread will eventually undo the
179
modification corresponding to node->roll_ptr. */
181
/* fputs("--------------------undoing a previous version\n",
186
node->row = row_build(ROW_COPY_DATA, clust_index, rec,
187
offsets, NULL, &node->ext, node->heap);
189
node->undo_row = dtuple_copy(node->row, node->heap);
190
row_upd_replace(node->undo_row, &node->undo_ext,
191
clust_index, node->update, node->heap);
193
node->undo_row = NULL;
194
node->undo_ext = NULL;
197
btr_pcur_store_position(&(node->pcur), &mtr);
202
btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
204
if (UNIV_LIKELY_NULL(heap)) {
210
/***************************************************************
211
Fetches an undo log record and does the undo for the recorded operation.
212
If none left, or a partial rollback completed, returns control to the
213
parent node, which is always a query thread node. */
218
/* out: DB_SUCCESS if operation successfully
219
completed, else error code */
220
undo_node_t* node, /* in: row undo node */
221
que_thr_t* thr) /* in: query thread */
226
ibool locked_data_dict;
232
if (node->state == UNDO_NODE_FETCH_NEXT) {
234
node->undo_rec = trx_roll_pop_top_rec_of_trx(trx,
238
if (!node->undo_rec) {
239
/* Rollback completed for this query thread */
241
thr->run_node = que_node_get_parent(node);
246
node->roll_ptr = roll_ptr;
247
node->undo_no = trx_undo_rec_get_undo_no(node->undo_rec);
249
if (trx_undo_roll_ptr_is_insert(roll_ptr)) {
251
node->state = UNDO_NODE_INSERT;
253
node->state = UNDO_NODE_MODIFY;
256
} else if (node->state == UNDO_NODE_PREV_VERS) {
258
/* Undo should be done to the same clustered index record
259
again in this same rollback, restoring the previous version */
261
roll_ptr = node->new_roll_ptr;
263
node->undo_rec = trx_undo_get_undo_rec_low(roll_ptr,
265
node->roll_ptr = roll_ptr;
266
node->undo_no = trx_undo_rec_get_undo_no(node->undo_rec);
268
if (trx_undo_roll_ptr_is_insert(roll_ptr)) {
270
node->state = UNDO_NODE_INSERT;
272
node->state = UNDO_NODE_MODIFY;
276
/* Prevent DROP TABLE etc. while we are rolling back this row.
277
If we are doing a TABLE CREATE or some other dictionary operation,
278
then we already have dict_operation_lock locked in x-mode. Do not
279
try to lock again, because that would cause a hang. */
281
locked_data_dict = (trx->dict_operation_lock_mode == 0);
283
if (locked_data_dict) {
285
row_mysql_lock_data_dictionary(trx);
288
if (node->state == UNDO_NODE_INSERT) {
290
err = row_undo_ins(node);
292
node->state = UNDO_NODE_FETCH_NEXT;
294
ut_ad(node->state == UNDO_NODE_MODIFY);
295
err = row_undo_mod(node, thr);
298
if (locked_data_dict) {
300
row_mysql_unlock_data_dictionary(trx);
303
/* Do some cleanup */
304
btr_pcur_close(&(node->pcur));
306
mem_heap_empty(node->heap);
308
thr->run_node = node;
313
/***************************************************************
314
Undoes a row operation in a table. This is a high-level function used
315
in SQL execution graphs. */
320
/* out: query thread to run next or NULL */
321
que_thr_t* thr) /* in: query thread */
329
srv_activity_count++;
331
trx = thr_get_trx(thr);
333
node = thr->run_node;
335
ut_ad(que_node_get_type(node) == QUE_NODE_UNDO);
337
err = row_undo(node, thr);
339
trx->error_state = err;
341
if (err != DB_SUCCESS) {
342
/* SQL error detected */
344
fprintf(stderr, "InnoDB: Fatal error %lu in rollback.\n",
347
if (err == DB_OUT_OF_FILE_SPACE) {
349
"InnoDB: Error 13 means out of tablespace.\n"
350
"InnoDB: Consider increasing"
351
" your tablespace.\n");