dtuple_t*
row_rec_to_index_entry(
/*===================*/
/* out, own: index entry built; see the
NOTE below! */
ulint type, /* in: ROW_COPY_DATA, or ROW_COPY_POINTERS:
the former copies also the data fields to
heap as the latter only places pointers to
data fields on the index page */
dict_index_t* index, /* in: index */
rec_t* rec, /* in: record in the index;
NOTE: in the case ROW_COPY_POINTERS
the data fields in the row will point
directly into this record, therefore,
the buffer page of this record must be
at least s-latched and the latch held
as long as the dtuple is used! */
mem_heap_t* heap) /* in: memory heap from which the memory
needed is allocated */
{
dtuple_t* entry;
dfield_t* dfield;
ulint i;
byte* field;
ulint len;
ulint rec_len;
byte* buf;
ut_ad(rec && heap && index);
if (type == ROW_COPY_DATA) {
/* Take a copy of rec to heap */
buf = mem_heap_alloc(heap, rec_get_size(rec));
rec = rec_copy(buf, rec);
}
rec_len = rec_get_n_fields(rec);
entry = dtuple_create(heap, rec_len);
dtuple_set_n_fields_cmp(entry,
dict_index_get_n_unique_in_tree(index));
ut_ad(rec_len == dict_index_get_n_fields(index));
dict_index_copy_types(entry, index, rec_len);
dtuple_set_info_bits(entry, rec_get_info_bits(rec));
for (i = 0; i < rec_len; i++) {
dfield = dtuple_get_nth_field(entry, i);
field = rec_get_nth_field(rec, i, &len);
dfield_set_data(dfield, field, len);
}
ut_ad(dtuple_check_typed(entry));
return(entry);
}
dtuple_t*
row_build_index_entry(
/*==================*/
/* out: index entry which should be inserted */
dtuple_t* row, /* in: row which should be inserted to the
table */
dict_index_t* index, /* in: index on the table */
mem_heap_t* heap) /* in: memory heap from which the memory for
the index entry is allocated */
{
dtuple_t* entry;
ulint entry_len;
dict_field_t* ind_field;
dfield_t* dfield;
dfield_t* dfield2;
dict_col_t* col;
ulint i;
ut_ad(row && index && heap);
ut_ad(dtuple_check_typed(row));
entry_len = dict_index_get_n_fields(index);
entry = dtuple_create(heap, entry_len);
if (index->type & DICT_UNIVERSAL) {
dtuple_set_n_fields_cmp(entry, entry_len);
} else {
dtuple_set_n_fields_cmp(entry,
dict_index_get_n_unique_in_tree(index));
}
for (i = 0; i < entry_len; i++) {
ind_field = dict_index_get_nth_field(index, i);
col = ind_field->col;
dfield = dtuple_get_nth_field(entry, i);
dfield2 = dtuple_get_nth_field(row, dict_col_get_no(col));
dfield_copy(dfield, dfield2);
/* If a column prefix index, take only the prefix */
if (ind_field->prefix_len > 0
&& dfield_get_len(dfield2) != UNIV_SQL_NULL
&& dfield_get_len(dfield2) > ind_field->prefix_len) {
dfield_set_len(dfield, ind_field->prefix_len);
}
}
ut_ad(dtuple_check_typed(entry));
return(entry);
}
/*******************************************************************//**
Converts an index record to a typed data tuple.
@return index entry built; does not set info_bits, and the data fields
in the entry will point directly to rec */
UNIV_INTERN
dtuple_t*
row_rec_to_index_entry_low(
/*=======================*/
const rec_t* rec, /*!< in: record in the index */
const dict_index_t* index, /*!< in: index */
const ulint* offsets,/*!< in: rec_get_offsets(rec, index) */
ulint* n_ext, /*!< out: number of externally
stored columns */
mem_heap_t* heap) /*!< in: memory heap from which
the memory needed is allocated */
{
dtuple_t* entry;
dfield_t* dfield;
ulint i;
const byte* field;
ulint len;
ulint rec_len;
ut_ad(rec && heap && index);
/* Because this function may be invoked by row0merge.c
on a record whose header is in different format, the check
rec_offs_validate(rec, index, offsets) must be avoided here. */
ut_ad(n_ext);
*n_ext = 0;
rec_len = rec_offs_n_fields(offsets);
entry = dtuple_create(heap, rec_len);
dtuple_set_n_fields_cmp(entry,
dict_index_get_n_unique_in_tree(index));
ut_ad(rec_len == dict_index_get_n_fields(index));
dict_index_copy_types(entry, index, rec_len);
for (i = 0; i < rec_len; i++) {
dfield = dtuple_get_nth_field(entry, i);
field = rec_get_nth_field(rec, offsets, i, &len);
dfield_set_data(dfield, field, len);
if (rec_offs_nth_extern(offsets, i)) {
dfield_set_ext(dfield);
(*n_ext)++;
}
}
ut_ad(dtuple_check_typed(entry));
return(entry);
}
/********************************************************//**
Opens a buffer for mlog, writes the initial log record and,
if needed, the field lengths of an index.
@return buffer, NULL if log mode MTR_LOG_NONE */
UNIV_INTERN
byte*
mlog_open_and_write_index(
/*======================*/
mtr_t* mtr, /*!< in: mtr */
const byte* rec, /*!< in: index record or page */
dict_index_t* index, /*!< in: record descriptor */
byte type, /*!< in: log item type */
ulint size) /*!< in: requested buffer size in bytes
(if 0, calls mlog_close() and returns NULL) */
{
byte* log_ptr;
const byte* log_start;
const byte* log_end;
ut_ad(!!page_rec_is_comp(rec) == dict_table_is_comp(index->table));
if (!page_rec_is_comp(rec)) {
log_start = log_ptr = mlog_open(mtr, 11 + size);
if (!log_ptr) {
return(NULL); /* logging is disabled */
}
log_ptr = mlog_write_initial_log_record_fast(rec, type,
log_ptr, mtr);
log_end = log_ptr + 11 + size;
} else {
ulint i;
ulint n = dict_index_get_n_fields(index);
/* total size needed */
ulint total = 11 + size + (n + 2) * 2;
ulint alloc = total;
/* allocate at most DYN_ARRAY_DATA_SIZE at a time */
if (alloc > DYN_ARRAY_DATA_SIZE) {
alloc = DYN_ARRAY_DATA_SIZE;
}
log_start = log_ptr = mlog_open(mtr, alloc);
if (!log_ptr) {
return(NULL); /* logging is disabled */
}
log_end = log_ptr + alloc;
log_ptr = mlog_write_initial_log_record_fast(rec, type,
log_ptr, mtr);
mach_write_to_2(log_ptr, n);
log_ptr += 2;
mach_write_to_2(log_ptr,
dict_index_get_n_unique_in_tree(index));
log_ptr += 2;
for (i = 0; i < n; i++) {
dict_field_t* field;
const dict_col_t* col;
ulint len;
field = dict_index_get_nth_field(index, i);
col = dict_field_get_col(field);
len = field->fixed_len;
ut_ad(len < 0x7fff);
if (len == 0
&& (col->len > 255 || col->mtype == DATA_BLOB)) {
/* variable-length field
with maximum length > 255 */
len = 0x7fff;
}
if (col->prtype & DATA_NOT_NULL) {
len |= 0x8000;
}
if (log_ptr + 2 > log_end) {
mlog_close(mtr, log_ptr);
ut_a(total > (ulint) (log_ptr - log_start));
total -= log_ptr - log_start;
alloc = total;
if (alloc > DYN_ARRAY_DATA_SIZE) {
alloc = DYN_ARRAY_DATA_SIZE;
}
log_start = log_ptr = mlog_open(mtr, alloc);
if (!log_ptr) {
return(NULL); /* logging is disabled */
}
log_end = log_ptr + alloc;
}
mach_write_to_2(log_ptr, len);
log_ptr += 2;
}
}
if (size == 0) {
mlog_close(mtr, log_ptr);
log_ptr = NULL;
} else if (log_ptr + size > log_end) {
mlog_close(mtr, log_ptr);
log_ptr = mlog_open(mtr, size);
}
return(log_ptr);
}
big_rec_t*
dtuple_convert_big_rec(
/*===================*/
/* out, own: created big record vector,
NULL if we are not able to shorten
the entry enough, i.e., if there are
too many short fields in entry */
dict_index_t* index, /* in: index */
dtuple_t* entry, /* in: index entry */
ulint* ext_vec,/* in: array of externally stored fields,
or NULL: if a field already is externally
stored, then we cannot move it to the vector
this function returns */
ulint n_ext_vec)/* in: number of elements is ext_vec */
{
mem_heap_t* heap;
big_rec_t* vector;
dfield_t* dfield;
ulint size;
ulint n_fields;
ulint longest;
ulint longest_i = ULINT_MAX;
ibool is_externally_stored;
ulint i;
ulint j;
ut_a(dtuple_check_typed_no_assert(entry));
size = rec_get_converted_size(index, entry);
if (UNIV_UNLIKELY(size > 1000000000)) {
fprintf(stderr,
"InnoDB: Warning: tuple size very big: %lu\n", (ulong) size);
fputs("InnoDB: Tuple contents: ", stderr);
dtuple_print(stderr, entry);
putc('\n', stderr);
}
heap = mem_heap_create(size + dtuple_get_n_fields(entry)
* sizeof(big_rec_field_t) + 1000);
vector = mem_heap_alloc(heap, sizeof(big_rec_t));
vector->heap = heap;
vector->fields = mem_heap_alloc(heap, dtuple_get_n_fields(entry)
* sizeof(big_rec_field_t));
/* Decide which fields to shorten: the algorithm is to look for
the longest field whose type is DATA_BLOB */
n_fields = 0;
while (rec_get_converted_size(index, entry)
>= ut_min(page_get_free_space_of_empty(
index->table->comp) / 2,
REC_MAX_DATA_SIZE)) {
longest = 0;
for (i = dict_index_get_n_unique_in_tree(index);
i < dtuple_get_n_fields(entry); i++) {
/* Skip over fields which already are externally
stored */
is_externally_stored = FALSE;
if (ext_vec) {
for (j = 0; j < n_ext_vec; j++) {
if (ext_vec[j] == i) {
is_externally_stored = TRUE;
}
}
}
if (!is_externally_stored) {
dfield = dtuple_get_nth_field(entry, i);
if (dfield->len != UNIV_SQL_NULL &&
dfield->len > longest) {
longest = dfield->len;
longest_i = i;
}
}
}
/* We do not store externally fields which are smaller than
DICT_MAX_INDEX_COL_LEN */
ut_a(DICT_MAX_INDEX_COL_LEN > REC_1BYTE_OFFS_LIMIT);
if (longest < BTR_EXTERN_FIELD_REF_SIZE + 10
+ DICT_MAX_INDEX_COL_LEN) {
/* Cannot shorten more */
mem_heap_free(heap);
return(NULL);
}
/* Move data from field longest_i to big rec vector;
we do not let data size of the remaining entry
drop below 128 which is the limit for the 2-byte
offset storage format in a physical record. This
we accomplish by storing 128 bytes of data in entry
itself, and only the remaining part to big rec vec.
We store the first bytes locally to the record. Then
we can calculate all ordering fields in all indexes
from locally stored data. */
dfield = dtuple_get_nth_field(entry, longest_i);
vector->fields[n_fields].field_no = longest_i;
ut_a(dfield->len > DICT_MAX_INDEX_COL_LEN);
vector->fields[n_fields].len = dfield->len
- DICT_MAX_INDEX_COL_LEN;
vector->fields[n_fields].data = mem_heap_alloc(heap,
vector->fields[n_fields].len);
/* Copy data (from the end of field) to big rec vector */
ut_memcpy(vector->fields[n_fields].data,
((byte*)dfield->data) + dfield->len
- vector->fields[n_fields].len,
vector->fields[n_fields].len);
dfield->len = dfield->len - vector->fields[n_fields].len
+ BTR_EXTERN_FIELD_REF_SIZE;
/* Set the extern field reference in dfield to zero */
memset(((byte*)dfield->data)
+ dfield->len - BTR_EXTERN_FIELD_REF_SIZE,
0, BTR_EXTERN_FIELD_REF_SIZE);
n_fields++;
}
vector->n_fields = n_fields;
return(vector);
}
/**************************************************************//**
Moves parts of long fields in entry to the big record vector so that
the size of tuple drops below the maximum record size allowed in the
database. Moves data only from those fields which are not necessary
to determine uniquely the insertion place of the tuple in the index.
@return own: created big record vector, NULL if we are not able to
shorten the entry enough, i.e., if there are too many fixed-length or
short fields in entry or the index is clustered */
UNIV_INTERN
big_rec_t*
dtuple_convert_big_rec(
/*===================*/
dict_index_t* index, /*!< in: index */
dtuple_t* entry, /*!< in/out: index entry */
ulint* n_ext) /*!< in/out: number of
externally stored columns */
{
mem_heap_t* heap;
big_rec_t* vector;
dfield_t* dfield;
dict_field_t* ifield;
ulint size;
ulint n_fields;
ulint local_len;
ulint local_prefix_len;
if (UNIV_UNLIKELY(!dict_index_is_clust(index))) {
return(NULL);
}
if (dict_table_get_format(index->table) < DICT_TF_FORMAT_ZIP) {
/* up to MySQL 5.1: store a 768-byte prefix locally */
local_len = BTR_EXTERN_FIELD_REF_SIZE + DICT_MAX_INDEX_COL_LEN;
} else {
/* new-format table: do not store any BLOB prefix locally */
local_len = BTR_EXTERN_FIELD_REF_SIZE;
}
ut_a(dtuple_check_typed_no_assert(entry));
size = rec_get_converted_size(index, entry, *n_ext);
if (UNIV_UNLIKELY(size > 1000000000)) {
fprintf(stderr,
"InnoDB: Warning: tuple size very big: %lu\n",
(ulong) size);
fputs("InnoDB: Tuple contents: ", stderr);
dtuple_print(stderr, entry);
putc('\n', stderr);
}
heap = mem_heap_create(size + dtuple_get_n_fields(entry)
* sizeof(big_rec_field_t) + 1000);
vector = mem_heap_alloc(heap, sizeof(big_rec_t));
vector->heap = heap;
vector->fields = mem_heap_alloc(heap, dtuple_get_n_fields(entry)
* sizeof(big_rec_field_t));
/* Decide which fields to shorten: the algorithm is to look for
a variable-length field that yields the biggest savings when
stored externally */
n_fields = 0;
while (page_zip_rec_needs_ext(rec_get_converted_size(index, entry,
*n_ext),
dict_table_is_comp(index->table),
dict_index_get_n_fields(index),
dict_table_zip_size(index->table))) {
ulint i;
ulint longest = 0;
ulint longest_i = ULINT_MAX;
byte* data;
big_rec_field_t* b;
for (i = dict_index_get_n_unique_in_tree(index);
i < dtuple_get_n_fields(entry); i++) {
ulint savings;
dfield = dtuple_get_nth_field(entry, i);
ifield = dict_index_get_nth_field(index, i);
/* Skip fixed-length, NULL, externally stored,
or short columns */
if (ifield->fixed_len
|| dfield_is_null(dfield)
|| dfield_is_ext(dfield)
|| dfield_get_len(dfield) <= local_len
|| dfield_get_len(dfield)
<= BTR_EXTERN_FIELD_REF_SIZE * 2) {
goto skip_field;
}
savings = dfield_get_len(dfield) - local_len;
/* Check that there would be savings */
if (longest >= savings) {
goto skip_field;
}
longest_i = i;
longest = savings;
skip_field:
continue;
}
if (!longest) {
/* Cannot shorten more */
mem_heap_free(heap);
return(NULL);
}
/* Move data from field longest_i to big rec vector.
We store the first bytes locally to the record. Then
we can calculate all ordering fields in all indexes
from locally stored data. */
dfield = dtuple_get_nth_field(entry, longest_i);
ifield = dict_index_get_nth_field(index, longest_i);
local_prefix_len = local_len - BTR_EXTERN_FIELD_REF_SIZE;
b = &vector->fields[n_fields];
b->field_no = longest_i;
b->len = dfield_get_len(dfield) - local_prefix_len;
b->data = (char*) dfield_get_data(dfield) + local_prefix_len;
/* Allocate the locally stored part of the column. */
data = mem_heap_alloc(heap, local_len);
/* Copy the local prefix. */
memcpy(data, dfield_get_data(dfield), local_prefix_len);
/* Clear the extern field reference (BLOB pointer). */
memset(data + local_prefix_len, 0, BTR_EXTERN_FIELD_REF_SIZE);
#if 0
/* The following would fail the Valgrind checks in
page_cur_insert_rec_low() and page_cur_insert_rec_zip().
The BLOB pointers in the record will be initialized after
the record and the BLOBs have been written. */
UNIV_MEM_ALLOC(data + local_prefix_len,
BTR_EXTERN_FIELD_REF_SIZE);
#endif
dfield_set_data(dfield, data, local_len);
dfield_set_ext(dfield);
n_fields++;
(*n_ext)++;
ut_ad(n_fields < dtuple_get_n_fields(entry));
}
vector->n_fields = n_fields;
return(vector);
}
/*****************************************************************//**
When an insert or purge to a table is performed, this function builds
the entry to be inserted into or purged from an index on the table.
@return index entry which should be inserted or purged, or NULL if the
externally stored columns in the clustered index record are
unavailable and ext != NULL */
UNIV_INTERN
dtuple_t*
row_build_index_entry(
/*==================*/
const dtuple_t* row, /*!< in: row which should be
inserted or purged */
row_ext_t* ext, /*!< in: externally stored column prefixes,
or NULL */
dict_index_t* index, /*!< in: index on the table */
mem_heap_t* heap) /*!< in: memory heap from which the memory for
the index entry is allocated */
{
dtuple_t* entry;
ulint entry_len;
ulint i;
ut_ad(row && index && heap);
ut_ad(dtuple_check_typed(row));
entry_len = dict_index_get_n_fields(index);
entry = dtuple_create(heap, entry_len);
if (UNIV_UNLIKELY(index->type & DICT_UNIVERSAL)) {
dtuple_set_n_fields_cmp(entry, entry_len);
/* There may only be externally stored columns
in a clustered index B-tree of a user table. */
ut_a(!ext);
} else {
dtuple_set_n_fields_cmp(
entry, dict_index_get_n_unique_in_tree(index));
}
for (i = 0; i < entry_len; i++) {
const dict_field_t* ind_field
= dict_index_get_nth_field(index, i);
const dict_col_t* col
= ind_field->col;
ulint col_no
= dict_col_get_no(col);
dfield_t* dfield
= dtuple_get_nth_field(entry, i);
const dfield_t* dfield2
= dtuple_get_nth_field(row, col_no);
ulint len
= dfield_get_len(dfield2);
dfield_copy(dfield, dfield2);
if (dfield_is_null(dfield)) {
continue;
}
if (ind_field->prefix_len == 0
&& (!dfield_is_ext(dfield)
|| dict_index_is_clust(index))) {
/* The dfield_copy() above suffices for
columns that are stored in-page, or for
clustered index record columns that are not
part of a column prefix in the PRIMARY KEY. */
continue;
}
/* If the column is stored externally (off-page) in
the clustered index, it must be an ordering field in
the secondary index. In the Antelope format, only
prefix-indexed columns may be stored off-page in the
clustered index record. In the Barracuda format, also
fully indexed long CHAR or VARCHAR columns may be
stored off-page. */
ut_ad(col->ord_part);
if (UNIV_LIKELY_NULL(ext)) {
/* See if the column is stored externally. */
const byte* buf = row_ext_lookup(ext, col_no,
&len);
if (UNIV_LIKELY_NULL(buf)) {
if (UNIV_UNLIKELY(buf == field_ref_zero)) {
return(NULL);
}
dfield_set_data(dfield, buf, len);
}
if (ind_field->prefix_len == 0) {
/* In the Barracuda format
(ROW_FORMAT=DYNAMIC or
ROW_FORMAT=COMPRESSED), we can have a
secondary index on an entire column
that is stored off-page in the
clustered index. As this is not a
prefix index (prefix_len == 0),
include the entire off-page column in
the secondary index record. */
continue;
}
} else if (dfield_is_ext(dfield)) {
/* This table is either in Antelope format
(ROW_FORMAT=REDUNDANT or ROW_FORMAT=COMPACT)
or a purge record where the ordered part of
the field is not external.
In Antelope, the maximum column prefix
index length is 767 bytes, and the clustered
index record contains a 768-byte prefix of
each off-page column. */
ut_a(len >= BTR_EXTERN_FIELD_REF_SIZE);
len -= BTR_EXTERN_FIELD_REF_SIZE;
dfield_set_len(dfield, len);
}
/* If a column prefix index, take only the prefix. */
if (ind_field->prefix_len) {
len = dtype_get_at_most_n_mbchars(
col->prtype, col->mbminlen, col->mbmaxlen,
ind_field->prefix_len, len,
dfield_get_data(dfield));
dfield_set_len(dfield, len);
}
}
ut_ad(dtuple_check_typed(entry));
return(entry);
}
/*********************************************************************//**
Updates the search info of an index about hash successes. NOTE that info
is NOT protected by any semaphore, to save CPU time! Do not assume its fields
are consistent. */
static
void
btr_search_info_update_hash(
/*========================*/
btr_search_t* info, /*!< in/out: search info */
btr_cur_t* cursor) /*!< in: cursor which was just positioned */
{
dict_index_t* index;
ulint n_unique;
int cmp;
#ifdef UNIV_SYNC_DEBUG
ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_SHARED));
ut_ad(!rw_lock_own(&btr_search_latch, RW_LOCK_EX));
#endif /* UNIV_SYNC_DEBUG */
index = cursor->index;
if (dict_index_is_ibuf(index)) {
/* So many deletes are performed on an insert buffer tree
that we do not consider a hash index useful on it: */
return;
}
n_unique = dict_index_get_n_unique_in_tree(index);
if (info->n_hash_potential == 0) {
goto set_new_recomm;
}
/* Test if the search would have succeeded using the recommended
hash prefix */
if (info->n_fields >= n_unique && cursor->up_match >= n_unique) {
increment_potential:
info->n_hash_potential++;
return;
}
cmp = ut_pair_cmp(info->n_fields, info->n_bytes,
cursor->low_match, cursor->low_bytes);
if (info->left_side ? cmp <= 0 : cmp > 0) {
goto set_new_recomm;
}
cmp = ut_pair_cmp(info->n_fields, info->n_bytes,
cursor->up_match, cursor->up_bytes);
if (info->left_side ? cmp <= 0 : cmp > 0) {
goto increment_potential;
}
set_new_recomm:
/* We have to set a new recommendation; skip the hash analysis
for a while to avoid unnecessary CPU time usage when there is no
chance for success */
info->hash_analysis = 0;
cmp = ut_pair_cmp(cursor->up_match, cursor->up_bytes,
cursor->low_match, cursor->low_bytes);
if (cmp == 0) {
info->n_hash_potential = 0;
/* For extra safety, we set some sensible values here */
info->n_fields = 1;
info->n_bytes = 0;
info->left_side = TRUE;
} else if (cmp > 0) {
info->n_hash_potential = 1;
if (cursor->up_match >= n_unique) {
info->n_fields = n_unique;
info->n_bytes = 0;
} else if (cursor->low_match < cursor->up_match) {
info->n_fields = cursor->low_match + 1;
info->n_bytes = 0;
} else {
info->n_fields = cursor->low_match;
info->n_bytes = cursor->low_bytes + 1;
}
info->left_side = TRUE;
} else {
info->n_hash_potential = 1;
if (cursor->low_match >= n_unique) {
info->n_fields = n_unique;
info->n_bytes = 0;
} else if (cursor->low_match > cursor->up_match) {
info->n_fields = cursor->up_match + 1;
info->n_bytes = 0;
} else {
info->n_fields = cursor->up_match;
info->n_bytes = cursor->up_bytes + 1;
}
info->left_side = FALSE;
}
}
/********************************************************//**
Opens a buffer for mlog, writes the initial log record and,
if needed, the field lengths of an index.
@return buffer, NULL if log mode MTR_LOG_NONE */
UNIV_INTERN
byte*
mlog_open_and_write_index(
/*======================*/
mtr_t* mtr, /*!< in: mtr */
const byte* rec, /*!< in: index record or page */
dict_index_t* index, /*!< in: record descriptor */
byte type, /*!< in: log item type */
ulint size) /*!< in: requested buffer size in bytes
(if 0, calls mlog_close() and returns NULL) */
{
byte* log_ptr;
const byte* log_start;
const byte* log_end;
ut_ad(!!page_rec_is_comp(rec) == dict_table_is_comp(index->table));
if (!page_rec_is_comp(rec)) {
log_start = log_ptr = mlog_open(mtr, 11 + size);
if (!log_ptr) {
return(NULL); /* logging is disabled */
}
log_ptr = mlog_write_initial_log_record_fast(rec, type,
log_ptr, mtr);
log_end = log_ptr + 11 + size;
} else {
ulint i;
ulint n = dict_index_get_n_fields(index);
ibool is_gcs_cluster = dict_index_is_gcs_clust_after_alter_table(index);
/* total size needed */
/* redo日志有可能需要多两字节，total需要根据实际情况分配空间！ */
ulint total = 11 + size + (n + 2 + (is_gcs_cluster ? 1 : 0)) * 2;
ulint alloc = total;
/* allocate at most DYN_ARRAY_DATA_SIZE at a time */
if (alloc > DYN_ARRAY_DATA_SIZE) {
alloc = DYN_ARRAY_DATA_SIZE;
}
log_start = log_ptr = mlog_open(mtr, alloc);
if (!log_ptr) {
return(NULL); /* logging is disabled */
}
log_end = log_ptr + alloc;
log_ptr = mlog_write_initial_log_record_fast(rec, type,
log_ptr, mtr);
/* 在第一次alter table前，所有gcs表可以当成compact表，即使是redo过程，这样redo log也兼容！ */
if (is_gcs_cluster)
{
//ut_ad(rec_is_gcs(rec) || rec == page_align(rec) || rec_get_status(rec) & REC_STATUS_NODE_PTR); /* rec有可能就是页头，如日志MLOG_COMP_LIST_END_COPY_CREATED */
mach_write_to_2(log_ptr, n | 0x8000); /* 标记是gcs表 */
}
else
{
ut_ad(rec == page_align(rec) || !rec_is_gcs(rec));
mach_write_to_2(log_ptr, n);
}
/* 对于gcs聚集索引，记录第一次alter table前聚集索引的字段数 */
if (is_gcs_cluster)
{
log_ptr += 2;
mach_write_to_2(log_ptr, (ulint)index->n_fields_before_alter);
ut_ad(!index->n_fields_before_alter == !dict_index_is_gcs_clust_after_alter_table(index));
}
log_ptr += 2;
mach_write_to_2(log_ptr,
dict_index_get_n_unique_in_tree(index));
log_ptr += 2;
for (i = 0; i < n; i++) {
dict_field_t* field;
const dict_col_t* col;
ulint len;
field = dict_index_get_nth_field(index, i);
col = dict_field_get_col(field);
len = field->fixed_len;
ut_ad(len < 0x7fff);
if (len == 0
&& (col->len > 255 || col->mtype == DATA_BLOB)) {
/* variable-length field
with maximum length > 255 */
len = 0x7fff;
}
if (col->prtype & DATA_NOT_NULL) {
len |= 0x8000;
}
if (log_ptr + 2 > log_end) {
mlog_close(mtr, log_ptr);
ut_a(total > (ulint) (log_ptr - log_start));
total -= log_ptr - log_start;
alloc = total;
if (alloc > DYN_ARRAY_DATA_SIZE) {
alloc = DYN_ARRAY_DATA_SIZE;
}
log_start = log_ptr = mlog_open(mtr, alloc);
if (!log_ptr) {
return(NULL); /* logging is disabled */
}
log_end = log_ptr + alloc;
}
mach_write_to_2(log_ptr, len);
log_ptr += 2;
}
}
if (size == 0) {
mlog_close(mtr, log_ptr);
log_ptr = NULL;
} else if (log_ptr + size > log_end) {
mlog_close(mtr, log_ptr);
log_ptr = mlog_open(mtr, size);
}
return(log_ptr);
}
