mirror/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-11-04 04:46:15 +01:00
Code Issues Projects Releases Packages Wiki Activity
mariadb/sql/rpl_utility_server.cc
Monty 232533978f MDEV-36290: ALTER TABLE with multi-master can cause data loss 
One can have data loss in multi-master setups when 1) both masters
update the same table, 2) ALTER TABLE is run on one master which
re-arranges the column ordering, and 3) transactions are binlogged
in ROW binlog_format.

This is because the slave assumes that all columns are in the same
order on the master and slave and all columns on the master also
exists on the slave. This happens even if binlog_row_metadata=FULL is
used.  If this is not the case, this will lead to silent data loss.

A new option for slave_type_conversions bit field,
ERROR_IF_MISSING_FIELD, has been added. This allows the user to define if
the slave should abort replication if it is missing some field that
existed on the master. This option is off by default to keep things
compatible with earlier versions.
If a field is missing on the slave and log_warnings >= 1, a warning
will be logged to the error log.

This patch fixes this, when binlog_row_metadata=FULL is used on the
master, by mapping fields with identical names on the master and slave.
If slave has fields that does not exist in the row event, these will
be set to their default value.

The main idea is that we added two conversion tables:
m_tabledef.master_to_slave_map[master_column_index] -> slave_column_index
and m_tabledef.master_to_slave_error[master_column_index] which contains
an error number if the master_column does not exists on the slave or
it is not possible to convert the master data to the slave column.
master_to_slave_error[#] contains 0 if the column exists and is compatible.

General code changes:
- Instead of looping over row fields in the order of slave table
  we are instead looping over fields in the order of the binary log.
- We are using table->write_set to know which fields should be updated
  on the slave. This is reflected in unpack_row
- We are calling TABLE::mark_columns_per_binlog_row_image() to ensure
  that rpl_write_set is properly set. This is needed if the slave also
  is doing binary logging.
- Before replication aborted if the master and slave tables where too
  different.  Now replication is only aborted if the row actually uses
  columns that does not exists on the slave (and ALLOW_MISSING_FIELDS
  is not used) or uses columns that cannot be converted.
  - Instead of giving errors in compatible_with(), used when table is
    accessed by first the row event, we are instead giving errors
    when we examine a row event and notice that it is accessing
    a not existing or not compatible field.

Other code changes:
- Removed conv_table argument from compatible_with() and store it
  directly in RPL_TABLE_LIST->m_conv_table
- table_def::compatible_with() returns now 1 on error (not 0).
- Remove m_width and skip arguments from prepare_record() as we are
  now using table->write_set() to check which elements need a default
  value.
- Moved DBUG_ENTER() to it's proper place (after variable
  declarations) in a few functions.
- Some changes in unpack_row():
  - Replaced null_mask and null_ptr with an indexed bit check for
    simplicity.
  - Removed check of rgi == null and table_found which never worked.
  - Updated comments to reflect current code.
  - Indentation changes as the code now uses 'continue' instead of
    'if-else' in the main loop.
  - The code to throw away 'extra master fields' is not needed as we
    are now looping over fields in binary log, not over fields in
    slave table.
- fill_extra_persistent_columns() is now using table->cond_set to know
  which columns where not updated from binlog.
- Simplified get_table_data(TABLE *table_arg) by returning found
  table_list.
- Errors for row events are now initialized in compatible_with(),
  checked in check_wrong_column_usage() and reported in
  give_compatibility_error().

Test cases and some code patchs provide by Brandon Nesterenko
<brandon.nesterenko@mariadb.com>
2025-05-19 19:57:47 +03:00

1368 lines
46 KiB
C++
Raw Permalink Blame History

/* Copyright (c) 2006, 2013, Oracle and/or its affiliates.
Copyright (c) 2011, 2013, Monty Program Ab
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1335 USA */
#include "mariadb.h"
#include <my_bit.h>
#include "rpl_utility.h"
#include "log_event.h"
#if defined(MYSQL_CLIENT)
#error MYSQL_CLIENT must not be defined here
#endif
#if !defined(MYSQL_SERVER)
#error MYSQL_SERVER must be defined here
#endif
#if defined(HAVE_REPLICATION)
#include "rpl_rli.h"
#include "sql_select.h"
#endif
/**
Compute the maximum display length of a field.
@param sql_type Type of the field
@param metadata The metadata from the master for the field.
@return Maximum length of the field in bytes.
The precise values calculated by field->max_display_length() and
calculated by max_display_length_for_field() can differ (by +1 or -1)
for integer data types (TINYINT, SMALLINT, MEDIUMINT, INT, BIGINT).
This slight difference is not important here, because we call
this function only for two *different* integer data types.
*/
static uint32
max_display_length_for_field(const Conv_source &source)
{
DBUG_PRINT("debug", ("sql_type: %s, metadata: 0x%x",
source.type_handler()->name().ptr(), source.metadata()));
return source.type_handler()->max_display_length_for_field(source);
}
/*
Compare the pack lengths of a source field (on the master) and a
target field (on the slave).
@param sh Source type handler
@param source_length Source length
@param th Target type hander
@param target_length Target length
@retval CONV_TYPE_SUBSET_TO_SUPERSET The length of the source field is
smaller than the target field.
@retval CONV_TYPE_PRECISE The length of the source and
the target fields are equal.
@retval CONV_TYPE_SUPERSET_TO_SUBSET The length of the source field is
greater than the target field.
*/
static enum_conv_type
compare_lengths(const Type_handler *sh, uint32 source_length,
const Type_handler *th, uint32 target_length)
{
DBUG_ENTER("compare_lengths");
DBUG_PRINT("debug", ("source_length: %lu, source_type: %s,"
" target_length: %lu, target_type: %s",
(unsigned long) source_length, sh->name().ptr(),
(unsigned long) target_length, th->name().ptr()));
enum_conv_type result=
source_length < target_length ? CONV_TYPE_SUBSET_TO_SUPERSET :
source_length > target_length ? CONV_TYPE_SUPERSET_TO_SUBSET :
CONV_TYPE_PRECISE;
DBUG_PRINT("result", ("%d", result));
DBUG_RETURN(result);
}
/**
Calculate display length for MySQL56 temporal data types from their metadata.
It contains fractional precision in the low 16-bit word.
*/
static uint32
max_display_length_for_temporal2_field(uint32 int_display_length,
unsigned int metadata)
{
metadata&= 0x00ff;
return int_display_length + metadata + (metadata ? 1 : 0);
}
uint32
Type_handler_newdecimal::max_display_length_for_field(const Conv_source &src)
const
{
return src.metadata() >> 8;
}
uint32
Type_handler_typelib::max_display_length_for_field(const Conv_source &src)
const
{
/*
Field_enum::rpl_conv_type_from() does not use compare_lengths().
So we should not come here.
*/
DBUG_ASSERT(0);
return src.metadata() & 0x00ff;
}
uint32
Type_handler_string::max_display_length_for_field(const Conv_source &src)
const
{
/*
ENUM and SET are transferred using as STRING,
with the exact type code in metadata.
Make sure that we previously detected ENUM/SET and
translated them into a proper type handler.
See table_def::field_type_handler() for details.
*/
DBUG_ASSERT((src.metadata() >> 8) != MYSQL_TYPE_SET);
DBUG_ASSERT((src.metadata() >> 8) != MYSQL_TYPE_ENUM);
/* This is taken from Field_string::unpack. */
return (((src.metadata() >> 4) & 0x300) ^ 0x300) + (src.metadata() & 0x00ff);
}
uint32
Type_handler_time2::max_display_length_for_field(const Conv_source &src)
const
{
return max_display_length_for_temporal2_field(MIN_TIME_WIDTH,
src.metadata());
}
uint32
Type_handler_timestamp2::max_display_length_for_field(const Conv_source &src)
const
{
return max_display_length_for_temporal2_field(MAX_DATETIME_WIDTH,
src.metadata());
}
uint32
Type_handler_datetime2::max_display_length_for_field(const Conv_source &src)
const
{
return max_display_length_for_temporal2_field(MAX_DATETIME_WIDTH,
src.metadata());
}
uint32
Type_handler_bit::max_display_length_for_field(const Conv_source &src)
const
{
/*
Decode the size of the bit field from the master.
*/
DBUG_ASSERT((src.metadata() & 0xff) <= 7);
return 8 * (src.metadata() >> 8U) + (src.metadata() & 0x00ff);
}
uint32
Type_handler_var_string::max_display_length_for_field(const Conv_source &src)
const
{
return src.metadata();
}
uint32
Type_handler_varchar::max_display_length_for_field(const Conv_source &src)
const
{
return src.metadata();
}
uint32
Type_handler_varchar_compressed::
max_display_length_for_field(const Conv_source &src) const
{
DBUG_ASSERT(src.metadata() > 0);
return src.metadata() - 1;
}
/*
The actual length for these types does not really matter since
they are used to calc_pack_length, which ignores the given
length for these types.
Since we want this to be accurate for other uses, we return the
maximum size in bytes of these BLOBs.
*/
uint32
Type_handler_tiny_blob::max_display_length_for_field(const Conv_source &src)
const
{
return (uint32) my_set_bits(1 * 8);
}
uint32
Type_handler_medium_blob::max_display_length_for_field(const Conv_source &src)
const
{
return (uint32) my_set_bits(3 * 8);
}
uint32
Type_handler_blob::max_display_length_for_field(const Conv_source &src)
const
{
/*
For the blob type, Field::real_type() lies and say that all
blobs are of type MYSQL_TYPE_BLOB. In that case, we have to look
at the length instead to decide what the max display size is.
*/
return (uint32) my_set_bits(src.metadata() * 8);
}
uint32
Type_handler_blob_compressed::max_display_length_for_field(const Conv_source &src)
const
{
return (uint32) my_set_bits(src.metadata() * 8);
}
uint32
Type_handler_long_blob::max_display_length_for_field(const Conv_source &src)
const
{
return (uint32) my_set_bits(4 * 8);
}
uint32
Type_handler_olddecimal::max_display_length_for_field(const Conv_source &src)
const
{
return ~(uint32) 0;
}
void Type_handler::show_binlog_type(const Conv_source &src, const Field &,
String *str) const
{
str->set_ascii(name().ptr(), name().length());
}
void Type_handler_var_string::show_binlog_type(const Conv_source &src,
const Field &dst,
String *str) const
{
CHARSET_INFO *cs= str->charset();
const char* fmt= dst.cmp_type() != STRING_RESULT || dst.has_charset()
? "char(%u octets)" : "binary(%u)";
size_t length= cs->cset->snprintf(cs, (char*) str->ptr(),
str->alloced_length(),
fmt, src.metadata());
str->length(length);
}
void Type_handler_varchar::show_binlog_type(const Conv_source &src,
const Field &dst,
String *str) const
{
CHARSET_INFO *cs= str->charset();
const char* fmt= dst.cmp_type() != STRING_RESULT || dst.has_charset()
? "varchar(%u octets)" : "varbinary(%u)";
size_t length= cs->cset->snprintf(cs, (char*) str->ptr(),
str->alloced_length(),
fmt, src.metadata());
str->length(length);
}
void Type_handler_varchar_compressed::show_binlog_type(const Conv_source &src,
const Field &dst,
String *str) const
{
CHARSET_INFO *cs= str->charset();
const char* fmt= dst.cmp_type() != STRING_RESULT || dst.has_charset()
? "varchar(%u octets) compressed" : "varbinary(%u) compressed";
size_t length= cs->cset->snprintf(cs, (char*) str->ptr(),
str->alloced_length(),
fmt, src.metadata());
str->length(length);
}
void Type_handler_bit::show_binlog_type(const Conv_source &src, const Field &,
String *str) const
{
CHARSET_INFO *cs= str->charset();
int bit_length= 8 * (src.metadata() >> 8) + (src.metadata() & 0xFF);
size_t length=
cs->cset->snprintf(cs, (char*) str->ptr(), str->alloced_length(),
"bit(%d)", bit_length);
str->length(length);
}
void Type_handler_olddecimal::show_binlog_type(const Conv_source &src,
const Field &,
String *str) const
{
CHARSET_INFO *cs= str->charset();
size_t length=
cs->cset->snprintf(cs, (char*) str->ptr(), str->alloced_length(),
"decimal(%d,?)/*old*/", src.metadata());
str->length(length);
}
void Type_handler_newdecimal::show_binlog_type(const Conv_source &src,
const Field &,
String *str) const
{
CHARSET_INFO *cs= str->charset();
size_t length=
cs->cset->snprintf(cs, (char*) str->ptr(), str->alloced_length(),
"decimal(%d,%d)",
src.metadata() >> 8, src.metadata() & 0xff);
str->length(length);
}
void Type_handler_blob_compressed::show_binlog_type(const Conv_source &src,
const Field &,
String *str) const
{
/*
Field::real_type() lies regarding the actual type of a BLOB, so
it is necessary to check the pack length to figure out what kind
of blob it really is.
*/
switch (src.metadata()) {
case 1:
str->set_ascii(STRING_WITH_LEN("tinyblob compressed"));
break;
case 2:
str->set_ascii(STRING_WITH_LEN("blob compressed"));
break;
case 3:
str->set_ascii(STRING_WITH_LEN("mediumblob compressed"));
break;
default:
DBUG_ASSERT(0);
// Fall through
case 4:
str->set_ascii(STRING_WITH_LEN("longblob compressed"));
}
}
void Type_handler_string::show_binlog_type(const Conv_source &src,
const Field &dst,
String *str) const
{
/*
This is taken from Field_string::unpack.
*/
CHARSET_INFO *cs= str->charset();
uint bytes= (((src.metadata() >> 4) & 0x300) ^ 0x300) +
(src.metadata() & 0x00ff);
const char* fmt= dst.cmp_type() != STRING_RESULT || dst.has_charset()
? "char(%u octets)" : "binary(%u)";
size_t length= cs->cset->snprintf(cs, (char*) str->ptr(),
str->alloced_length(),
fmt, bytes);
str->length(length);
}
enum_conv_type
Field::rpl_conv_type_from_same_data_type(uint16 metadata,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (metadata == 0) // Metadata can only be zero if no metadata was provided
{
/*
If there is no metadata, we either have an old event where no
metadata were supplied, or a type that does not require any
metadata. In either case, conversion can be done but no
conversion table is necessary.
*/
DBUG_PRINT("debug", ("Base types are identical, but there is no metadata"));
return CONV_TYPE_PRECISE;
}
DBUG_PRINT("debug", ("Base types are identical, doing field size comparison"));
int order= 0;
if (!compatible_field_size(metadata, rli, param.table_def_flags(), &order))
return CONV_TYPE_IMPOSSIBLE;
return order == 0 ? CONV_TYPE_PRECISE :
order < 0 ? CONV_TYPE_SUBSET_TO_SUPERSET :
CONV_TYPE_SUPERSET_TO_SUBSET;
}
enum_conv_type
Field_new_decimal::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (binlog_type() == source.real_field_type())
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
if (source.type_handler() == &type_handler_olddecimal ||
source.type_handler() == &type_handler_newdecimal ||
source.type_handler() == &type_handler_float ||
source.type_handler() == &type_handler_double)
{
/*
Then the other type is either FLOAT, DOUBLE, or old style
DECIMAL, so we require lossy conversion.
*/
return CONV_TYPE_SUPERSET_TO_SUBSET;
}
return CONV_TYPE_IMPOSSIBLE;
}
/*
This covers FLOAT, DOUBLE and old DECIMAL
*/
enum_conv_type
Field_real::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (binlog_type() == source.real_field_type())
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
if (source.type_handler() == &type_handler_olddecimal ||
source.type_handler() == &type_handler_newdecimal)
return CONV_TYPE_SUPERSET_TO_SUBSET; // Always require lossy conversions
if (source.type_handler() == &type_handler_float ||
source.type_handler() == &type_handler_double)
{
enum_conv_type order= compare_lengths(source.type_handler(),
max_display_length_for_field(source),
type_handler(), max_display_length());
DBUG_ASSERT(order != CONV_TYPE_PRECISE);
return order;
}
return CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_int::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (binlog_type() == source.real_field_type())
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
/*
The length comparison check will do the correct job of comparing
the field lengths (in bytes) of two integer types.
*/
if (source.type_handler() == &type_handler_stiny ||
source.type_handler() == &type_handler_sshort ||
source.type_handler() == &type_handler_sint24 ||
source.type_handler() == &type_handler_slong ||
source.type_handler() == &type_handler_slonglong)
{
/*
max_display_length_for_field() is not fully precise for the integer
data types. So its result cannot be compared to the result of
max_dispay_length() when the table field and the binlog field
are of the same type.
This code should eventually be rewritten not to use
compare_lengths(), to detect subtype/supetype relations
just using the type codes.
*/
DBUG_ASSERT(source.real_field_type() != real_type());
enum_conv_type order= compare_lengths(source.type_handler(),
max_display_length_for_field(source),
type_handler(), max_display_length());
DBUG_ASSERT(order != CONV_TYPE_PRECISE);
return order;
}
return CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_enum::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
/*
For some reasons Field_enum and Field_set store MYSQL_TYPE_STRING
as a type code in the binary log and encode the real type in metadata.
So we need to test real_type() here instread of binlog_type().
*/
return real_type() == source.real_field_type() ?
rpl_conv_type_from_same_data_type(source.metadata(), rli, param) :
CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_longstr::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
/**
@todo
Implement Field_varstring_compressed::real_type() and
Field_blob_compressed::real_type() properly. All occurencies
of Field::real_type() have to be inspected and adjusted if needed.
Until it is not ready we have to compare source_type against
binlog_type() when replicating from or to compressed data types.
@sa Comment for Field::binlog_type()
*/
bool same_type;
if (source.real_field_type() == MYSQL_TYPE_VARCHAR_COMPRESSED ||
source.real_field_type() == MYSQL_TYPE_BLOB_COMPRESSED ||
binlog_type() == MYSQL_TYPE_VARCHAR_COMPRESSED ||
binlog_type() == MYSQL_TYPE_BLOB_COMPRESSED)
same_type= binlog_type() == source.real_field_type();
else if (Type_handler_json_common::is_json_type_handler(type_handler()))
same_type= type_handler()->type_handler_base() == source.type_handler();
else
same_type= type_handler() == source.type_handler();
if (same_type)
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
if (source.type_handler() == &type_handler_tiny_blob ||
source.type_handler() == &type_handler_medium_blob ||
source.type_handler() == &type_handler_long_blob ||
source.type_handler() == &type_handler_blob ||
source.type_handler() == &type_handler_blob_compressed ||
source.type_handler() == &type_handler_string ||
source.type_handler() == &type_handler_var_string ||
source.type_handler() == &type_handler_varchar ||
source.type_handler() == &type_handler_varchar_compressed)
{
enum_conv_type order= compare_lengths(source.type_handler(),
max_display_length_for_field(source),
type_handler(), max_display_length());
/*
Here we know that the types are different, so if the order
gives that they do not require any conversion, we still need
to have non-lossy conversion enabled to allow conversion
between different (string) types of the same length.
Also, if all conversions are disabled, it is not allowed to convert
between these types. Since the TEXT vs. BINARY is distinguished by
the charset, and the charset is not replicated, we cannot
currently distinguish between , e.g., TEXT and BLOB.
*/
if (order == CONV_TYPE_PRECISE)
order= CONV_TYPE_SUBSET_TO_SUPERSET;
return order;
}
return CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_newdate::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (real_type() == source.real_field_type())
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
if (source.type_handler() == &type_handler_datetime2)
return CONV_TYPE_SUPERSET_TO_SUBSET;
return CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_time::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (binlog_type() == source.real_field_type())
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
// 'MySQL56 TIME(N)' -> 'MariaDB-5.3 TIME(N)' is non-lossy
if (decimals() == source.metadata() &&
source.type_handler() == &type_handler_time2)
return CONV_TYPE_VARIANT; // TODO: conversion from FSP1>FSP2
return CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_timef::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (binlog_type() == source.real_field_type())
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
/*
See comment in Field_datetimef::rpl_conv_type_from()
'MariaDB-5.3 TIME(0)' to 'MySQL56 TIME(0)' is non-lossy
*/
if (source.metadata() == 0 && source.type_handler() == &type_handler_time)
return CONV_TYPE_VARIANT;
return CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_timestamp::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (binlog_type() == source.real_field_type())
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
// 'MySQL56 TIMESTAMP(N)' -> MariaDB-5.3 TIMESTAMP(N)' is non-lossy
if (source.metadata() == decimals() &&
source.type_handler() == &type_handler_timestamp2)
return CONV_TYPE_VARIANT; // TODO: conversion from FSP1>FSP2
return CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_timestampf::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (binlog_type() == source.real_field_type())
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
/*
See comment in Field_datetimef::rpl_conv_type_from()
'MariaDB-5.3 TIMESTAMP(0)' to 'MySQL56 TIMESTAMP(0)' is non-lossy
*/
if (source.metadata() == 0 &&
source.type_handler() == &type_handler_timestamp)
return CONV_TYPE_VARIANT;
return CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_datetime::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (binlog_type() == source.real_field_type())
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
// 'MySQL56 DATETIME(N)' -> MariaDB-5.3 DATETIME(N) is non-lossy
if (source.metadata() == decimals() &&
source.type_handler() == &type_handler_datetime2)
return CONV_TYPE_VARIANT; // TODO: conversion from FSP1>FSP2
if (source.type_handler() == &type_handler_newdate)
return CONV_TYPE_SUBSET_TO_SUPERSET;
return CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_datetimef::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
if (binlog_type() == source.real_field_type())
return rpl_conv_type_from_same_data_type(source.metadata(), rli, param);
/*
'MariaDB-5.3 DATETIME(N)' does not provide information about fractional
precision in metadata. So we assume the precision on the master is equal
to the precision on the slave.
TODO: See MDEV-17394 what happend in case precisions are in case different
'MariaDB-5.3 DATETIME(0)' to 'MySQL56 DATETIME(0)' is non-lossy
*/
if (source.metadata() == 0 &&
source.type_handler() == &type_handler_datetime)
return CONV_TYPE_VARIANT;
if (source.type_handler() == &type_handler_newdate)
return CONV_TYPE_SUBSET_TO_SUPERSET;
return CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_date::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
// old DATE
return binlog_type() == source.real_field_type() ?
rpl_conv_type_from_same_data_type(source.metadata(), rli, param) :
CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_bit::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
return binlog_type() == source.real_field_type() ?
rpl_conv_type_from_same_data_type(source.metadata(), rli, param) :
CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_year::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
return binlog_type() == source.real_field_type() ?
rpl_conv_type_from_same_data_type(source.metadata(), rli, param) :
CONV_TYPE_IMPOSSIBLE;
}
enum_conv_type
Field_null::rpl_conv_type_from(const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param) const
{
DBUG_ASSERT(0);
return CONV_TYPE_IMPOSSIBLE;
}
/**********************************************************************/
#if defined(HAVE_REPLICATION)
/**
*/
static void show_sql_type(const Conv_source &src, const Field &dst,
String *str)
{
DBUG_ENTER("show_sql_type");
DBUG_ASSERT(src.type_handler() != NULL);
DBUG_PRINT("enter", ("type: %s, metadata: 0x%x",
src.type_handler()->name().ptr(), src.metadata()));
src.type_handler()->show_binlog_type(src, dst, str);
DBUG_VOID_RETURN;
}
/**
Check the order variable and print errors if the order is not
acceptable according to the current settings.
@param order The computed order of the conversion needed.
@param rli The relay log info data structure: for error reporting.
*/
static bool is_conversion_ok(enum_conv_type type, const Relay_log_info *rli,
ulonglong type_conversion_options)
{
DBUG_ENTER("is_conversion_ok");
bool allow_non_lossy, allow_lossy;
allow_non_lossy= type_conversion_options &
(1ULL << SLAVE_TYPE_CONVERSIONS_ALL_NON_LOSSY);
allow_lossy= type_conversion_options &
(1ULL << SLAVE_TYPE_CONVERSIONS_ALL_LOSSY);
DBUG_PRINT("enter", ("order: %d, flags:%s%s", (int) type,
allow_non_lossy ? " ALL_NON_LOSSY" : "",
allow_lossy ? " ALL_LOSSY" : ""));
switch (type) {
case CONV_TYPE_PRECISE:
case CONV_TYPE_VARIANT:
DBUG_RETURN(true);
case CONV_TYPE_SUBSET_TO_SUPERSET:
/* !!! Add error message saying that non-lossy conversions need to be allowed. */
DBUG_RETURN(allow_non_lossy);
case CONV_TYPE_SUPERSET_TO_SUBSET:
/* !!! Add error message saying that lossy conversions need to be allowed. */
DBUG_RETURN(allow_lossy);
case CONV_TYPE_IMPOSSIBLE:
DBUG_RETURN(false);
}
DBUG_RETURN(false);
}
/**
Can a type potentially be converted to another type?
This function check if the types are convertible and what
conversion is required.
If conversion is not possible, and error is printed.
If conversion is possible:
- *order will be set to -1 if source type is smaller than target
type and a non-lossy conversion can be required. This includes
the case where the field types are different but types could
actually be converted in either direction.
- *order will be set to 0 if no conversion is required.
- *order will be set to 1 if the source type is strictly larger
than the target type and that conversion is potentially lossy.
@param[in] field Target field
@param[in] type Source field type
@param[in] metadata Source field metadata
@param[in] rli Relay log info (for error reporting)
@param[in] mflags Flags from the table map event
@param[out] order Order between source field and target field
@return @c true if conversion is possible according to the current
settings, @c false if conversion is not possible according to the
current setting.
*/
static enum_conv_type
can_convert_field_to(Field *field, const Conv_source &source,
const Relay_log_info *rli,
const Conv_param &param)
{
DBUG_ENTER("can_convert_field_to");
#ifndef DBUG_OFF
char field_type_buf[MAX_FIELD_WIDTH];
String field_type(field_type_buf, sizeof(field_type_buf), &my_charset_latin1);
field->sql_type(field_type);
DBUG_PRINT("enter", ("field_type: %s, target_type: %d, source_type: %d, source_metadata: 0x%x",
field_type.c_ptr_safe(), field->real_type(),
source.real_field_type(), source.metadata()));
#endif
DBUG_RETURN(field->rpl_conv_type_from(source, rli, param));
}
const Type_handler *table_def::field_type_handler(uint col) const
{
enum_field_types typecode= binlog_type(col);
uint16 metadata= field_metadata(col);
DBUG_ASSERT(typecode != MYSQL_TYPE_ENUM);
DBUG_ASSERT(typecode != MYSQL_TYPE_SET);
if (typecode == MYSQL_TYPE_BLOB)
{
switch (metadata & 0xff) {
case 1: return &type_handler_tiny_blob;
case 2: return &type_handler_blob;
case 3: return &type_handler_medium_blob;
case 4: return &type_handler_long_blob;
default: return NULL;
}
}
if (typecode == MYSQL_TYPE_STRING)
{
uchar typecode2= metadata >> 8;
if (typecode2 == MYSQL_TYPE_SET)
return &type_handler_set;
if (typecode2 == MYSQL_TYPE_ENUM)
return &type_handler_enum;
return &type_handler_string;
}
/*
This type has not been used since before row-based replication,
so we can safely assume that it really is MYSQL_TYPE_NEWDATE.
*/
if (typecode == MYSQL_TYPE_DATE)
return &type_handler_newdate;
return Type_handler::get_handler_by_real_type(typecode);
}
/**
Is the definition compatible with a table?
Compare the master table with an existing table on the slave and
create a conversion map for fields that needs to be converted and
update master_to_slave_error[] map with fields that does not exist
on the slave or are not compatible with the field with the same name
on the slave.
If any fields need to be converted, a temporary conversion table
is created with the fields that needs conversions
Compatibility checking will be done for each row event by calling
check_wrong_column_usage()
@param rli_arg[in] Relay log info, for error reporting.
@param table[in] Table to compare with
The conversion table will be stored in table_list->conv_table
table_list->m_tabledef.master_to_slave_error[X] will hold the error to be
reported if the row_event will contain column master column X.
@return 0 ok
@return 1 Something went wrong (OOM?)
*/
bool
table_def::compatible_with(THD *thd, rpl_group_info *rgi,
RPL_TABLE_LIST *table_list)
const
{
/*
We only check the initial columns for the tables.
*/
Relay_log_info *rli= rgi->rli;
TABLE *table= table_list->table, *tmp_table= NULL;
uint master_cols= size(), conv_table_idx= 0;
for (uint col= 0 ; col < master_cols ; ++col)
{
uint slave_idx;
/* Skip columns on the master that where not replicated */
if (master_to_slave_error[col])
continue; // Field is not usable on the slave
slave_idx= master_to_slave_map[col];
Field *const field= table->field[slave_idx];
const Type_handler *field_handler= field_type_handler(col);
if (!field_handler)
{
master_to_slave_error[col]= SLAVE_FIELD_UNKNOWN_TYPE;
continue;
}
Conv_source source(field_handler, field_metadata(col), field->charset());
enum_conv_type convtype= can_convert_field_to(field, source, rli,
Conv_param(m_flags));
if (is_conversion_ok(convtype, rli, slave_type_conversions_options))
{
DBUG_PRINT("debug", ("Checking column %d -"
" field '%s' can be converted - order: %d",
col, field->field_name.str, convtype));
/*
If conversion type is not CONV_TYPE_RECISE, a conversion is required,
so we need to set up the conversion table.
*/
if (convtype != CONV_TYPE_PRECISE && tmp_table == NULL)
{
/*
This will create the full table with all fields. This is
necessary to ge the correct field lengths for the record.
*/
tmp_table= create_conversion_table(thd, rgi, table_list);
if (tmp_table == NULL)
return true;
/*
Clear all fields up to, but not including, this column, as
they do not need conversions.
The conversion table has one field for every used field on
the master that also exists on the slave in the master order.
*/
for (uint i= 0; i < conv_table_idx; ++i)
tmp_table->field[i]= NULL;
}
if (convtype == CONV_TYPE_PRECISE && tmp_table != NULL)
tmp_table->field[conv_table_idx]= NULL;
conv_table_idx++;
}
else
{
DBUG_PRINT("debug", ("Checking column %d -"
" field '%s' can not be converted",
col, field->field_name.str));
DBUG_ASSERT(col < size() && col < table->s->fields);
DBUG_ASSERT(table->s->db.str && table->s->table_name.str);
DBUG_ASSERT(table->in_use);
master_to_slave_error[col]= SLAVE_FIELD_WRONG_TYPE;
}
}
#ifndef DBUG_OFF
if (tmp_table)
{
for (unsigned int col= 0; col < tmp_table->s->fields; ++col)
if (tmp_table->field[col])
{
char source_buf[MAX_FIELD_WIDTH];
char target_buf[MAX_FIELD_WIDTH];
String source_type(source_buf, sizeof(source_buf), &my_charset_latin1);
String target_type(target_buf, sizeof(target_buf), &my_charset_latin1);
tmp_table->field[col]->sql_type(source_type);
table->field[col]->sql_type(target_type);
DBUG_PRINT("debug", ("Field %s - conversion required."
" Source type: '%s', Target type: '%s'",
tmp_table->field[col]->field_name.str,
source_type.c_ptr_safe(), target_type.c_ptr_safe()));
}
}
#endif
table_list->m_conv_table= tmp_table;
return false;
}
/*
Check if there are any not supported columns are used
*/
bool RPL_TABLE_LIST::check_wrong_column_usage(rpl_group_info *rgi,
MY_BITMAP *m_cols)
{
DBUG_ENTER("RPL_TABLE_LIST::check_wrong_column_usage");
for (uint col= 0 ; col < m_tabledef.size() ; col++)
{
if (!bitmap_is_set(m_cols, col))
continue;
if (m_tabledef.master_to_slave_error[col])
{
if (give_compatibility_error(rgi, col))
DBUG_RETURN(1);
}
}
DBUG_RETURN(0);
}
/*
Give an error if we are trying to access a wrong column
@return 0 error was ignored
@return 1 error, abort replication
*/
bool RPL_TABLE_LIST::give_compatibility_error(rpl_group_info *rgi, uint col)
{
enum loglevel error_level= ERROR_LEVEL;
switch (m_tabledef.master_to_slave_error[col]) {
case SLAVE_FIELD_NAME_MISSING:
if (!(slave_type_conversions_options &
SLAVE_TYPE_CONVERSIONS_ERROR_IF_MISSING_FIELD))
error_level= WARNING_LEVEL;
if (error_level == ERROR_LEVEL || table->in_use->variables.log_warnings >= 1)
rgi->rli->report(error_level, ER_SLAVE_CORRUPT_EVENT, rgi->gtid_info(),
"Column '%s' missing from table '%s.%s",
m_tabledef.master_column_name[col],
table->s->db.str, table->s->table_name.str);
break;
case SLAVE_FIELD_NR_MISSING:
{
char number[LONGLONG_BUFFER_SIZE];
if (!(slave_type_conversions_options &
SLAVE_TYPE_CONVERSIONS_ERROR_IF_MISSING_FIELD))
error_level= WARNING_LEVEL;
if (error_level == ERROR_LEVEL || table->in_use->variables.log_warnings >= 1)
rgi->rli->report(error_level, ER_SLAVE_CORRUPT_EVENT, rgi->gtid_info(),
"Column %s missing from table '%s.%s",
llstr(col+1, number),
table->s->db.str, table->s->table_name.str);
return 0;
break;
}
case SLAVE_FIELD_UNKNOWN_TYPE:
{
Field *field= table->field[m_tabledef.master_to_slave_map[col]];
rgi->rli->report(ERROR_LEVEL, ER_SLAVE_CORRUPT_EVENT, rgi->gtid_info(),
"In RBR mode, Slave received unknown field type field %d "
"for column Name: %s.%s.%s",
m_tabledef.binlog_type(col),
field->table->s->db.str,
field->table->s->table_name.str,
field->field_name.str);
break;
}
case SLAVE_FIELD_WRONG_TYPE:
{
Field *field= table->field[m_tabledef.master_to_slave_map[col]];
const char *db_name= table->s->db.str;
const char *tbl_name= table->s->table_name.str;
StringBuffer<MAX_FIELD_WIDTH> source_type(&my_charset_latin1);
StringBuffer<MAX_FIELD_WIDTH> target_type(&my_charset_latin1);
THD *thd= table->in_use;
const Type_handler *h= m_tabledef.field_type_handler(col);
Conv_source source(h, m_tabledef.field_metadata(col), field->charset());
show_sql_type(source, *field, &source_type);
field->sql_rpl_type(&target_type);
DBUG_ASSERT(source_type.length() > 0);
DBUG_ASSERT(target_type.length() > 0);
rgi->rli->report(ERROR_LEVEL, ER_SLAVE_CONVERSION_FAILED, rgi->gtid_info(),
ER_THD(thd, ER_SLAVE_CONVERSION_FAILED),
col, db_name, tbl_name,
source_type.c_ptr_safe(), target_type.c_ptr_safe());
break;
}
}
return 1;
}
/**
A wrapper to Virtual_tmp_table, to get access to its constructor,
which is protected for safety purposes (against illegal use on stack).
*/
class Virtual_conversion_table: public Virtual_tmp_table
{
public:
Virtual_conversion_table(THD *thd) :Virtual_tmp_table(thd) { }
/**
Add a new field into the virtual table.
@param handler - The type handler of the field.
@param metadata - The RBR binary log metadata for this field.
@param target_field - The field from the target table, to get extra
attributes from (e.g. typelib in case of ENUM).
*/
bool add(const Type_handler *handler,
uint16 metadata, const Field *target_field)
{
Field *tmp= handler->make_conversion_table_field(in_use->mem_root,
this, metadata,
target_field);
if (!tmp)
return true;
Virtual_tmp_table::add(tmp);
DBUG_PRINT("debug", ("sql_type: %s, target_field: '%s', max_length: %d, decimals: %d,"
" maybe_null: %d, unsigned_flag: %d, pack_length: %u",
handler->name().ptr(), target_field->field_name.str,
tmp->field_length, tmp->decimals(), TRUE,
tmp->flags, tmp->pack_length()));
return false;
}
/* Make last inserted field not null */
void make_not_null()
{
DBUG_ASSERT(s->fields > 0);
/* Resetting flag and null_ptr makes the field not null */
field[s->fields-1]->flags |= NOT_NULL_FLAG;
field[s->fields-1]->null_ptr= 0;
}
};
/**
Create a conversion table
If the function is unable to create the conversion table, an error
will be printed and NULL will be returned.
@return Pointer to conversion table, or NULL if unable to create
conversion table.
The conversion table contains one field for every field in the binlog
for which there exists a field on the slave.
*/
TABLE *table_def::create_conversion_table(THD *thd, rpl_group_info *rgi,
RPL_TABLE_LIST *table_list) const
{
Virtual_conversion_table *conv_table;
Relay_log_info *rli= rgi->rli;
TABLE *target_table= table_list->table;
uint const cols_to_create= MY_MIN(size(), target_table->s->fields);
DBUG_ENTER("table_def::create_conversion_table");
if (!(conv_table= new(thd) Virtual_conversion_table(thd)) ||
conv_table->init(cols_to_create))
goto err;
/*
Iterate through the number of columns logged on the master, and if
skip any that is missing on the slave. Skipped columns are not
added to the conv_table, as there is no column on the slave to use
as the reference for the target_field.
*/
for (uint col= 0 ; col < cols_to_create; col++)
{
Field *field;
if (master_to_slave_error[col])
continue; // Slave does not have field
const Type_handler *ha= field_type_handler(col);
if (!ha)
{
/* This can happen as we have not checked all columns in the caller */
master_to_slave_error[col]= SLAVE_FIELD_UNKNOWN_TYPE;
continue;
}
field= target_table->field[master_to_slave_map[col]];
if (conv_table->add(ha, field_metadata(col), field))
{
DBUG_PRINT("debug", ("binlog_type: %d, metadata: %04X, target_field: '%s'"
" make_conversion_table_field() failed",
binlog_type(col), field_metadata(col),
field->field_name.str));
goto err;
}
/*
We only use the conversion table for not null values
This also avoids a bug in Virtual_conversion_table where the null
pointer for created fields points to uninitialized memory.
*/
conv_table->make_not_null();
}
conv_table->fix_field_count();
if (conv_table->open())
goto err; // Could not allocate record buffer?
DBUG_RETURN(conv_table);
err:
if (conv_table)
delete conv_table;
rli->report(ERROR_LEVEL, ER_SLAVE_CANT_CREATE_CONVERSION, rgi->gtid_info(),
ER_THD(thd, ER_SLAVE_CANT_CREATE_CONVERSION),
target_table->s->db.str,
target_table->s->table_name.str);
DBUG_RETURN(NULL);
}
Deferred_log_events::Deferred_log_events(Relay_log_info *rli) : last_added(NULL)
{
my_init_dynamic_array(PSI_INSTRUMENT_ME, &array, sizeof(Log_event *), 32, 16, MYF(0));
}
Deferred_log_events::~Deferred_log_events()
{
delete_dynamic(&array);
}
int Deferred_log_events::add(Log_event *ev)
{
last_added= ev;
insert_dynamic(&array, (uchar*) &ev);
return 0;
}
bool Deferred_log_events::is_empty()
{
return array.elements == 0;
}
bool Deferred_log_events::execute(rpl_group_info *rgi)
{
bool res= false;
DBUG_ENTER("Deferred_log_events::execute");
DBUG_ASSERT(rgi->deferred_events_collecting);
rgi->deferred_events_collecting= false;
for (uint i= 0; !res && i < array.elements; i++)
{
Log_event *ev= (* (Log_event **)
dynamic_array_ptr(&array, i));
res= ev->apply_event(rgi);
}
rgi->deferred_events_collecting= true;
DBUG_RETURN(res);
}
void Deferred_log_events::rewind()
{
/*
Reset preceding Query log event events which execution was
deferred because of slave side filtering.
*/
if (!is_empty())
{
for (uint i= 0; i < array.elements; i++)
{
Log_event *ev= *(Log_event **) dynamic_array_ptr(&array, i);
delete ev;
}
last_added= NULL;
if (array.elements > array.max_element)
freeze_size(&array);
reset_dynamic(&array);
}
last_added= NULL;
}
/*
Create column mapping from the master table to the slave table
Mapping stored in master_to_slave_map[].
Errors stored in master_to_slave_error[]. Error will be given
on usage.
Store master column names in master_column_name[].
Note that we map all columns as we at this point do not know which
columns will be used by the row events.
*/
bool RPL_TABLE_LIST::create_column_mapping(rpl_group_info *rgi)
{
ulong master_cols= m_tabledef.size();
DBUG_ENTER("RPL_TABLE_LIST::create_column_mapping");
DBUG_ASSERT(table->s);
if (!m_tabledef.optional_metadata.length)
{
default_column_mapping:
uint col, min_cols= MY_MIN(master_cols, table->s->fields);
for (col= 0; col < min_cols; col++)
m_tabledef.master_to_slave_map[col]= col;
for ( ; col < master_cols ; col++)
{
m_tabledef.master_to_slave_map[col]= INT_MAX32;
m_tabledef.master_to_slave_error[col]= SLAVE_FIELD_NR_MISSING;
}
DBUG_RETURN(0);
}
Table_map_log_event::Optional_metadata_fields
opt_metadata((uchar*) m_tabledef.optional_metadata.str,
m_tabledef.optional_metadata.length);
if (!opt_metadata.m_column_name.size())
{
/*
If there are no column names provided in the optional metadata
use the default column mapping.
This can happen when reading an event from MySQL 8.
*/
goto default_column_mapping;
}
for (uint col= 0; col < master_cols; col++)
{
std::string master_col_name_cppstr= opt_metadata.m_column_name[col];
LEX_CSTRING field_name=
{ master_col_name_cppstr.c_str(), master_col_name_cppstr.length() };
Field *field= table->find_field_by_name(&field_name);
if (unlikely(!field))
{
m_tabledef.master_column_name[col]= field_name.str;
m_tabledef.master_to_slave_map[col]= INT_MAX32; // Not used
m_tabledef.master_to_slave_error[col]= SLAVE_FIELD_NAME_MISSING;
continue; // ok that field did not exists
}
m_tabledef.master_to_slave_map[col]= field->field_index;
DBUG_PRINT("info", ("Found mapping for %s", field_name.str));
}
DBUG_RETURN(false);
}
#endif // defined(HAVE_REPLICATION)
Reference in a new issue View git blame Copy permalink