diff options
Diffstat (limited to 'sql')
-rw-r--r-- | sql/field.cc | 236 | ||||
-rw-r--r-- | sql/field.h | 33 | ||||
-rw-r--r-- | sql/log_event.cc | 261 | ||||
-rw-r--r-- | sql/log_event.h | 10 | ||||
-rw-r--r-- | sql/rpl_record.cc | 43 | ||||
-rw-r--r-- | sql/rpl_rli.h | 10 | ||||
-rw-r--r-- | sql/rpl_utility.cc | 99 | ||||
-rw-r--r-- | sql/rpl_utility.h | 132 |
8 files changed, 737 insertions, 87 deletions
diff --git a/sql/field.cc b/sql/field.cc index a970a6e4318..49433deca74 100644 --- a/sql/field.cc +++ b/sql/field.cc @@ -1372,6 +1372,51 @@ int Field::store(const char *to, uint length, CHARSET_INFO *cs, } +/** + Unpack a field from row data. + + This method is used to unpack a field from a master whose size + of the field is less than that of the slave. + + @param to Destination of the data + @param from Source of the data + @param param_data Pack length of the field data + + @return New pointer into memory based on from + length of the data +*/ +const uchar *Field::unpack(uchar* to, + const uchar *from, + uint param_data) +{ + uint length=pack_length(); + int from_type= 0; + /* + If from length is > 255, it has encoded data in the upper bits. Need + to mask it out. + */ + if (param_data > 255) + { + from_type= (param_data & 0xff00) >> 8U; // real_type. + param_data= param_data & 0x00ff; // length. + } + uint len= (param_data && (param_data < length)) ? + param_data : length; + /* + If the length is the same, use old unpack method. + If the param_data is 0, use the old unpack method. + This is possible if the table map was generated from a down-level + master or if the data was not available on the master. + If the real_types are not the same, use the old unpack method. + */ + if ((length == param_data) || + (param_data == 0) || + (from_type != real_type())) + return(unpack(to, from)); + memcpy(to, from, param_data > length ? length : len); + return from+len; +} + + my_decimal *Field::val_decimal(my_decimal *decimal) { /* This never have to be called */ @@ -2657,6 +2702,52 @@ uint Field_new_decimal::is_equal(Create_field *new_field) } +/** + Unpack a decimal field from row data. + + This method is used to unpack a decimal or numeric field from a master + whose size of the field is less than that of the slave. + + @param to Destination of the data + @param from Source of the data + @param param_data Precision (upper) and decimal (lower) values + + @return New pointer into memory based on from + length of the data +*/ +const uchar *Field_new_decimal::unpack(uchar* to, + const uchar *from, + uint param_data) +{ + uint from_precision= (param_data & 0xff00) >> 8U; + uint from_decimal= param_data & 0x00ff; + uint length=pack_length(); + uint from_pack_len= my_decimal_get_binary_size(from_precision, from_decimal); + uint len= (param_data && (from_pack_len < length)) ? + from_pack_len : length; + if (from_pack_len && (from_pack_len < length)) + { + /* + If the master's data is smaller than the slave, we need to convert + the binary to decimal then resize the decimal converting it back to + a decimal and write that to the raw data buffer. + */ + decimal_digit_t dec_buf[DECIMAL_MAX_PRECISION]; + decimal_t dec; + dec.len= from_precision; + dec.buf= dec_buf; + /* + Note: bin2decimal does not change the length of the field. So it is + just the first step the resizing operation. The second step does the + resizing using the precision and decimals from the slave. + */ + bin2decimal((uchar *)from, &dec, from_precision, from_decimal); + decimal2bin(&dec, to, precision, decimals()); + } + else + memcpy(to, from, len); // Sizes are the same, just copy the data. + return from+len; +} + /**************************************************************************** ** tiny int ****************************************************************************/ @@ -6306,6 +6397,38 @@ uchar *Field_string::pack(uchar *to, const uchar *from, uint max_length) } +/** + Unpack a string field from row data. + + This method is used to unpack a string field from a master whose size + of the field is less than that of the slave. Note that there can be a + variety of field types represented with this class. Certain types like + ENUM or SET are processed differently. Hence, the upper byte of the + @c param_data argument contains the result of field->real_type() from + the master. + + @param to Destination of the data + @param from Source of the data + @param param_data Real type (upper) and length (lower) values + + @return New pointer into memory based on from + length of the data +*/ +const uchar *Field_string::unpack(uchar *to, + const uchar *from, + uint param_data) +{ + uint from_len= param_data & 0x00ff; // length. + uint length= 0; + uint f_length; + f_length= (from_len < field_length) ? from_len : field_length; + DBUG_ASSERT(f_length <= 255); + length= (uint) *from++; + bitmap_set_bit(table->write_set,field_index); + store((const char *)from, length, system_charset_info); + return from+length; +} + + const uchar *Field_string::unpack(uchar *to, const uchar *from) { uint length; @@ -6797,6 +6920,44 @@ uchar *Field_varstring::pack_key_from_key_image(uchar *to, const uchar *from, } +/** + Unpack a varstring field from row data. + + This method is used to unpack a varstring field from a master + whose size of the field is less than that of the slave. + + @param to Destination of the data + @param from Source of the data + @param param_data Length bytes from the master's field data + + @return New pointer into memory based on from + length of the data +*/ +const uchar *Field_varstring::unpack(uchar *to, + const uchar *from, + uint param_data) +{ + uint length; + uint l_bytes= (param_data && (param_data < field_length)) ? + (param_data <= 255) ? 1 : 2 : length_bytes; + if (l_bytes == 1) + { + to[0]= *from++; + length= to[0]; + if (length_bytes == 2) + to[1]= 0; + } + else + { + length= uint2korr(from); + to[0]= *from++; + to[1]= *from++; + } + if (length) + memcpy(to+ length_bytes, from, length); + return from+length; +} + + /* unpack field packed with Field_varstring::pack() */ @@ -7493,6 +7654,29 @@ uchar *Field_blob::pack(uchar *to, const uchar *from, uint max_length) } +/** + Unpack a blob field from row data. + + This method is used to unpack a blob field from a master whose size of + the field is less than that of the slave. Note: This method is included + to satisfy inheritance rules, but is not needed for blob fields. It + simply is used as a pass-through to the original unpack() method for + blob fields. + + @param to Destination of the data + @param from Source of the data + @param param_data <not used> + + @return New pointer into memory based on from + length of the data +*/ +const uchar *Field_blob::unpack(uchar *to, + const uchar *from, + uint param_data) +{ + return unpack(to, from); +} + + const uchar *Field_blob::unpack(uchar *to, const uchar *from) { memcpy(to,from,packlength); @@ -8540,6 +8724,58 @@ uchar *Field_bit::pack(uchar *to, const uchar *from, uint max_length) } +/** + Unpack a bit field from row data. + + This method is used to unpack a bit field from a master whose size + of the field is less than that of the slave. + + @param to Destination of the data + @param from Source of the data + @param param_data Bit length (upper) and length (lower) values + + @return New pointer into memory based on from + length of the data +*/ +const uchar *Field_bit::unpack(uchar *to, + const uchar *from, + uint param_data) +{ + uint const from_len= (param_data >> 8U) & 0x00ff; + uint const from_bit_len= param_data & 0x00ff; + /* + If the master and slave have the same sizes, then use the old + unpack() method. + */ + if ((from_bit_len == bit_len) && + (from_len == bytes_in_rec)) + return(unpack(to, from)); + /* + We are converting a smaller bit field to a larger one here. + To do that, we first need to construct a raw value for the original + bit value stored in the from buffer. Then that needs to be converted + to the larger field then sent to store() for writing to the field. + Lastly the odd bits need to be masked out if the bytes_in_rec > 0. + Otherwise stray bits can cause spurious values. + */ + uint new_len= (field_length + 7) / 8; + char *value= (char *)my_alloca(new_len); + bzero(value, new_len); + uint len= from_len + ((from_bit_len > 0) ? 1 : 0); + memcpy(value + (new_len - len), from, len); + /* + Mask out the unused bits in the partial byte. + TODO: Add code to the master to always mask these bits and remove + the following. + */ + if ((from_bit_len > 0) && (from_len > 0)) + value[new_len - len]= value[new_len - len] & ((1U << from_bit_len) - 1); + bitmap_set_bit(table->write_set,field_index); + store(value, new_len, system_charset_info); + my_afree(value); + return from + len; +} + + const uchar *Field_bit::unpack(uchar *to, const uchar *from) { if (bit_len > 0) diff --git a/sql/field.h b/sql/field.h index 8bf087c7ebd..c1130d53569 100644 --- a/sql/field.h +++ b/sql/field.h @@ -345,6 +345,7 @@ public: memcpy(to,from,length); return to+length; } + virtual const uchar *unpack(uchar* to, const uchar *from, uint param_data); virtual const uchar *unpack(uchar* to, const uchar *from) { uint length=pack_length(); @@ -628,6 +629,7 @@ public: uint size_of() const { return sizeof(*this); } uint32 pack_length() const { return (uint32) bin_size; } uint is_equal(Create_field *new_field); + virtual const uchar *unpack(uchar* to, const uchar *from, uint param_data); }; @@ -1164,6 +1166,7 @@ public: void sort_string(uchar *buff,uint length); void sql_type(String &str) const; uchar *pack(uchar *to, const uchar *from, uint max_length=~(uint) 0); + virtual const uchar *unpack(uchar* to, const uchar *from, uint param_data); const uchar *unpack(uchar* to, const uchar *from); int pack_cmp(const uchar *a,const uchar *b,uint key_length, my_bool insert_or_update); @@ -1240,6 +1243,7 @@ public: uchar *pack_key(uchar *to, const uchar *from, uint max_length); uchar *pack_key_from_key_image(uchar* to, const uchar *from, uint max_length); + virtual const uchar *unpack(uchar* to, const uchar *from, uint param_data); const uchar *unpack(uchar* to, const uchar *from); const uchar *unpack_key(uchar* to, const uchar *from, uint max_length); int pack_cmp(const uchar *a, const uchar *b, uint key_length, @@ -1295,6 +1299,9 @@ public: l_char_length <= 16777215 ? 3 : 4; } } + Field_blob(uint32 packlength_arg) + :Field_longstr((uchar*) 0, 0, (uchar*) "", 0, NONE, "temp", system_charset_info), + packlength(packlength_arg) {} enum_field_types type() const { return MYSQL_TYPE_BLOB;} enum ha_base_keytype key_type() const { return binary() ? HA_KEYTYPE_VARBINARY2 : HA_KEYTYPE_VARTEXT2; } @@ -1316,6 +1323,17 @@ public: void sort_string(uchar *buff,uint length); uint32 pack_length() const { return (uint32) (packlength+table->s->blob_ptr_size); } + + /** + Return the packed length without the pointer size added. + + This is used to determine the size of the actual data in the row + buffer. + + @retval The length of the raw data itself without the pointer. + */ + uint32 pack_length_no_ptr() const + { return (uint32) (packlength); } uint32 sort_length() const; inline uint32 max_data_length() const { @@ -1331,7 +1349,18 @@ public: { store_length(ptr, packlength, number); } - + + /** + Return the packed length plus the length of the data. + + This is used to determine the size of the data plus the + packed length portion in the row data. + + @retval The length in the row plus the size of the data. + */ + uint32 get_packed_size(const uchar *ptr) + {return packlength + get_length((const uchar *)ptr);} + inline uint32 get_length(uint row_offset=0) { return get_length(ptr+row_offset); } uint32 get_length(const uchar *ptr); @@ -1379,6 +1408,7 @@ public: uchar *pack_key(uchar *to, const uchar *from, uint max_length); uchar *pack_key_from_key_image(uchar* to, const uchar *from, uint max_length); + virtual const uchar *unpack(uchar *to, const uchar *from, uint param_data); const uchar *unpack(uchar *to, const uchar *from); const uchar *unpack_key(uchar* to, const uchar *from, uint max_length); int pack_cmp(const uchar *a, const uchar *b, uint key_length, @@ -1554,6 +1584,7 @@ public: uint32 pack_length_in_rec() const { return bytes_in_rec; } void sql_type(String &str) const; uchar *pack(uchar *to, const uchar *from, uint max_length=~(uint) 0); + virtual const uchar *unpack(uchar *to, const uchar *from, uint param_data); const uchar *unpack(uchar* to, const uchar *from); virtual void set_default(); diff --git a/sql/log_event.cc b/sql/log_event.cc index dc492d08c10..22b5a96753d 100644 --- a/sql/log_event.cc +++ b/sql/log_event.cc @@ -6028,11 +6028,8 @@ int Rows_log_event::do_apply_event(RELAY_LOG_INFO const *rli) #ifdef HAVE_QUERY_CACHE query_cache.invalidate_locked_for_write(rli->tables_to_lock); #endif - const_cast<RELAY_LOG_INFO*>(rli)->clear_tables_to_lock(); } - DBUG_ASSERT(rli->tables_to_lock == NULL && rli->tables_to_lock_count == 0); - TABLE* table= const_cast<RELAY_LOG_INFO*>(rli)->m_table_map.get_table(m_table_id); if (table) @@ -6128,6 +6125,13 @@ int Rows_log_event::do_apply_event(RELAY_LOG_INFO const *rli) } } + /* + We need to delay this clear until the table def is no longer needed. + The table def is needed in unpack_row(). + */ + if (rli->tables_to_lock && get_flags(STMT_END_F)) + const_cast<RELAY_LOG_INFO*>(rli)->clear_tables_to_lock(); + if (error) { /* error has occured during the transaction */ rli->report(ERROR_LEVEL, thd->net.last_errno, @@ -6378,6 +6382,163 @@ void Rows_log_event::print_helper(FILE *file, Table_map_log_event member functions and support functions **************************************************************************/ +/** + * Calculate field metadata size based on the real_type of the field. + * + * @returns int Size of field metadata. + */ +#if !defined(MYSQL_CLIENT) +const int Table_map_log_event::calc_field_metadata_size() +{ + DBUG_ENTER("Table_map_log_event::calc_field_metadata_size"); + int size= 0; + for (unsigned int i= 0 ; i < m_table->s->fields ; i++) + { + switch (m_table->s->field[i]->real_type()) { + case MYSQL_TYPE_TINY_BLOB: + case MYSQL_TYPE_BLOB: + case MYSQL_TYPE_MEDIUM_BLOB: + case MYSQL_TYPE_LONG_BLOB: + case MYSQL_TYPE_DOUBLE: + case MYSQL_TYPE_FLOAT: + { + size++; // Store one byte here. + break; + } + case MYSQL_TYPE_BIT: + case MYSQL_TYPE_NEWDECIMAL: + case MYSQL_TYPE_ENUM: + case MYSQL_TYPE_STRING: + case MYSQL_TYPE_VARCHAR: + case MYSQL_TYPE_SET: + { + size= size + sizeof(short int); // Store short int here. + break; + } + default: + break; + } + } + m_field_metadata_size= size; + DBUG_PRINT("info", ("Table_map_log_event: %d bytes in field metadata.", + (int)m_field_metadata_size)); + DBUG_RETURN(m_field_metadata_size); +} +#endif /* !defined(MYSQL_CLIENT) */ + +/** + @page How replication of field metadata works. + + When a table map is created, the master first calls + Table_map_log_event::get_field_metadata_size() which calculates how many + values will be in the field metadata. Only those fields that require the + extra data are added (see table above). The master then loops through all + of the fields in the table calling the method + Table_map_log_event::get_field_metadata() which returns the values for the + field that will be saved in the metadata and replicated to the slave. Once + all fields have been processed, the table map is written to the binlog + adding the size of the field metadata and the field metadata to the end of + the body of the table map. + + When a table map is read on the slave, the field metadata is read from the + table map and passed to the table_def class constructor which saves the + field metadata from the table map into an array based on the type of the + field. Field metadata values not present (those fields that do not use extra + data) in the table map are initialized as zero (0). The array size is the + same as the columns for the table on the slave. + +*/ + +/** + Save the field metadata based on the real_type of the field. + The metadata saved depends on the type of the field. Some fields + store a single byte for pack_length() while others store two bytes + for field_length (max length). + + @retval 0 Ok. + + TODO: We may want to consider changing the encoding of the information. + Currently, the code attempts to minimize the number of bytes written to + the tablemap. There are at least two other alternatives; 1) using + net_store_length() to store the data allowing it to choose the number of + bytes that are appropriate thereby making the code much easier to + maintain (only 1 place to change the encoding), or 2) use a fixed number + of bytes for each field. The problem with option 1 is that net_store_length() + will use one byte if the value < 251, but 3 bytes if it is > 250. Thus, + for fields like CHAR which can be no larger than 255 characters, the method + will use 3 bytes when the value is > 250. Further, every value that is + encoded using 2 parts (e.g., pack_length, field_length) will be numerically + > 250 therefore will use 3 bytes for eah value. The problem with option 2 + is less wasteful for space but does waste 1 byte for every field that does + not encode 2 parts. +*/ +#if !defined(MYSQL_CLIENT) +int Table_map_log_event::save_field_metadata() +{ + DBUG_ENTER("Table_map_log_event::save_field_metadata"); + int index= 0; + for (unsigned int i= 0 ; i < m_table->s->fields ; i++) + { + switch (m_table->s->field[i]->real_type()) { + case MYSQL_TYPE_NEWDECIMAL: + { + m_field_metadata[index++]= + (uchar)((Field_new_decimal *)m_table->s->field[i])->precision; + m_field_metadata[index++]= + (uchar)((Field_new_decimal *)m_table->s->field[i])->decimals(); + break; + } + case MYSQL_TYPE_TINY_BLOB: + case MYSQL_TYPE_BLOB: + case MYSQL_TYPE_MEDIUM_BLOB: + case MYSQL_TYPE_LONG_BLOB: + { + m_field_metadata[index++]= + (uchar)((Field_blob *)m_table->s->field[i])->pack_length_no_ptr(); + break; + } + case MYSQL_TYPE_DOUBLE: + case MYSQL_TYPE_FLOAT: + { + m_field_metadata[index++]= (uchar)m_table->s->field[i]->pack_length(); + break; + } + case MYSQL_TYPE_BIT: + { + m_field_metadata[index++]= + (uchar)((Field_bit *)m_table->s->field[i])->bit_len; + m_field_metadata[index++]= + (uchar)((Field_bit *)m_table->s->field[i])->bytes_in_rec; + break; + } + case MYSQL_TYPE_VARCHAR: + { + char *ptr= (char *)&m_field_metadata[index]; + int2store(ptr, m_table->s->field[i]->field_length); + index= index + sizeof(short int); + break; + } + case MYSQL_TYPE_STRING: + { + m_field_metadata[index++]= (uchar)m_table->s->field[i]->real_type(); + m_field_metadata[index++]= m_table->s->field[i]->field_length; + break; + } + case MYSQL_TYPE_ENUM: + case MYSQL_TYPE_SET: + { + m_field_metadata[index++]= (uchar)m_table->s->field[i]->real_type(); + m_field_metadata[index++]= m_table->s->field[i]->pack_length(); + break; + } + default: + break; + } + } + DBUG_RETURN(0); +} +#endif /* !defined(MYSQL_CLIENT) */ + /* Constructor used to build an event for writing to the binary log. Mats says tbl->s lives longer than this event so it's ok to copy pointers @@ -6392,9 +6553,8 @@ Table_map_log_event::Table_map_log_event(THD *thd, TABLE *tbl, ulong tid, m_dblen(m_dbnam ? tbl->s->db.length : 0), m_tblnam(tbl->s->table_name.str), m_tbllen(tbl->s->table_name.length), - m_colcnt(tbl->s->fields), m_coltype(0), - m_table_id(tid), - m_flags(flags) + m_colcnt(tbl->s->fields), m_field_metadata(0), + m_table_id(tid), m_null_bits(0), m_flags(flags) { DBUG_ASSERT(m_table_id != ~0UL); /* @@ -6413,6 +6573,16 @@ Table_map_log_event::Table_map_log_event(THD *thd, TABLE *tbl, ulong tid, m_data_size+= m_dblen + 2; // Include length and terminating \0 m_data_size+= m_tbllen + 2; // Include length and terminating \0 m_data_size+= 1 + m_colcnt; // COLCNT and column types + m_field_metadata_size= calc_field_metadata_size(); + + /* + Now set the size of the data to the size of the field metadata array + plus one or two bytes for number of elements in the field metadata array. + */ + if (m_field_metadata_size > 255) + m_data_size+= m_field_metadata_size + 2; + else + m_data_size+= m_field_metadata_size + 1; /* If malloc fails, catched in is_valid() */ if ((m_memory= (uchar*) my_malloc(m_colcnt, MYF(MY_WME)))) @@ -6421,6 +6591,28 @@ Table_map_log_event::Table_map_log_event(THD *thd, TABLE *tbl, ulong tid, for (unsigned int i= 0 ; i < m_table->s->fields ; ++i) m_coltype[i]= m_table->field[i]->type(); } + + /* + Calculate a bitmap for the results of maybe_null() for all columns. + The bitmap is used to determine when there is a column from the master + that is not on the slave and is null and thus not in the row data during + replication. + */ + uint num_null_bytes= (m_table->s->fields + 7) / 8; + m_data_size+= num_null_bytes; + m_meta_memory= (uchar *)my_multi_malloc(MYF(MY_WME), + &m_null_bits, num_null_bytes, + &m_field_metadata, m_field_metadata_size, + NULL); + bzero(m_null_bits, num_null_bytes); + for (unsigned int i= 0 ; i < m_table->s->fields ; ++i) + if (m_table->field[i]->maybe_null()) + m_null_bits[(i / 8)]+= 1 << (i % 8); + + /* + Create an array for the field metadata and store it. + */ + save_field_metadata(); } #endif /* !defined(MYSQL_CLIENT) */ @@ -6436,8 +6628,10 @@ Table_map_log_event::Table_map_log_event(const char *buf, uint event_len, #ifndef MYSQL_CLIENT m_table(NULL), #endif - m_memory(NULL) + m_memory(NULL), + m_field_metadata(0), m_field_metadata_size(0) { + unsigned int bytes_read= 0; DBUG_ENTER("Table_map_log_event::Table_map_log_event(const char*,uint,...)"); uint8 common_header_len= description_event->common_header_len; @@ -6508,6 +6702,22 @@ Table_map_log_event::Table_map_log_event(const char *buf, uint event_len, strncpy(const_cast<char*>(m_dbnam), (const char*)ptr_dblen + 1, m_dblen + 1); strncpy(const_cast<char*>(m_tblnam), (const char*)ptr_tbllen + 1, m_tbllen + 1); memcpy(m_coltype, ptr_after_colcnt, m_colcnt); + + ptr_after_colcnt= ptr_after_colcnt + m_colcnt; + bytes_read= ptr_after_colcnt - (uchar *)buf; + DBUG_PRINT("info", ("Bytes read: %d.\n", bytes_read)); + if (bytes_read < event_len) + { + m_field_metadata_size= net_field_length(&ptr_after_colcnt); + uint num_null_bytes= (m_colcnt + 7) / 8; + m_meta_memory= (uchar *)my_multi_malloc(MYF(MY_WME), + &m_null_bits, num_null_bytes, + &m_field_metadata, m_field_metadata_size, + NULL); + memcpy(m_field_metadata, ptr_after_colcnt, m_field_metadata_size); + ptr_after_colcnt= (uchar*)ptr_after_colcnt + m_field_metadata_size; + memcpy(m_null_bits, ptr_after_colcnt, num_null_bytes); + } } DBUG_VOID_RETURN; @@ -6516,6 +6726,7 @@ Table_map_log_event::Table_map_log_event(const char *buf, uint event_len, Table_map_log_event::~Table_map_log_event() { + my_free(m_meta_memory, MYF(MY_ALLOW_ZERO_PTR)); my_free(m_memory, MYF(MY_ALLOW_ZERO_PTR)); } @@ -6646,7 +6857,8 @@ int Table_map_log_event::do_apply_event(RELAY_LOG_INFO const *rli) inside st_relay_log_info::clear_tables_to_lock() by calling the table_def destructor explicitly. */ - new (&table_list->m_tabledef) table_def(m_coltype, m_colcnt); + new (&table_list->m_tabledef) table_def(m_coltype, m_colcnt, + m_field_metadata, m_field_metadata_size, m_null_bits); table_list->m_tabledef_valid= TRUE; /* @@ -6718,12 +6930,22 @@ bool Table_map_log_event::write_data_body(IO_CACHE *file) uchar *const cbuf_end= net_store_length(cbuf, (size_t) m_colcnt); DBUG_ASSERT(static_cast<size_t>(cbuf_end - cbuf) <= sizeof(cbuf)); + /* + Store the size of the field metadata. + */ + uchar mbuf[sizeof(m_field_metadata_size)]; + uchar *const mbuf_end= net_store_length(mbuf, + (size_t) m_field_metadata_size); + return (my_b_safe_write(file, dbuf, sizeof(dbuf)) || my_b_safe_write(file, (const uchar*)m_dbnam, m_dblen+1) || my_b_safe_write(file, tbuf, sizeof(tbuf)) || my_b_safe_write(file, (const uchar*)m_tblnam, m_tbllen+1) || my_b_safe_write(file, cbuf, (size_t) (cbuf_end - cbuf)) || - my_b_safe_write(file, m_coltype, m_colcnt)); + my_b_safe_write(file, m_coltype, m_colcnt) || + my_b_safe_write(file, mbuf, (size_t) (mbuf_end - mbuf)) || + my_b_safe_write(file, m_field_metadata, m_field_metadata_size), + my_b_safe_write(file, m_null_bits, (m_colcnt + 7) / 8)); } #endif @@ -6946,6 +7168,7 @@ copy_extra_record_fields(TABLE *table, size_t master_reclength, my_ptrdiff_t master_fields) { + DBUG_ENTER("copy_extra_record_fields(table, master_reclen, master_fields)"); DBUG_PRINT("info", ("Copying to 0x%lx " "from field %lu at offset %lu " "to field %d at offset %lu", @@ -6956,7 +7179,11 @@ copy_extra_record_fields(TABLE *table, Copying the extra fields of the slave that does not exist on master into record[0] (which are basically the default values). */ - DBUG_ASSERT(master_reclength <= table->s->reclength); + + if (table->s->fields < (uint) master_fields) + DBUG_RETURN(0); + + DBUG_ASSERT(master_reclength <= table->s->reclength); if (master_reclength < table->s->reclength) bmove_align(table->record[0] + master_reclength, table->record[1] + master_reclength, @@ -7013,7 +7240,7 @@ copy_extra_record_fields(TABLE *table, } } } - return 0; // All OK + DBUG_RETURN(0); // All OK } #define DBUG_PRINT_BITSET(N,FRM,BS) \ @@ -7603,12 +7830,6 @@ int Delete_rows_log_event::do_prepare_row(THD *thd, RELAY_LOG_INFO const *rli, uchar const **const row_end) { DBUG_ASSERT(row_start && row_end); - /* - This assertion actually checks that there is at least as many - columns on the slave as on the master. - */ - DBUG_ASSERT(table->s->fields >= m_width); - if (int error= unpack_row(rli, table, m_width, row_start, &m_cols, row_end, &m_master_reclength, table->read_set, DELETE_ROWS_EVENT)) { @@ -7787,12 +8008,6 @@ int Update_rows_log_event::do_prepare_row(THD *thd, RELAY_LOG_INFO const *rli, int error; DBUG_ASSERT(row_start && row_end); /* - This assertion actually checks that there is at least as many - columns on the slave as on the master. - */ - DBUG_ASSERT(table->s->fields >= m_width); - - /* We need to perform some juggling below since unpack_row() always unpacks into table->record[0]. For more information, see the comments for unpack_row(). diff --git a/sql/log_event.h b/sql/log_event.h index 6787aa08016..e22a9785736 100644 --- a/sql/log_event.h +++ b/sql/log_event.h @@ -2049,6 +2049,8 @@ public: virtual int get_data_size() { return m_data_size; } #ifndef MYSQL_CLIENT + virtual const int calc_field_metadata_size(); + virtual int save_field_metadata(); virtual bool write_data_header(IO_CACHE *file); virtual bool write_data_body(IO_CACHE *file); virtual const char *get_db() { return m_dbnam; } @@ -2085,6 +2087,14 @@ private: flag_set m_flags; size_t m_data_size; + + uchar *m_field_metadata; // buffer for field metadata + /* + The size of field metadata buffer set by calling calc_field_metadata_size() + */ + ulong m_field_metadata_size; + uchar *m_null_bits; + uchar *m_meta_memory; }; diff --git a/sql/rpl_record.cc b/sql/rpl_record.cc index 94778948bf9..094ccbcdd7d 100644 --- a/sql/rpl_record.cc +++ b/sql/rpl_record.cc @@ -17,6 +17,8 @@ #include "rpl_rli.h" #include "rpl_record.h" #include "slave.h" // Need to pull in slave_print_msg +#include "rpl_utility.h" +#include "rpl_rli.h" /** Pack a record of data for a table into a format suitable for @@ -143,7 +145,8 @@ pack_row(TABLE *table, MY_BITMAP const* cols, @param rli Relay log info @param table Table to unpack into @param colcnt Number of columns to read from record - @param row Packed row data + @param row_data + Packed row data @param cols Pointer to columns data to fill in @param row_end Pointer to variable that will hold the value of the one-after-end position for the row @@ -191,7 +194,9 @@ unpack_row(RELAY_LOG_INFO const *rli, unsigned int null_mask= 1U; // The "current" null bits unsigned int null_bits= *null_ptr++; - for (field_ptr= begin_ptr ; field_ptr < end_ptr ; ++field_ptr) + uint i= 0; + table_def *tabledef= ((RELAY_LOG_INFO*)rli)->get_tabledef(table); + for (field_ptr= begin_ptr ; field_ptr < end_ptr && *field_ptr ; ++field_ptr) { Field *const f= *field_ptr; @@ -220,14 +225,42 @@ unpack_row(RELAY_LOG_INFO const *rli, f->set_notnull(); /* - We only unpack the field if it was non-null - */ - pack_ptr= f->unpack(f->ptr, pack_ptr); + We only unpack the field if it was non-null. + Use the master's size information if available else call + normal unpack operation. + */ + if (tabledef && tabledef->field_metadata(i)) + pack_ptr= f->unpack(f->ptr, pack_ptr, tabledef->field_metadata(i)); + else + pack_ptr= f->unpack(f->ptr, pack_ptr); } bitmap_set_bit(rw_set, f->field_index); null_mask <<= 1; } + i++; + } + + /* + throw away master's extra fields + */ + uint max_cols= min(tabledef->size(), cols->n_bits); + for (; i < max_cols; i++) + { + if (bitmap_is_set(cols, i)) + { + if ((null_mask & 0xFF) == 0) + { + DBUG_ASSERT(null_ptr < row_data + master_null_byte_count); + null_mask= 1U; + null_bits= *null_ptr++; + } + DBUG_ASSERT(null_mask & 0xFF); // One of the 8 LSB should be set + + if (!((null_bits & null_mask) && tabledef->maybe_null(i))) + pack_ptr+= tabledef->calc_field_size(i, (uchar *) pack_ptr); + null_mask <<= 1; + } } /* diff --git a/sql/rpl_rli.h b/sql/rpl_rli.h index c458318594a..7ae21897e3f 100644 --- a/sql/rpl_rli.h +++ b/sql/rpl_rli.h @@ -18,6 +18,7 @@ #include "rpl_tblmap.h" #include "rpl_reporting.h" +#include "rpl_utility.h" struct RPL_TABLE_LIST; @@ -301,6 +302,15 @@ typedef struct st_relay_log_info : public Slave_reporting_capability uint tables_to_lock_count; /* RBR: Count of tables to lock */ table_mapping m_table_map; /* RBR: Mapping table-id to table */ + inline table_def *get_tabledef(TABLE *tbl) + { + table_def *td= 0; + for (TABLE_LIST *ptr= tables_to_lock; ptr && !td; ptr= ptr->next_global) + if (ptr->table == tbl) + td= &((RPL_TABLE_LIST *)ptr)->m_tabledef; + return (td); + } + /* Last charset (6 bytes) seen by slave SQL thread is cached here; it helps the thread save 3 get_charset() per Query_log_event if the charset is not diff --git a/sql/rpl_utility.cc b/sql/rpl_utility.cc index a04bcd1fab9..2a124c7b14e 100644 --- a/sql/rpl_utility.cc +++ b/sql/rpl_utility.cc @@ -16,17 +16,45 @@ #include "rpl_utility.h" #include "rpl_rli.h" -uint32 -field_length_from_packed(enum_field_types const field_type, - uchar const *const data) +/********************************************************************* + * table_def member definitions * + *********************************************************************/ + +/* + This function returns the field size in raw bytes based on the type + and the encoded field data from the master's raw data. +*/ +uint32 table_def::calc_field_size(uint col, uchar *master_data) { uint32 length; - switch (field_type) { - case MYSQL_TYPE_DECIMAL: + switch (type(col)) { case MYSQL_TYPE_NEWDECIMAL: - length= ~(uint32) 0; + length= my_decimal_get_binary_size(m_field_metadata[col] >> 8, + m_field_metadata[col] - ((m_field_metadata[col] >> 8) << 8)); + break; + case MYSQL_TYPE_DECIMAL: + case MYSQL_TYPE_FLOAT: + case MYSQL_TYPE_DOUBLE: + length= m_field_metadata[col]; + break; + case MYSQL_TYPE_SET: + case MYSQL_TYPE_ENUM: + case MYSQL_TYPE_STRING: + { + if (((m_field_metadata[col] & 0xff00) == (MYSQL_TYPE_SET << 8)) || + ((m_field_metadata[col] & 0xff00) == (MYSQL_TYPE_ENUM << 8))) + length= m_field_metadata[col] & 0x00ff; + else + { + length= m_field_metadata[col] & 0x00ff; + if (length > 255) + length= uint2korr(master_data) + 2; + else + length= (uint) *master_data + 1; + } break; + } case MYSQL_TYPE_YEAR: case MYSQL_TYPE_TINY: length= 1; @@ -45,12 +73,6 @@ field_length_from_packed(enum_field_types const field_type, length= 8; break; #endif - case MYSQL_TYPE_FLOAT: - length= sizeof(float); - break; - case MYSQL_TYPE_DOUBLE: - length= sizeof(double); - break; case MYSQL_TYPE_NULL: length= 0; break; @@ -58,8 +80,6 @@ field_length_from_packed(enum_field_types const field_type, length= 3; break; case MYSQL_TYPE_DATE: - length= 4; - break; case MYSQL_TYPE_TIME: length= 3; break; @@ -69,41 +89,33 @@ field_length_from_packed(enum_field_types const field_type, case MYSQL_TYPE_DATETIME: length= 8; break; - break; case MYSQL_TYPE_BIT: - length= ~(uint32) 0; - break; - default: - /* This case should never be chosen */ - DBUG_ASSERT(0); - /* If something goes awfully wrong, it's better to get a string than die */ - case MYSQL_TYPE_STRING: - length= uint2korr(data); + { + uint from_len= (m_field_metadata[col] >> 8U) & 0x00ff; + uint from_bit_len= m_field_metadata[col] & 0x00ff; + length= from_len + ((from_bit_len > 0) ? 1 : 0); break; - - case MYSQL_TYPE_ENUM: - case MYSQL_TYPE_SET: - case MYSQL_TYPE_VAR_STRING: + } case MYSQL_TYPE_VARCHAR: - length= ~(uint32) 0; // NYI + length= m_field_metadata[col] > 255 ? 2 : 1; // c&p of Field_varstring::data_length() + length+= length == 1 ? (uint32) *master_data : uint2korr(master_data); break; - case MYSQL_TYPE_TINY_BLOB: case MYSQL_TYPE_MEDIUM_BLOB: case MYSQL_TYPE_LONG_BLOB: case MYSQL_TYPE_BLOB: case MYSQL_TYPE_GEOMETRY: - length= ~(uint32) 0; // NYI + { + Field_blob fb(m_field_metadata[col]); + length= fb.get_packed_size(master_data); break; } - + default: + length= -1; + } return length; } -/********************************************************************* - * table_def member definitions * - *********************************************************************/ - /* Is the definition compatible with a table? @@ -121,23 +133,6 @@ table_def::compatible_with(RELAY_LOG_INFO const *rli_arg, TABLE *table) TABLE_SHARE const *const tsh= table->s; - /* - To get proper error reporting for all columns of the table, we - both check the width and iterate over all columns. - */ - if (tsh->fields < size()) - { - DBUG_ASSERT(tsh->db.str && tsh->table_name.str); - error= 1; - char buf[256]; - my_snprintf(buf, sizeof(buf), "Table width mismatch - " - "received %u columns, %s.%s has %u columns", - (uint) size(), tsh->db.str, tsh->table_name.str, - tsh->fields); - rli->report(ERROR_LEVEL, ER_BINLOG_ROW_WRONG_TABLE_DEF, - ER(ER_BINLOG_ROW_WRONG_TABLE_DEF), buf); - } - for (uint col= 0 ; col < cols_to_check ; ++col) { if (table->field[col]->type() != type(col)) diff --git a/sql/rpl_utility.h b/sql/rpl_utility.h index 79e69aecaeb..c91113c9e2a 100644 --- a/sql/rpl_utility.h +++ b/sql/rpl_utility.h @@ -25,8 +25,6 @@ struct st_relay_log_info; typedef st_relay_log_info RELAY_LOG_INFO; -uint32 -field_length_from_packed(enum_field_types field_type, uchar const *data); /** A table definition from the master. @@ -57,19 +55,96 @@ public: @param types Array of types @param size Number of elements in array 'types' + @param field_metadata Array of extra information about fields + @param metadata_size Size of the field_metadata array + @param null_bitmap The bitmap of fields that can be null */ - table_def(field_type *types, ulong size) - : m_size(size), m_type(new unsigned char [size]) + table_def(field_type *types, ulong size, uchar *field_metadata, + int metadata_size, uchar *null_bitmap) + : m_size(size), m_type(0), + m_field_metadata(0), m_null_bits(0), m_memory(NULL) { + m_memory= (uchar *)my_multi_malloc(MYF(MY_WME), + &m_type, size, + &m_field_metadata, size * sizeof(short), + &m_null_bits, (m_size + 7) / 8, + NULL); if (m_type) memcpy(m_type, types, size); else m_size= 0; + /* + Extract the data from the table map into the field metadata array + iff there is field metadata. The variable metadata_size will be + 0 if we are replicating from an older version server since no field + metadata was written to the table map. This can also happen if + there were no fields in the master that needed extra metadata. + */ + if (m_size && metadata_size) + { + int index= 0; + for (unsigned int i= 0; i < m_size; i++) + { + switch (m_type[i]) { + case MYSQL_TYPE_TINY_BLOB: + case MYSQL_TYPE_BLOB: + case MYSQL_TYPE_MEDIUM_BLOB: + case MYSQL_TYPE_LONG_BLOB: + case MYSQL_TYPE_DOUBLE: + case MYSQL_TYPE_FLOAT: + { + /* + These types store a single byte. + */ + m_field_metadata[i]= (uchar)field_metadata[index]; + index++; + break; + } + case MYSQL_TYPE_SET: + case MYSQL_TYPE_ENUM: + case MYSQL_TYPE_STRING: + { + short int x= field_metadata[index++] << 8U; // real_type + x = x + field_metadata[index++]; // pack or field length + m_field_metadata[i]= x; + break; + } + case MYSQL_TYPE_BIT: + { + short int x= field_metadata[index++]; + x = x + (field_metadata[index++] << 8U); + m_field_metadata[i]= x; + break; + } + case MYSQL_TYPE_VARCHAR: + { + /* + These types store two bytes. + */ + char *ptr= (char *)&field_metadata[index]; + m_field_metadata[i]= sint2korr(ptr); + index= index + sizeof(short int); + break; + } + case MYSQL_TYPE_NEWDECIMAL: + { + short int x= field_metadata[index++] << 8U; // precision + x = x + field_metadata[index++]; // decimals + m_field_metadata[i]= x; + break; + } + default: + m_field_metadata[i]= 0; + break; + } + } + } + if (m_size && null_bitmap) + memcpy(m_null_bits, null_bitmap, (m_size + 7) / 8); } ~table_def() { - if (m_type) - delete [] m_type; + my_free(m_memory, MYF(0)); #ifndef DBUG_OFF m_type= 0; m_size= 0; @@ -99,6 +174,48 @@ public: return m_type[index]; } + + /* + This function allows callers to get the extra field data from the + table map for a given field. If there is no metadata for that field + or there is no extra metadata at all, the function returns 0. + + The function returns the value for the field metadata for column at + position indicated by index. As mentioned, if the field was a type + that stores field metadata, that value is returned else zero (0) is + returned. This method is used in the unpack() methods of the + corresponding fields to properly extract the data from the binary log + in the event that the master's field is smaller than the slave. + */ + uint16 field_metadata(uint index) const + { + DBUG_ASSERT(index < m_size); + if (m_field_metadata) + return m_field_metadata[index]; + else + return 0; + } + + /* + This function returns whether the field on the master can be null. + This value is derived from field->maybe_null(). + */ + my_bool maybe_null(uint index) const + { + DBUG_ASSERT(index < m_size); + return ((m_null_bits[(index / 8)] & + (1 << (index % 8))) == (1 << (index %8))); + } + + /* + This function returns the field size in raw bytes based on the type + and the encoded field data from the master's raw data. This method can + be used for situations where the slave needs to skip a column (e.g., + WL#3915) or needs to advance the pointer for the fields in the raw + data from the master to a specific column. + */ + uint32 calc_field_size(uint col, uchar *master_data); + /** Decide if the table definition is compatible with a table. @@ -121,6 +238,9 @@ public: private: ulong m_size; // Number of elements in the types array field_type *m_type; // Array of type descriptors + short int *m_field_metadata; + uchar *m_null_bits; + uchar *m_memory; }; /** |