Recovery of lost or corrupted inno db tables(mysql uc 2010)

Recovery of lost or corrupted InnoDB tables MySQL User Conference 2010, Santa Clara [email_address] Percona Inc. http://MySQLPerformanceBlog.com

Agenda InnoDB format overview Internal system tables SYS_INDEXES and SYS_TABLES InnoDB Primary and Secondary keys Typical failure scenarios InnoDB recovery tool - - Three things are certain: Death, taxes and lost data. Guess which has occurred?

How MySQL stores data in InnoDB A table space (ibdata1) System tablespace(data dictionary, undo, insert buffer, etc.) PRIMARY indices (PK + data) SECONDARY indices (SK + PK) If the key is (f1, f2) it is stored as (f1, f2, PK) file per table (.ibd) PRIMARY index SECONDARY indices InnoDB pages size 16k (uncompressed) Every index is identified by index_id

How MySQL stores data in InnoDB Page identifier index_id TABLE: name test/site_folders, id 0 119, columns 9, indexes 1, appr.rows 1 COLUMNS: id: DATA_INT len 4 prec 0; name: type 12 len 765 prec 0; sites_count: DATA_INT len 4 prec 0; created_at: DATA_INT len 8 prec 0; updated_at: DATA_INT len 8 prec 0; DB_ROW_ID: DATA_SYS prtype 256 len 6 prec 0; DB_TRX_ID: DATA_SYS prtype 257 len 6 prec 0; DB_ROLL_PTR: DATA_SYS prtype 258 len 7 prec 0; INDEX: name PRIMARY, id 0 254 , fields 1/7, type 3 root page 271, appr.key vals 1, leaf pages 1, size pages 1 FIELDS: id DB_TRX_ID DB_ROLL_PTR name sites_count created_at updated_at mysql> CREATE TABLE innodb_table_monitor(x int) engine=innodb Error log:

InnoDB page format Fil Trailer Page Directory FREE SPACE USER RECORDS INFINUM+SUPREMUM RECORDS PAGE_HEADER FIL HEADER

InnoDB page format Fil Header the latest archived log file number at the time that FIL_PAGE_FILE_FLUSH_LSN was written (in the log) 4 FIL_PAGE_ARCH_LOG_NO "the file has been flushed to disk at least up to this lsn" (log serial number), valid only on the first page of the file 8 FIL_PAGE_FILE_FLUSH_LSN current defined types are: FIL_PAGE_INDEX , FIL_PAGE_UNDO_LOG , FIL_PAGE_INODE , FIL_PAGE_IBUF_FREE_LIST 2 FIL_PAGE_TYPE log serial number of page's latest log record 8 FIL_PAGE_LSN offset of next page in key order 4 FIL_PAGE_NEXT offset of previous page in key order 4 FIL_PAGE_PREV ordinal page number from start of space 4 FIL_PAGE_OFFSET 4 ID of the space the page is in 4 FIL_PAGE_SPACE Remarks Size Name Data are stored in FIL_PAGE_INODE == 0x03

InnoDB page format Page Header "file segment header for the non-leaf pages in a B-tree" (this is irrelevant here) 10 PAGE_BTR_SEG_TOP "file segment header for the leaf pages in a B-tree" (this is irrelevant here) 10 PAGE_BTR_SEG_LEAF identifier of the index the page belongs to 8 PAGE_INDEX_ID level within the index (0 for a leaf page) 2 PAGE_LEVEL the highest ID of a transaction which might have changed a record on the page (only set for secondary indexes) 8 PAGE_MAX_TRX_ID number of user records 2 PAGE_N_RECS number of consecutive inserts in the same direction, e.g. "last 5 were all to the left" 2 PAGE_N_DIRECTION either PAGE_LEFT , PAGE_RIGHT , or PAGE_NO_DIRECTION 2 PAGE_DIRECTION record pointer to the last inserted record 2 PAGE_LAST_INSERT "number of bytes in deleted records" 2 PAGE_GARBAGE record pointer to first free record 2 PAGE_FREE number of heap records; initial value = 2 2 PAGE_N_HEAP record pointer to first record in heap 2 PAGE_HEAP_TOP number of directory slots in the Page Directory part; initial value = 2 2 PAGE_N_DIR_SLOTS Remarks Size Name index_id Highest bit is row format(1 -COMPACT, 0 - REDUNDANT )

InnoDB page format (REDUNDANT) Extra bytes pointer to next record in page 16 bits next 16 bits 1 if each Field Start Offsets is 1 byte long (this item is also called the "short" flag) 1 bit 1byte_offs_flag number of fields in this record, 1 to 1023 10 bits n_fields record's order number in heap of index page 13 bits heap_no number of records owned by this record 4 bits n_owned 1 if record is predefined minimum record 1 bit min_rec_flag 1 if record is deleted 1 bit deleted_flag _ORDINAR Y, _NODE_PTR , _INFIMUM , _SUPREMUM 2 bit s record_status Description Size Name

InnoDB page format (COMPACT) Extra bytes a relative pointer to the next record in the page 16 next 16 bits 000=conventional, 001=node pointer (inside B-tree), 010=infimum, 011=supremum, 1xx=reserved 3 record type the order number of this record in the heap of the index page 1 3 heap_no the number of records owned by this record (this term is explained in page0page.h) 4 n_owned 4 bits used to delete mark a record, and mark a predefined minimum record in alphabetical order 4 record_statu s deleted_fla g min_rec_flag Description Size , bits Name

How to check row format? The highest bit of the PAGE_N_HEAP from the page header 0 stands for version REDUNDANT , 1 - for COMACT dc -e "2o `hexdump –C d pagefile | grep 00000020 | awk '{ print $12}'` p" | sed 's/./& /g' | awk '{ print $1}'

Rows in an InnoDB page Rows in a single pages is a linked list The first record INFIMUM The last record SUPREMUM Sorted by Primar y key infimum next supremum 0 100 data... next 101 data... next 102 data... next 103 data... next

Records are saved in insert order insert into t1 values(10, 'aaa'); insert into t1 values(30, ' ccc '); insert into t1 values(20, ' bbb '); JG....................N<E....... ................................ .............................2.. ...infimum......supremum......6. ........)....2.. aaa ............. ...*....2.. ccc .... ...........+. ...2.. bbb ....................... ................................

Row format EXAMPLE: CREATE TABLE ` t1 ` ( ` ID ` int( 11 ) unsigned NOT NULL, ` NAME ` varchar(120), ` N_FIELDS ` int(10), PRIMARY KEY (`ID`) ) ENGINE=InnoDB DEFAULT CHARSET=latin1 depends on content Field Contents 6 bytes (5 bytes if COMPACT format) Extra Bytes (F*1) or (F*2) bytes Field Start Offsets Size Name

REDUNDANT A row: (10 , ‘abcdef’, 20 ) 4 6 7 Actualy stored as: (10 , TRX_ID, PTR_ID, ‘abcdef’, 20 ) 6 4 Field Offsets … . next Extra 6 bytes: 0x00 00 00 0A record_status deleted_flag min_rec_flag n_owned heap_no n_fields 1byte_offs_flag Fields ... ... abcdef 0x80 00 00 14

COMPACT A row: (10 , ‘abcdef’, 20 ) 6 NULLS Actualy stored as: (10 , TRX_ID, PTR_ID, ‘abcdef’, 20 ) Field Offsets … . next Extra 5 bytes: 0x00 00 00 0A Fields ... ... abcdef 0x80 00 00 14 A bit per NULL-able field

Data types INT types (fixed-size) String types VARCHAR(x) – variable-size CHAR(x) – fixed-size, variable-size if UTF-8 DECIMAL Stored in strings before 5.0.3, variable in size Binary format after 5.0.3, fixed-size.

BLOB and other long fields Field length (so called offset) is one or two byte long Page size is 16k If record size < (UNIV_PAGE_SIZE/2-200) == ~7k – the record is stored internally (in a PK page) Otherwise – 768 bytes internally, the rest in an external page

2 . Internal system tables SYS_INDEXES and SYS_TABLES

Why are SYS_* tables needed? Correspondence “table name” -> “index_id” Storage for other internal information

How MySQL stores data in InnoDB SYS_TABLES and SYS_INDEXES Always REDUNDANT format! CREATE TABLE `SYS_INDEXES` ( ` TABLE_ID ` bigint(20) unsigned NOT NULL default '0', ` ID ` bigint(20) unsigned NOT NULL default '0', ` NAME ` varchar(120) default NULL, ` N_FIELDS ` int(10) unsigned default NULL, ` TYPE ` int(10) unsigned default NULL, ` SPACE ` int(10) unsigned default NULL, ` PAGE_NO ` int(10) unsigned default NULL, PRIMARY KEY (`TABLE_ID`,`ID`) ) ENGINE=InnoDB DEFAULT CHARSET=latin1 CREATE TABLE `SYS_TABLES` ( ` NAME ` varchar(255) NOT NULL default '', ` ID ` bigint(20) unsigned NOT NULL default '0', ` N_COLS ` int(10) unsigned default NULL, ` TYPE ` int(10) unsigned default NULL, ` MIX_ID ` bigint(20) unsigned default NULL, ` MIX_LEN ` int(10) unsigned default NULL, ` CLUSTER_NAME ` varchar(255) default NULL, ` SPACE ` int(10) unsigned default NULL, PRIMARY KEY (`NAME`) ) ENGINE=InnoDB DEFAULT CHARSET=latin1 index_id = 0-3 index_id = 0-1 Name: PRIMARY GEN_CLUSTER_ID or unique index name

How MySQL stores data in InnoDB NAME ID … "archive/msg_store" 40 8 1 0 0 NULL 0 "archive/msg_store" 40 8 1 0 0 NULL 0 "archive/msg_store" 40 8 1 0 0 NULL 0 TABLE_ID ID NAME … 40 196389 "PRIMARY" 2 3 0 21031026 4 0 196390 "msg_hash" 1 0 0 21031028 SYS_TABLES SYS_INDEXES Example:

3. InnoDB Primary and Secondary keys

Primary key The table: CREATE TABLE `t1` ( `ID` int(11), `NAME` varchar(120), `N_FIELDS` int(10), PRIMARY KEY (`ID`), KEY `NAME` (`NAME`) ) ENGINE=InnoDB DEFAULT CHARSET=latin1 Fields in the PK: ID DB_TRX_ID DB_ROLL_PTR NAME N_FIELDS

Secondary key The table: CREATE TABLE `t1` ( `ID` int(11), `NAME` varchar(120), `N_FIELDS` int(10), PRIMARY KEY (`ID`), KEY `NAME` (`NAME`) ) ENGINE=InnoDB DEFAULT CHARSET=latin1 Fields in the SK: NAME ID  Primary key

Deleted records DELETE FROM table WHERE id = 5; Forgotten WHERE clause Band-aid: Stop/kill mysqld ASAP

How delete is performed? "row/row0upd.c“ : “… /* How is a delete performed?...The delete is performed by setting the delete bit in the record and substituting the id of the deleting transaction for the original trx id, and substituting a new roll ptr for previous roll ptr. The old trx id and roll ptr are saved in the undo log record. Thus, no physical changes occur in the index tree structure at the time of the delete . Only when the undo log is purged, the index records will be physically deleted from the index trees.…”

Dropped table/database DROP TABLE table; DROP DATABASE database; Often happens when restoring from SQL dump Bad because .FRM file goes away Especially painful when innodb_file_per_table Band-aid: Stop/kill mysqld ASAP Stop IO on an HDD or mount read-only or take a raw image

Corrupted InnoDB tablespace Hardware failures OS or filesystem failures InnoDB bugs Corrupted InnoDB tablespace by other processes Band-aid: Stop mysqld Take a copy of InnoDB files

Wrong UPDATE statement UPDATE user SET Password = PASSWORD(‘qwerty’) WHERE User=‘root’ ; Again forgotten WHERE clause Bad because changes are applied in a PRIMARY index immediately Old version goes to UNDO segment Band-aid: Stop/kill mysqld ASAP

Recovery prerequisites Media ibdata1 *.ibd HDD image Tables structure SQL dump *.FRM files

table_defs.h { /* int(11) unsigned */ name: “ I D", type: FT_UINT, fixed_length: 4, has_limits: TRUE, limits: { can_be_null: FALSE, uint_min_val: 0, uint_max_val: 4294967295ULL }, can_be_null: FALSE }, { /* varchar(120) */ name: "NAME", type: FT_CHAR, min_length: 0, max_length: 120, has_limits: TRUE, limits: { can_be_null: TRUE, char_min_len: 0, char_max_len: 120, char_ascii_only: TRUE }, can_be_null: TRUE }, generated by create_defs.pl

How to get CREATE info from .frm files 1. CREATE TABLE t1 (id int) Engine=INNODB; 2. Replace t1.frm with the one’s you need to get scheme 3. R un “show create table t1” If mysqld crashes See the end of bvi t1.frm : .ID.NAME.N_FIELDS.. 2. *.FRM viewer !TODO

InnoDB recovery tool http:// launchpad.net / percona -innodb-recovery-tool / Written in Percona Contributed by Percona and community Supported by Percona Consists of two major tools page_parser – splits InnoDB tablespace into 16k pages constraints_parser – scans a page and finds good records

InnoDB recovery tool server # ./page_parser -4 -f /var/lib/mysql/ibdata1 Opening file: /var/lib/mysql/ibdata1 Read data from fn=3... Read page #0.. saving it to pages-1259793800/0-18219008/0-00000000.page Read page #1.. saving it to pages-1259793800/0-0/1-00000001.page Read page #2.. saving it to pages-1259793800/4294967295-65535/2-00000002.page Read page #3.. saving it to pages-1259793800/0-0/3-00000003.page page_parser

Page signature check 0{....guatda.com/cmx.p0...4...4......=..E....... ........<..~...A.......|........ ................................ ... infimum ...... supremum f .....qT M/T/196001834/ XXXXX XXXXXXXXXXX L X X X X X X X X X XXXXXX X X XX X X X INFIMUM and SUPREMUM records are in fixed positions Works with corrupted pages

InnoDB recovery tool server # ./constraints_parser -4 -f pages-1259793800/0-16/51-00000051.page constraints_parser Table structure is defined in "include/table_defs.h" See HOWTO for details http://code.google.com/p/innodb-tools/wiki/InnodbRecoveryHowto Filters inside table_defs.h are very important

Check InnoDB page before reading recs # ./constraints_parser -5 -U -f pages/0-418/12665-00012665.page -V Initializing table definitions... Processing table: document_type_fieldsets_link - total fields: 5 - nullable fields: 0 - minimum header size: 5 - minimum rec size: 25 - maximum rec size: 25 Read data from fn=3... Page id: 12665 Checking a page Infimum offset: 0x63 Supremum offset: 0x70 Next record at offset: 0x9F (159) Next record at offset: 0xB0 (176) Next record at offset: 0x3D95 (15765) … Next record at offset: 0x70 (112) Page is good Check if the tool can follow all records by addresses If so, find a rec. exactly at the position where the record is. Helps a lot for COMPACT format!

Import result t1 1 "browse" 10 t1 2 "dashboard" 20 t1 3 "addFolder" 18 t1 4 "editFolder" 15 mysql> LOAD DATA INFILE '/path/to/datafile' REPLACE INTO TABLE <table_name> FIELDS TERMINATED BY '\t' OPTIONALLY ENCLOSED BY '"' LINES STARTING BY '<table_name>\t' ;

Questions ? Thank you for coming! References http://www.mysqlperformanceblog.com/ http://percona.com/ http://www.slideshare.net/guest808c167/recovery-of-lost-or-corrupted-inno-db-tablesmysql-uc-2010 - - Applause :-)

Recovery of lost or corrupted inno db tables(mysql uc 2010)

More Related Content

What's hot (20)

Similar to Recovery of lost or corrupted inno db tables(mysql uc 2010) (20)

More from Aleksandr Kuzminsky (8)

Recently uploaded (20)

Recovery of lost or corrupted inno db tables(mysql uc 2010)