1. Chapter 17: Recovery

2. Chapter 17: Recovery System Failure Classification Storage Structure Recovery and Atomicity Log-Based Recovery Shadow Paging Recovery With Concurrent Transactions Buffer Management Failure with Loss of Nonvolatile Storage Advanced Recovery Techniques ARIES Recovery Algorithm Remote Backup Systems

3. Database Backup Database backup and recovery constitiute important functions of a DBMS  Automatic scheduling to backup storage device  A full backup or dump of the database  A differential backup with only last modifications done after the previous backup copy  Only the transaction logs. Backups are stored in secured sites and in stable storage devices

4. Recovery Concepts Database recovery is the process of restoring the database to a correct / consistent state in the event of failure. Recovery process ensures the atomicity and durability properties of transaction There are many different types of failure, each of which must be handled appropriately. Among the causes of failure are:  System crashes due to hardware or software errors.  Media failures.  Natural disasters such as fire, flood, etc.  Human error on the part of operators or users, application software errors  Sabotage Whatever the cause of failure, there are two main effects that must be considered:  The loss of main memory, including database buffers  The loss of the disk copy of the database. The recovery scheme must also provide high availability, i.e. minimize the time for which the database is not usable after the crash.

5. Failure Classification Hardware failures  Disk failure / Media failure : a head crash or similar disk failure destroys all or part of disk storage Destruction is assumed to be detectable: disk drives use checksums to detect failures  Memory errors, bad disk sectors, disk full errors Software failures  Failures related to OS, DBMS etc.  Application Software errors : Logical errors in programs that access the database Network failures  Can occur when using client-server configuration or distributed database systems where multiple database systems are connected by a common network. Transaction failure :  Logical errors: transaction cannot complete due to some internal error condition  System errors: the database system must terminate an active transaction due to an error condition (e.g., deadlock) System crash: a power failure or other hardware or software failure causes the system to crash (loss of main memory) Natural physical disasters : fire, floods, earthquake etc. Human errors : due to carelessness, unintentional destruction of data by operators, users etc. Sabotage Fail-stop assumption: non-volatile storage contents are assumed to not be corrupted by system crash Database systems have numerous integrity checks to prevent corruption of disk data

6. Non-Catastrophic Failures A computer failure (system crash):  A hardware failure,  A software failure,  A network failure A transaction error:  Integer overflow,  Division by zero,  Logical error,  User interruption,  Exception condition Concurrency control enforcement:  Violated serializability  Deadlock

7. Recovery Algorithms Recovery algorithms are techniques to ensure database consistency and transaction atomicity and durability despite failures Recovery algorithms have two parts 1. Actions taken during normal transaction processing to ensure enough information exists to recover from failures 2. Actions taken after a failure to recover the database contents to a state that ensures atomicity, consistency and durability

8. Recovery Classification Recovery process first determines the type and extent of recovery required Entire database Only unstable committed records According to the type of a failure, recovery procedures classify to:  Recovery from a catastrophic (like disk crash) failure, and  Recovery from a non catastrophic failure Recovery from catastrophic failure is based on restoring a database back_up copy by redoing operations of committed transactions (stored in an archived log file) up to the time of the failure

9. Recovery Classification (continued) If a database becomes inconsistent due to a non catastrophic failure, the strategy is to reverse only those changes that made database inconsistent It is accomplished by undoing (and sometimes also redoing) some operations. A memory log file is used here From now on we consider only recovery from non disk crash failures (we suppose data on disk are safe) The recovery from non catastrophic failures can be based on many algorithms, as:  Log based recovery  Shadow paging

10. Storage Structure Volatile storage:  does not survive system crashes  examples: main memory, cache memory  Access to volatile storage is extremely fast Nonvolatile storage:  survives system crashes  examples: disk (online storage), tape, flash memory, magnetic tapes (archival storage), optical disk  Access to volatile storage is slower than volatile storage by several orders of magnitude Stable storage:  a mythical form of storage that survives all failures  approximated by maintaining multiple copies on distinct nonvolatile media

11. Stable-Storage Implementation To implement stable storage we need to replicate the needed information in several non-volatile storage media. Store archival backups on tapes Maintain multiple copies of each block  on separate disks at same location (Disk mirroring, RAID level 1)  Maintain copies at remote sites to protect against disasters such as fire or flooding. Failure during data transfer can still result in inconsistent copies: Block transfer between memory and disk can result in  Successful completion  Partial failure: destination block has incorrect information  Total failure: destination block was never updated If a data transfer failure occurs, the recovery procedure must detect and restore the block to a consistent stage.

12. Data Access Database systems are partitioned into fixed length storage units called blocks. Blocks are the unit of transfer between memory and disk Physical blocks are those blocks residing on the disk. Buffer blocks are the blocks residing temporarily in main memory. The area of memory where the blocks reside temporarily is called Disk Buffer Block movements between disk and main memory are initiated through the following two operations:  input(B) transfers the physical block B to main memory.  output(B) transfers the buffer block B to the disk, and replaces the appropriate physical block there. Each transaction T i has its private work-area in which local copies of all data items accessed and updated by it are kept.  T i 's local copy of a data item X is called x i. We assume, for simplicity, that each data item fits in, and is stored inside, a single block.

13. Data Access (Cont.) Transaction transfers data items between system buffer blocks and its private work-area using the following operations :  read(X) assigns the value of data item X to the local variable x i.  write(X) assigns the value of local variable x i to data item {X} in the buffer block.  both these commands may necessitate the issue of an input(B X ) instruction before the assignment, if the block B X in which X resides is not already in memory. Transactions  Perform read(X) while accessing X for the first time;  All subsequent accesses are to the local copy.  After last access, transaction executes write(X). output(B X ) need not immediately follow write(X). System can perform the output (or force output) operation when it deems fit.

14. Example of Data Access X Y A B x1x1 y1y1 buffer Buffer Block A Buffer Block B input(A) output(B) read(X) write(Y) disk work area of T 1 work area of T 2 memory x2x2

15. Recovery and Atomicity Modifying the database without ensuring that the transaction will commit may leave the database in an inconsistent state. Consider transaction T i that transfers $50 from account A to account B; goal is either to perform all database modifications made by T i or none at all. Several output operations may be required for T i (to output A and B). A failure may occur after one of these modifications have been made but before all of them are made.

16. Recovery and Atomicity (Cont.) To ensure atomicity despite failures, we first output information describing the modifications to stable storage without modifying the database itself. We study two approaches:  log-based recovery, and  shadow-paging We assume (initially) that transactions run serially, that is, one after the other.

17. Log Based Recovery In case of failure, a transaction must be aborted or committed to maintain data integrity Transaction Log plays an important role in database recovery Transaction is the basic unit of recovery A transaction begins with and ends with Two types of recovery  Forward Recovery  Backward Recovery

18. Forward Recovery (REDO) Forward recovery is also called Roll Forward Used in case of physical damage. For example disk crash, failures during writing of data to database buffers or during flushing buffers to secondary storage. Intermediate result of transaction is stored in database buffers. From buffers is transferred to secondary storage. Flushing operation of buffers could be triggered by COMMIT operation of transaction or automatically in the event of buffer becoming full. If failure occurs between writing to buffers and flushing of buffers to secondary storage, Recovery manger determines the status of transaction that perform WRITE at time of failures. If COMMIT is already issued, recovery manager will redo (roll forward) so that updates are written to database. Forward recovery guarantees durability of transaction.

19. Forward Recovery (REDO) Last Database Copy Transaction Log Recreated Database Roll forward program To recreate database 1.Read most recent copy of database 2.Read the Transaction Log 3.The Roll forward program reads log entries starting from the first and continuing upto the last entry 4.For each of the log entries, changes the data value of the database to the ‘after’ value

20. Backward Recovery (UNDO) Backward Recovery is also called Roll-Backward Used in case an error occurs in the midst of normal operation of database. Error could be a human keying in values or programs ending abnormally and leaving some of changes to database that is was suppose to made. If transaction had not committed at the time of failure, it causes inconsistency in the database. Recovery manager must undo (roll backward) any effects of the transaction to the database. Backward Recovery guarantees the atomicity property of the transaction

21. Backward Recovery (UNDO) Current Database Transaction Log Corrected Database Roll backward program To recreate database 1.Read the database in its current state 2.Read the Transaction Log and position it in the last entry 3.Read backward and reset each updated value to its ‘before image’ until you reach the point where the error occured 4.The Roll-backward program ‘undoes’ each transaction in the reverse order

22. Log-Based Recovery To be able to recover from failures DBMS maintains a log file A log is kept on stable storage.  The log is a sequence of log records, and maintains a record of update activities on the database. Typically, a log file contains records with following contents: [start_transaction, T ] [write_item, T, X, old_value, new_value] [read_item,T, X ] [commit, T ] [abort, T ] When transaction T i starts, it registers itself by writing a log record Before T i executes write(X), a log record is written, where V 1 is the value of X before the write, and V 2 is the value to be written to X.  Log record notes that T i has performed a write on data item X j X j had value V 1 before the write, and will have value V 2 after the write. When T i finishes it last statement, the log record is written. Two approaches using logs  Deferred database modification  Immediate database modification

23. Deferred Update The idea behind the deferred update is to postpone updates to the database until transaction reaches its commit command During the execution, updates are recorded in a log file and in cache buffers with database pages (all in RAM) When the COMMIT is reached, before it is executed all log updates are first written to the log file on disk, and then the transaction commits After that, corresponding updates are written from buffers to the database

24. Deferred Update Layout Database Log File Main Memory Database Buffers Transaction Program Log Buffer database updates log records Written before the actual COMMIT Written after the actual COMMIT

25. Deferred Update (continued) If a transaction fails before reaching COMMIT, there is no need to make any recovery Redo of operations is needed, if a system crash occurs after COMMIT, but before all changes are recorded in the database on disk Then the operations have to be redone from the log file (that is already on disk) to the database Using after images (item values intended to be written to the database) does redo Deferred update recovery log file has to contain only after images - the new database item values

26. Deferred Database Modification Transaction starts by writing record to log. A write(X) operation results in a log record being written, where V is the new value for X  Note: old value is not needed for this scheme The write is not performed on X at this time, but is deferred. When T i partially commits, is written to the log Finally, the log records are read and used to actually execute the previously deferred writes.

27. Deferred Database Modification (Cont.) During recovery after a crash, a transaction needs to be redone if and only if both and are there in the log. Redoing a transaction T i ( redoT i ) sets the value of all data items updated by the transaction to the new values. Crashes can occur while  the transaction is executing the original updates, or  while recovery action is being taken example transactions T 0 and T 1 (T 0 executes before T 1 ): T 0 : read (A)T 1 : read (C) A: - A - 50 C:-C- 100 Write (A) write (C) read (B) B:- B + 50 write (B)

28. Deferred Database Modification (Cont.) Below we show the log as it appears at three instances of time. If log on stable storage at time of crash is as in case: (a) No redo actions need to be taken (b) redo(T 0 ) must be performed since is present (c) redo(T 0 ) must be performed followed by redo(T 1 ) since and are present

29. Immediate Database Modification The main idea of immediate update:  When a transaction issues an update command, before and after images are recorded into a log file on disk, and (thereupon) the database can (but does not have to be) immediately updated The immediate database modification scheme allows database updates of an uncommitted transaction to be made as the writes are issued  since undoing may be needed, update logs must have both old value and new value Update log record must be written before database item is written  We assume that the log record is output directly to stable storage  Can be extended to postpone log record output, so long as prior to execution of an output(B) operation for a data block B, all log records corresponding to items B must be flushed to stable storage Output of updated blocks can take place at any time before or after transaction commit Order in which blocks are output can be different from the order in which they are written.

30. Immediate Update Layout Database Log File Main Memory Database Buffers Transaction Program Log Buffer database updates log records Written immediately after each update Written immediately or later

31. Immediate Database Modification (Cont.) Recovery procedure has two operations instead of one:  undo(T i ) restores the value of all data items updated by T i to their old values, going backwards from the last log record for T i  redo(T i ) sets the value of all data items updated by T i to the new values, going forward from the first log record for T i Both operations must be idempotent  That is, even if the operation is executed multiple times the effect is the same as if it is executed once Needed since operations may get re-executed during recovery When recovering after failure:  Transaction T i needs to be undone if the log contains the record, but does not contain the record.  Transaction T i needs to be redone if the log contains both the record and the record. Undo operations are performed first, then redo operations.

32. Immediate DB Modification Recovery Example Below we show the log as it appears at three instances of time. Recovery actions in each case above are: (a) undo (T 0 ): B is restored to 2000 and A to 1000. (b) undo (T 1 ) and redo (T 0 ): C is restored to 700, and then A and B are set to 950 and 2050 respectively. (c) redo (T 0 ) and redo (T 1 ): A and B are set to 950 and 2050 respectively. Then C is set to 600

33. Checkpoints Some DBMS use CHECKPOINT records in the log file to prevent unnecessary redo operations To take a checkpoint record, DBMS has to:  Temporarily to suspend operations of all transactions,  To force write results of all update operations of committed transactions from main memory buffers to disk,  Write a checkpoint record into the log file and force write log to disk  Resume executing transactions Only changes made by transactions that committed between the last checkpoint and a system failure have to be redone

34. Checkpoints (Cont.) During recovery we need to consider only the most recent transaction T i that started before the checkpoint, and transactions that started after T i. 1. Scan backwards from end of log to find the most recent record 2. Continue scanning backwards till a record is found. 3. Need only consider the part of log following the start record. Earlier part of log can be ignored during recovery, and can be erased whenever desired. 4. For all transactions (starting from T i or later) with no, execute undo(T i ). (Done only in case of immediate modification.) 5. Scanning forward in the log, for all transactions starting from T i or later with a, execute redo(T i ).

35. Example of Checkpoints T 1 can be ignored (updates already output to disk due to checkpoint) T 2 and T 3 redone. T 4 undone TcTc TfTf T1T1 T2T2 T3T3 T4T4 checkpoint system failure

36. Shadow Paging Shadow paging is an alternative to log-based recovery; this scheme is useful if transactions execute serially Idea: maintain two page tables during the lifetime of a transaction – the current page table, and the shadow page table Store the shadow page table in nonvolatile storage, such that state of the database prior to transaction execution may be recovered.  Shadow page table is never modified during execution To start with, both the page tables are identical. Only current page table is used for data item accesses during execution of the transaction. Whenever any page is about to be written for the first time  A copy of this page is made onto an unused page.  The current page table is then made to point to the copy  The update is performed on the copy

37. Sample Page Table

38. Example of Shadow Paging Shadow and current page tables after write to page 4

39. Shadow Paging (Cont.) To commit a transaction : 1. Flush all modified pages in main memory to disk 2. Output current page table to disk 3. Make the current page table the new shadow page table, as follows:  keep a pointer to the shadow page table at a fixed (known) location on disk.  to make the current page table the new shadow page table, simply update the pointer to point to current page table on disk Once pointer to shadow page table has been written, transaction is committed. No recovery is needed after a crash — new transactions can start right away, using the shadow page table. Pages not pointed to from current/shadow page table should be freed (garbage collected).

40. Show Paging (Cont.) Advantages of shadow-paging over log-based schemes  no overhead of writing log records  recovery is trivial Disadvantages :  Copying the entire page table is very expensive Can be reduced by using a page table structured like a B + -tree  No need to copy entire tree, only need to copy paths in the tree that lead to updated leaf nodes  Commit overhead is high even with above extension Need to flush every updated page, and page table  Data gets fragmented (related pages get separated on disk)  After every transaction completion, the database pages containing old versions of modified data need to be garbage collected  Hard to extend algorithm to allow transactions to run concurrently Easier to extend log based schemes

41. Recovery With Concurrent Transactions We modify the log-based recovery schemes to allow multiple transactions to execute concurrently.  All transactions share a single disk buffer and a single log  A buffer block can have data items updated by one or more transactions We assume concurrency control using strict two-phase locking;  i.e. the updates of uncommitted transactions should not be visible to other transactions Otherwise how to perform undo if T1 updates A, then T2 updates A and commits, and finally T1 has to abort? Logging is done as described earlier.  Log records of different transactions may be interspersed in the log. The checkpointing technique and actions taken on recovery have to be changed  since several transactions may be active when a checkpoint is performed.

42. Recovery With Concurrent Transactions (Cont.) Checkpoints are performed as before, except that the checkpoint log record is now of the form where L is the list of transactions active at the time of the checkpoint  We assume no updates are in progress while the checkpoint is carried out When the system recovers from a crash, it first does the following: 1. Initialize undo-list and redo-list to empty 2. Scan the log backwards from the end, stopping when the first record is found. For each record found during the backward scan: H if the record is, add T i to redo-list H if the record is, then if T i is not in redo-list, add T i to undo-list 3. For every T i in L, if T i is not in redo-list, add T i to undo-list

43. Recovery With Concurrent Transactions (Cont.) At this point undo-list consists of incomplete transactions which must be undone, and redo-list consists of finished transactions that must be redone. Recovery now continues as follows: 1. Scan log backwards from most recent record, stopping when records have been encountered for every T i in undo- list. n During the scan, perform undo for each log record that belongs to a transaction in undo-list. 2. Locate the most recent record. 3. Scan log forwards from the record till the end of the log. n During the scan, perform redo for each log record that belongs to a transaction on redo-list

44. Example of Recovery Go over the steps of the recovery algorithm on the following log: /* Scan at step 1 comes up to here */

45. Log Record Buffering Log record buffering: log records are buffered in main memory, instead of being output directly to stable storage.  Log records are output to stable storage when a block of log records in the buffer is full, or a log force operation is executed. Log force is performed to commit a transaction by forcing all its log records (including the commit record) to stable storage. Several log records can thus be output using a single output operation, reducing the I/O cost.

46. Log Record Buffering (Cont.) The rules below must be followed if log records are buffered:  Log records are output to stable storage in the order in which they are created.  Transaction T i enters the commit state only when the log record has been output to stable storage.  Before a block of data in main memory is output to the database, all log records pertaining to data in that block must have been output to stable storage. This rule is called the write-ahead logging or WAL rule

47. Database Buffering Database maintains an in-memory buffer of data blocks  When a new block is needed, if buffer is full an existing block needs to be removed from buffer  If the block chosen for removal has been updated, it must be output to disk If a block with uncommitted updates is output to disk, log records with undo information for the updates are output to the log on stable storage first  (Write ahead logging) No updates should be in progress on a block when it is output to disk. Can be ensured as follows.  Before writing a data item, transaction acquires exclusive lock on block containing the data item  Lock can be released once the write is completed. Such locks held for short duration are called latches.  Before a block is output to disk, the system acquires an exclusive latch on the block Ensures no update can be in progress on the block

48. Buffer Management (Cont.) Database buffer can be implemented either  in an area of real main-memory reserved for the database, or  in virtual memory Implementing buffer in reserved main-memory has drawbacks:  Memory is partitioned before-hand between database buffer and applications, limiting flexibility.  Needs may change, and although operating system knows best how memory should be divided up at any time, it cannot change the partitioning of memory.

49. Buffer Management (Cont.) Database buffers are generally implemented in virtual memory in spite of some drawbacks:  When operating system needs to evict a page that has been modified, the page is written to swap space on disk.  When database decides to write buffer page to disk, buffer page may be in swap space, and may have to be read from swap space on disk and output to the database on disk, resulting in extra I/O! Known as dual paging problem.  Ideally when OS needs to evict a page from the buffer, it should pass control to database, which in turn should 1. Output the page to database instead of to swap space (making sure to output log records first), if it is modified 2. Release the page from the buffer, for the OS to use Dual paging can thus be avoided, but common operating systems do not support such functionality.

50. Failure with Loss of Nonvolatile Storage So far we assumed no loss of non-volatile storage Technique similar to checkpointing used to deal with loss of non-volatile storage  Periodically dump the entire content of the database to stable storage  No transaction may be active during the dump procedure; a procedure similar to checkpointing must take place Output all log records currently residing in main memory onto stable storage. Output all buffer blocks onto the disk. Copy the contents of the database to stable storage. Output a record to log on stable storage.

51. Recovering from Failure of Non-Volatile Storage To recover from disk failure  restore database from most recent dump.  Consult the log and redo all transactions that committed after the dump The dump of the database is also referred to as archival dump

52. Remote Backup Systems

53. Remote backup systems provide high availability by allowing transaction processing to continue even if the primary site is destroyed.

54. Remote Backup Systems (Cont.) Detection of failure: Backup site must detect when primary site has failed  to distinguish primary site failure from link failure maintain several communication links between the primary and the remote backup.  Heart-beat messages Transfer of control:  To take over control backup site first perform recovery using its copy of the database and all the log records it has received from the primary. Thus, completed transactions are redone and incomplete transactions are rolled back.  When the backup site takes over processing it becomes the new primary  To transfer control back to old primary when it recovers, old primary must receive redo logs from the old backup and apply all updates locally.

55. Remote Backup Systems (Cont.) Time to recover: To reduce delay in takeover, backup site periodically processes the redo log records (in effect, performing recovery from previous database state), performs a checkpoint, and can then delete earlier parts of the log. Hot-Spare configuration permits very fast takeover:  Backup continually processes redo log record as they arrive, applying the updates locally.  When failure of the primary is detected the backup rolls back incomplete transactions, and is ready to process new transactions. Alternative to remote backup: distributed database with replicated data  Remote backup is faster and cheaper, but less tolerant to failure

56. Remote Backup Systems (Cont.) Time to commit  Ensure durability of updates by delaying transaction commit until update is logged at backup; avoid this delay by permitting lower degrees of durability. One-safe: commit as soon as transaction’s commit log record is written at primary  Problem: updates may not arrive at backup before it takes over. Two-very-safe: commit when transaction’s commit log record is written at primary and backup  Reduces availability since transactions cannot commit if either site fails. Two-safe: proceed as in two-very-safe if both primary and backup are active. If only the primary is active, the transaction commits as soon as is commit log record is written at the primary.  Better availability than two-very-safe; avoids problem of lost transactions in one-safe.

57. Database Mirroring : How it works Mirror Principal Witness Log Application SQL Server 2 2 4 5 1 Data Log 3>2>3 Mirror is always redoing – it remains current Commit

58. End of Chapter