1. PMIT-6102 Advanced Database Systems By- Jesmin Akhter Assistant Professor, IIT, Jahangirnagar University

2. Distributed DBMSSlide 2 Lecture -13 Reliability

3. Distributed DBMSSlide 3 Outline Reliability  Recovery Information  Logging  Checkpoints  Handing Media Failures –Full Architecture

4. Distributed DBMSSlide 4 Recovery Information  When a system failure occurs, the volatile database is lost. Therefore the DBMS has to maintain some information about the state at the time of its failure in order to be able to bring the database to the state that it was in when the failure occurred. This is called the recovery information.

5. Distributed DBMSSlide 5 Recovery Information The recovery information that the system maintains is dependent on the method of executing updates. Two possibilities are in-place updating and out- of-place updating. In-place update  Each update causes a change in one or more data values on pages in the database buffers  As a result, the previous values are lost. Out-of-place update  does not change the value of the data item in the stable database but maintains the new value separately.

6. Distributed DBMSSlide 6 In-Place Update Recovery Information New stable database state Database Log Update Operation Old stable database state Since in-place updates cause previous values of the affected data items to be lost, it is necessary to keep enough information about the database state changes to facilitate the recovery of the database to a consistent state following a failure. This information is typically maintained in a database log. Thus each update transaction not only changes the database but the change is also recorded in the database log

7. Distributed DBMSSlide 7 Logging The log contains information necessary to recover the database state following a failure. This information may include  transaction identifier  type of operation (action)  items accessed by the transaction to perform the action  old value (state) of item (before image)  new value (state) of item (after image) …

8. Distributed DBMSSlide 8 Why Logging? Upon recovery:  all of T 1 's effects should be reflected in the database (REDO if necessary due to a failure)  none of T 2 's effects should be reflected in the database (UNDO if necessary) 0t time system crash T1T1 Begin End BeginT2T2

9. Distributed DBMSSlide 9 REDO'ing an action means performing it again. The REDO operation uses the log information and performs the action that might have been done before, or not done due to failures. The REDO operation generates the new image. REDO Protocol Database Log REDO Old stable database state New stable database state

10. Distributed DBMSSlide 10 UNDO'ing an action means to restore the object to its before image. The UNDO operation uses the log information and restores the old value of the object. UNDO Protocol New stable database state Database Log UNDO Old stable database state

11. Distributed DBMSSlide 11 When to Write Log Records Into Stable Store Assume a transaction T updates a page P Fortunate case  System writes P in stable database  System updates stable log for this update  SYSTEM FAILURE OCCURS!... (before T commits) We can recover (undo) by restoring P to its old state by using the log Unfortunate case  System writes P in stable database  SYSTEM FAILURE OCCURS!... (before stable log is updated) We cannot recover from this failure because there is no log record to restore the old value. Solution: Write-Ahead Log (WAL) protocol

12. Distributed DBMSSlide 12 Write–Ahead Log Protocol Notice:  If a system crashes before a transaction is committed, then all the operations must be undone. Only need the before images ( undo portion of the log).  Once a transaction is committed, some of its actions might have to be redone. Need the after images ( redo portion of the log). WAL protocol :  Before a stable database is updated, the undo portion of the log should be written to the stable log  When a transaction commits, the redo portion of the log must be written to stable log prior to the updating of the stable database.

13. Distributed DBMSSlide 13 Logging Interface (see book) Read Write Read Main memory Local Recovery Manager Database Buffer Manager Fetch, Flush Secondary storage Stable log Stable database Database buffers (Volatile database) Log buffers Write Read

14. Distributed DBMSSlide 14 Typical techniques for out-of-place updating are shadowing and differential files Shadowing  When an update occurs, don't change the old page, but create a shadow page with the new values and write it into the stable database.  Update the access paths so that subsequent accesses are to the new shadow page.  The old page retained for recovery. Differential files  For each file F maintain  a read only part FR  a differential file consisting of insertions part DF+ and deletions part DF-  Thus, F = (FR  DF+) – DF-  Updates treated as delete old value, insert new value Out-of-Place Update Recovery Information

15. Distributed DBMSSlide 15 Commands to consider: begin_transaction read write commit abort recover Independent of execution strategy for LRM Execution of LRM Commands

16. Distributed DBMSSlide 16 There are five commands that form the interface to the LRM.  begin transaction, read, write, commit, and abort commands. A sixth interface command to the LRM: recover. The recover command is the interface that the operating system has to the LRM. It is used during recovery from system failures when the operating system asks the DBMS to recover the database to the state that existed when the failure occurred. Execution of LRM Commands

17. Distributed DBMSSlide 17 The execution of some of these commands abort, commit, and recover is quite dependent on the specific LRM algorithms and on the interaction of the LRM with the buffer manager. Others begin transaction, read, and write are quite independent of these considerations. Execution of LRM Commands

18. Distributed DBMSSlide 18 Begin transaction, Read, and Write Commands Begin transaction.  It causes the LRM to write a begin transaction record into the log. Read.  The read command specifies a data item. The LRM tries to read the specified data item from the buffer pages that belong to the transaction. If the data item is not in one of these pages, it issues a fetch command to the buffer manager in order to make the data available. Upon reading the data, the LRM returns it to the scheduler.

19. Distributed DBMSSlide 19 Write.  The write command specifies the data item and the new value. As with a read command, if the data item is available in the buffers of the transaction, its value is modified in the database buffers (i.e., the volatile database).  If it is not in the private buffer pages, a fetch command is issued to the buffer manager, and the data is made available and updated. The before image of the data page, as well as its after image, are recorded in the log. The local recovery manager then informs the scheduler that the operation has been completed successfully. Execution of Commands

20. Distributed DBMSSlide 20 The fundamental design decision in the implementation of the local recovery manager, the buffer manager, and the interaction between the two components is  whether or not the buffer manager obeys the local recovery manager’s instructions as to when to write the database buffer pages to stable storage. Specifically, two decisions are involved. The first one is  whether the buffer manager may write the buffer pages updated by a transaction into stable storage during the execution of that transaction,  or it waits for the LRM to instruct it to write them back. We call this the fix/no-fix decision. Execution of LRM Commands

21. Distributed DBMSSlide 21 The second decision is  whether the buffer manager will be forced to flush the buffer pages updated by a transaction into the stable storage at the end of that transaction (i.e., the commit point),  or the buffer manager flushes them out whenever it needs to according to its buffer management algorithm. We call this the flush/no-flush decision. It is also called the force/no-force decision Execution of LRM Commands

22. Distributed DBMSSlide 22 Possible execution strategies:  no-fix/no-flush  no-fix/flush  fix/no-flush  fix/flush Execution Strategies

23. Distributed DBMSSlide 23 This type of LRM algorithm is called a redo/undo algorithm, since it requires both the redo and undo operations upon recovery. It is called steal/no-force. Abort  Since the buffer manager may have written the updated pages into the stable database, abort will have to undo the actions of the transaction.  LRM performs transaction undo (or partial undo) Commit  LRM writes an “end_of_transaction” record into the log. Recover  For those transactions that have both a “begin_transaction” and an “end_of_transaction” record in the log, a partial redo is initiated by LRM  For those transactions that only have a “begin_transaction” in the log, a global undo is executed by LRM No-Fix/No-Flush

24. Distributed DBMSSlide 24 The LRM algorithms that use this strategy are called undo/no-redo and steal/force Abort  Buffer manager may have written some of the updated pages into stable database  LRM performs transaction undo (or partial undo) Commit  LRM issues a flush command to the buffer manager for all updated pages  LRM writes an “end_of_transaction” record into the log. Recover  No need to perform redo  Perform global undo No-Fix/Flush

25. Distributed DBMSSlide 25 The LRM controls the writing of the volatile database pages into stable storage. The key here is not to permit the buffer manager to write any updated volatile database page into the stable database until at least the transaction commit point. Abort  None of the updated pages have been written into stable database  Release the fix ed pages Commit  LRM writes an “end_of_transaction” record into the log.  LRM sends an unfix command to the buffer manager for all pages that were previously fix ed Recover  Perform partial redo  No need to perform global undo Fix/No-Flush

26. Distributed DBMSSlide 26 This is the case where the LRM forces the buffer manager to write the updated volatile database pages into the stable database at precisely the commit point—not before and not after. This strategy is called no-undo/no-redo and no-steal/force Abort  None of the updated pages have been written into stable database  Release the fix ed pages Commit (the following have to be done atomically)  LRM issues a flush command to the buffer manager for all updated pages  LRM sends an unfix command to the buffer manager for all pages that were previously fix ed  LRM writes an “end_of_transaction” record into the log. Recover  No need to do anything Fix/Flush

27. Distributed DBMSSlide 27 In most of the LRM implementation strategies, the execution of the recovery action requires searching the entire log.  The LRM is trying to find all the transactions that need to be undone and redone.  The overhead can be reduced if it is possible to build a wall which signifies that the database at that point is up-to-date and consistent.  In that case, the redo has to start from that point on  the undo only has to go back to that point.  This process of building the wall is called checkpointing. Checkpoints

28. Distributed DBMSSlide 28 Simplifies the task of determining actions of transactions that need to be undone or redone when a failure occurs. A checkpoint record contains a list of active transactions. Checkpointing is achieved in three steps  Write a begin_checkpoint record into the log  Collect the checkpoint data into the stable storage  Write an end_checkpoint record into the log Checkpoints

29. Distributed DBMSSlide 29 Handing Media Failures –Full Architecture Read Write Read Main memory Local Recovery Manager Database Buffer Manager Fetch, Flush Archive log Archive database Secondary storage Stable log Stable database Database buffers (Volatile database) Log buffers Write Read

30. Distributed DBMSSlide 30 Handing Media Failures –Full Architecture Media failures may either be quite catastrophic, causing the loss of the stable database or of the stable log, or they can simply result in partial loss of the database or the. To cope with catastrophic media failures, an archive copy of both the database and the log is maintained on a different (tertiary) storage medium, which is typically the magnetic tape or CD- ROM. Thus the DBMS deals with three levels of memory hierarchy: the main memory, random access disk storage, and magnetic tape (Figure). To deal with less catastrophic failures, having duplicate copies of the database and log may be sufficient. When a media failure occurs, the database is recovered from the archive copy by redoing and undoing the transactions as stored in the archive log.

31. Distributed DBMSSlide 31 Thank you