1. 1 Crash Recovery Chapter 18

2. 2 Review: The ACID properties  A  A tomicity: All actions of the Xact happen, or none happen.  C  C onsistency: If each Xact is consistent, and the DB starts consistent, it ends up consistent.  I  I solation: Execution of one Xact is isolated from that of other Xacts.  D  D urability: If a Xact commits, its effects persist.  The Recovery Manager guarantees Atomicity & Durability.

3. 3 Failure and Recovery  Failure and consistency  Transaction failures  System failures  Media failures  Principle of recovery  Redundancy  Database can be protected by ensuring that its correct state can be reconstructed from information stored redundantly in the system  Recovering database – restart operation  Bringing the stable DB to a consistent state by removing effects of uncommitted transactions and applying missing effects of committed transactions.

4. 4 Recovery and Restart  Types of storage media  Volatile storage: fast, but not surviving system failures  Non-volatile storage  Stable storage: information never lost (practically)  Recovery  Ideally, stable DB should contain, for each data item, the last value written by committed transaction  Practically, stable DB may contain values written by uncommitted transactions, or may not contain the last committed values.  Why? 1) Updating of uncommitted T 2) Buffering of committed values in the cache

5. 5 Function of Recovery Manager  Atomicity:  Transactions may abort (“Rollback”).  Durability:  What if DBMS stops running? (Causes?) crash! v Desired Behavior after system restarts: –T1, T2 & T3 should be durable. –T4 & T5 should be aborted (effects not seen). T1 T2 T3 T4 T5

6. 6 Recovery Management  Design rules for recovery manager  Undo rule: committed values must be saved before overwritten by uncommitted values in the stable DB  Redo rule: before commit, new values it wrote must be in the stable storage (DB or log)  Restart activity  Preparation: during normal operation  Actual recovery: after failure  Preparation  Logging  Checkpointing

7. 7 Cache Manager  Two operations: fetch and flush  Use dirty bit for deciding flushing operation  Flush: if the slot in cache is not dirty, do nothing; otherwise, copy the value into stable storage  Fetch: select a slot, using replacement algorithm if full (and flush if necessary), copy the value into slot, reset dirty bit, update cache directory  When to flush?  Depends on recovery strategy of the system  Different recovery algorithms use different strategies  Idempotence of restart  Any sequence of incomplete execution, followed by a complete execution of restart has the same effect of just one complete execution

8. 8

9. 9 Doctors Quotes from medical records as dictated by physicians: 1. The patient has been depressed ever since she began seeing me in 1983. 2. The patient refused an autopsy. 3. The patient has no past history of suicides. 4. The patient's past medical history has been remarkably insignificant with only a 40 lb weight gain in the last 3 days. 5. She slipped on the ice and apparently her legs went in different directions in early December. 6. The patient had waffles for breakfast and anorexia for lunch. 7. She is numb from her toes down. 8. While in the ER, she was examined, x-rated and sent home. 9. Occasional, constant, infrequent headaches. 10. Coming from Detroit, this man has no children.

10. 10 Assumptions  Concurrency control is in effect.  Strict 2PL, in particular.  Updates are happening “in place”.  i.e. data is overwritten on (deleted from) the disk.  A simple scheme to guarantee Atomicity and Durability?

11. 11 Handling the Memory Pool  Write to disk: force/no-force  Cache page: steal/no-steal  Force every write to disk?  Poor response time.  But provides durability.  Steal buffer-pool frames from uncommited Xacts?  If not, poor throughput.  If so, how can we ensure atomicity? Force No Force No Steal Steal Trivial Desired

12. 12 More on Steal and Force  STEAL (why enforcing Atomicity is hard)  To steal frame F: Current page in F (say P) is written to disk; some Xact holds lock on P. What if the Xact with the lock on P aborts? Must remember the old value of P at steal time (to support UNDOing the write to page P).  NO FORCE (why enforcing Durability is hard)  What if system crashes before a modified page is written to disk?  Write as little as possible, in a convenient place, at commit time, to support REDOing modifications.

13. 13 Recovery Algorithms  Undo/redo algorithm  Most complicated of the four recovery algorithms  Flexible in deciding when to flush (no-force)  Maximize efficiency during normal operation at the expense of less efficient recovery  Comparison with other recovery algorithms  Issues: disk I/O, log space, recovery time  No-redo requires more frequent flush (force)  Uncommitted transaction is allowed to replace dirty slot for in-place update – undo might be necessary  Restart procedure  Process log forward and backward for redo and undo

14. 14 Undo/Redo Recovery  A transaction T writes vale V to data object X. What will happen?  System fetches X if it is not already in cache  Record V in the log and in X’s slot C  No need for the cache manager to flush C  If cache manager replaces C (steal), and either T aborts or system fails before T commits, undo is required  If T commits and system fails before C is flushed (no force), redo is required

15. 15 Restart Procedure for Undo/Redo Recovery 1. Discard all cache slots 2. Scan the log to analyze which transactions committed, aborted, or active, to determine data for redo/undo 3. Redo all actions that were committed but not recorded in the stable DB 4. Undo all actions of transactions that were aborted or active at the time of failure

16. 16 Undo/No-redo Recovery  It records all the updates of a transaction in the stable DB before commit  Before adding T to the commit list, flush each data updated by T in the cache  Commit procedure  Flush all updated by T  Add T to the commit list  Acknowledge commitment of T  Delete T from active list  Log records for update after images not necessary  Increase response time for increased I/O workload

17. 17

18. 18 Break Time  OFFICE ARITHMETIC  Smart boss + smart employee = profit  Smart boss + dumb employee = production  Dumb boss + smart employee = promotion  Dumb boss + dumb employee = overtime  SHOPPING MATH  A man will pay $2 for a $1 item he needs.  A woman will pay $1 for a $2 item that she doesn't need.

19. 19 Key Idea: Logging  Record REDO and UNDO information, for every update, in a log.  Sequential writes to log (put it on a separate disk).  Minimal info written to log, so multiple updates fit in a single log page.  Log: An ordered list of REDO/UNDO actions  Log record contains:  and additional control info (which we’ll see soon).

20. 20 Write-Ahead Logging (WAL)  The Write-Ahead Logging Protocol:  Must force the log record for an update before the corresponding data page gets to disk. ‚ Must write all log records for a Xact before commit.  #1 guarantees Atomicity. How?  #2 guarantees Durability. How?  Exactly how is logging and recovery done?  Example: ARIES algorithms.

21. 21 Crash Recovery: Big Picture v Start from a checkpoint (found via master record). v Three phases. Need to: –Figure out which Xacts committed since checkpoint, which failed (Analysis). –REDO all actions. u (repeat history) –UNDO effects of failed Xacts. Oldest log rec. of Xact active at crash Smallest recLSN in dirty page table after Analysis Last chkpt CRASH A RU

22. 22 ARIES  Each log record has a unique Log Sequence Number (LSN).  LSNs always increasing.  Each data page contains a pageLSN.  The LSN of the most recent log record for an update to that page.  System keeps track of flushedLSN.  The max LSN flushed so far.  WAL: Before a page is written,  pageLSN  flushedLSN LSNs DB pageLSNs RAM flushedLSN pageLSN Log records flushed to disk “Log tail” in RAM

23. 23 Log Records Possible log record types:  Update  Commit  Abort  End (signifies end of commit or abort)  Compensation Log Records (CLRs)  for UNDO actions prevLSN XID type length pageID offset before-image after-image LogRecord fields: update records only

24. 24 Other Log-Related State  Transaction Table:  One entry per active Xact.  Contains XID, status (running/commited/aborted), and lastLSN.  Dirty Page Table:  One entry per dirty page in buffer pool.  Contains recLSN -- the LSN of the log record which first caused the page to be dirty.

25. 25 Normal Execution of an Xact  Series of reads & writes, followed by commit or abort.  We will assume that write is atomic on disk. In practice, additional details to deal with non-atomic writes.  Strict 2PL.  STEAL, NO-FORCE buffer management, with Write-Ahead Logging.

26. 26 Checkpointing  Periodically, the DBMS creates a checkpoint, in order to minimize the time taken to recover in the event of a system crash. Write to log:  begin_checkpoint record: Indicates when chkpt began.  end_checkpoint record: Contains current Xact table and dirty page table. This is a `fuzzy checkpoint’: Other Xacts continue to run; so these tables accurate only as of the time of the begin_checkpoint record. No attempt to force dirty pages to disk; effectiveness of checkpoint limited by oldest unwritten change to a dirty page. (So it’s a good idea to periodically flush dirty pages to disk!)  Store LSN of chkpt record in a safe place ( master record).

27. 27 The Big Picture: What’s Stored Where DB Data pages each with a pageLSN Xact Table lastLSN status Dirty Page Table recLSN flushedLSN RAM prevLSN XID type length pageID offset before-image after-image LogRecords LOG master record

28. 28 Simple Transaction Abort  For now, consider an explicit abort of a Xact.  No crash involved.  We want to “play back” the log in reverse order, UNDO ing updates.  Get lastLSN of Xact from Xact table.  Can follow chain of log records backward via the prevLSN field.  Before starting UNDO, write an Abort log record. For recovering from crash during UNDO!

29. 29 Abort, cont.  To perform UNDO, must have a lock on data!  No problem - Why?  Before restoring old value of a page, write a CLR:  You continue logging while you UNDO!!  CLR has one extra field: undonextLSN Points to the next LSN to undo (i.e. the prevLSN of the record we’re currently undoing).  CLRs never Undone (but they might be Redone when repeating history: guarantees Atomicity!)  At end of UNDO, write an “end” log record.

30. 30 Transaction Commit  Write commit record to log.  All log records up to Xact’s lastLSN are flushed.  Guarantees that flushedLSN  lastLSN.  Note that log flushes are sequential, synchronous writes to disk.  Many log records per log page.  Commit() returns.  Write end record to log.

31. 31 Recovery: The Analysis Phase  Reconstruct state at checkpoint.  via end_checkpoint record.  Scan log forward from checkpoint.  End record: Remove Xact from Xact table.  Other records: Add Xact to Xact table, set lastLSN=LSN, change Xact status on commit.  Update record: If P not in Dirty Page Table, Add P to D.P.T., set its recLSN=LSN.

32. 32 Recovery: The REDO Phase  We repeat History to reconstruct state at crash:  Reapply all updates (even of aborted Xacts!), redo CLRs.  Scan forward from log rec containing smallest recLSN in D.P.T. For each CLR or update log rec LSN, REDO the action unless:  Affected page is not in the Dirty Page Table, or  Affected page is in D.P.T., but has recLSN > LSN, or  pageLSN (in DB)  LSN.  To REDO an action:  Reapply logged action.  Set pageLSN to LSN. No additional logging!

33. 33 Recovery: The UNDO Phase ToUndo={ l | l a lastLSN of a “loser” Xact} Repeat:  Choose largest LSN among ToUndo.  If this LSN is a CLR and undonextLSN==NULL Write an End record for this Xact.  If this LSN is a CLR, and undonextLSN != NULL Add undonextLSN to ToUndo  Else this LSN is an update. Undo the update, write a CLR, add prevLSN to ToUndo. Until ToUndo is empty.

34. 34 Example of Recovery begin_checkpoint end_checkpoint update: T1 writes P5 update T2 writes P3 T1 abort CLR: Undo T1 LSN 10 T1 End update: T3 writes P1 update: T2 writes P5 CRASH, RESTART LSN LOG 00 05 10 20 30 40 45 50 60 Xact Table lastLSN status Dirty Page Table recLSN flushedLSN ToUndo prevLSNs RAM

35. 35 Example: Crash During Restart! begin_checkpoint, end_checkpoint update: T1 writes P5 update T2 writes P3 T1 abort CLR: Undo T1 LSN 10, T1 End update: T3 writes P1 update: T2 writes P5 CRASH, RESTART CLR: Undo T2 LSN 60 CLR: Undo T3 LSN 50, T3 end CRASH, RESTART CLR: Undo T2 LSN 20, T2 end LSN LOG 00,05 10 20 30 40,45 50 60 70 80,85 90 Xact Table lastLSN status Dirty Page Table recLSN flushedLSN ToUndo undonextLSN RAM

36. 36 Additional Crash Issues  What happens if system crashes during Analysis? During REDO ?  How do you limit the amount of work in REDO ?  Flush asynchronously in the background.  Watch “hot spots”!  How do you limit the amount of work in UNDO ?  Avoid long-running Xacts.

37. 37

38. 38 Now I know why Prof. X gets good evaluations…

39. 39 Summary of Logging/Recovery  Recovery Manager guarantees Atomicity & Durability.  Use WAL to allow STEAL/NO-FORCE w/o sacrificing correctness.  LSNs identify log records; linked into backwards chains per transaction (via prevLSN).  pageLSN allows comparison of data page and log records.

40. 40 Summary, Cont.  Checkpointing: A quick way to limit the amount of log to scan on recovery.  Recovery works in 3 phases:  Analysis: Forward from checkpoint.  Redo: Forward from oldest recLSN.  Undo: Backward from end to first LSN of oldest Xact alive at crash.  Upon Undo, write CLRs.  Redo “repeats history”: Simplifies the logic!