Introduction to Database Management Systems

Transcription

1 Database Administration Transaction Processing Why Concurrency Control? Locking Database Recovery Query Optimization DB Administration 1 Transactions Transaction -- A sequence of operations that is regarded as a single logical operation - i.e. a logical unit of work Atomicity of Transactions -- Either all operations in a transaction are executed completely or none of them is executed. Transaction Manager -- A component of a DBMS which is responsible for transaction processing. DB Administration 2 Commit and Rollback of Transactions TRANS: PROC OPTIONS (MAIN) ; /* declarations omitted */ EXEC SQL WHENEVER SQLERROR GO TO UNDO; GET LIST (SX, SY ) /* get values from enduser */ EXEC SQL UPDATE SUPPLIER SET S# = :SY WHERE S# = :SX; EXEC SQL UPDATE SHIPMENT SET S# = :SY WHERE S# = :SX; EXEC SQL COMMIT; GO TO FINISH UNDO: EXEC SQL ROLLBACK; FINISH: RETURN; END /*TRANS */; DB Administration 3 David McDonald,, GSU 11-1

2 Transaction Processing The execution of a transaction involves three steps: Transaction Start Signals the beginning a transaction execution Acquire required resources ( e.g., private workspace ) Transaction Execution -- Issues a sequence of reads/writes A read brings data from the DB to the private workspace A write updates data in the private workspace and the DB Transaction Commit/Rollback Commit -- All changes become permanent in the DB Rollback -- Stops the execution and undo all changes DB Administration 4 Synchronization Points A synchpoint represents a boundary between two consecutive transactions The point at which a database is guaranteed to be in a consistent state The only operations that establish a synchpoint are COMMIT, ROLLBACK and program initiation COMMITS and ROLLBACKS terminate the transaction, not the program All updates made by the program since the previous synchpoint are committed (COMMIT) or undone (ROLLBACK) All database positioning is lost ( all open cursors are closed) All record locks are released DB Administration 5 System Failure and Media Recovery System failures (e.g. power failure) - affect all transactions currently in progress, but do not physically damage the d.b. Media failure (e.g. head crash on the disk) - cause damage to the database, or to some portion of it, and affect at least those transactions currently using that portion System checkpoint - Taking a checkpoint involves physically writing the contents of the database buffers out to the physical database, and b) physically writing a special checkpoint record out to the physical log Undo a transaction - any transaction that was in progress at the time of failure Redo a transaction - a transaction that was completed, but was not written to disk successfully before a system failure DB Administration 6 David McDonald,, GSU 11-2

3 System & Media Recovery (cont d) Checkpoint Failure T3 T4 tc T5 tf A system failure has occurred at time tf The most recent checkpoint prior to time tf was taken at time tc Transactions of type were completed prior to time tc Transaction of type ed prior to time tc and completed after time tc and before time tf Transactions of type T3 also ed prior to time tc but did not complete by time tf Transactions of type T4 ed after time tc and completed before time tf Transactions of type T5 also ed after time tc but did not complete by time tf DB Administration 7 System Recovery (cont d) and T4 need to be Redone, while T3 and T5 need to be Undone To do this: 1. Start with two lists of transactions, the UNDO-list and the REDO-list Set the UNDO-list equal to the list of all transactions given in the checkpoint record; set the REDO-list to empty. 2. Search forward through the log, ing from the checkpoint record. 3. If a of transaction log entry is found for transaction T, add T to the UNDO-list. 4. If a commit log entry is found for transaction T, move T from the UNDOlist to the REDO-list 5. When the end of the log is reached, the UNDO- and REDO- lists identify, respectively, transactions of types T3 and T5, and transactions of type and T4. 6. The system now works backward through the log, undoing transactions in the UNDO-list; then it works forward again, redoing trans. in the REDO-list. DB Administration 8 Deferred Update Updates are not written to the database until after a transaction has reached its commit point. If transaction fails before commit, it will not have modified database and so no undoing of changes required. May be necessary to redo updates of committed transactions as their effect may not have reached database. DB Administration 9 David McDonald,, GSU 11-3

4 Immediate Update Updates are applied to database as they occur. When a transaction s, write a transaction to the log file. Once the log is written, write a record containing the necessary data to the log file. Once the log record is written, write the updates to the database buffers The updates to the database itself are written when the buffers are next flushed to secondary storage. Essential that log records are written before write to database. Write-ahead log protocol. Need to redo updates of committed transactions following a failure. May need to undo effects of transactions that had not committed at time of failure. DB Administration 10 Immediate Update If no transaction commit record in log, then that transaction was active at failure and must be undone. Undo operations are performed in reverse order in which they were written to log. DB Administration 11 Recovery Facilities Backup copies of the database Write-ahead logging (system log or journal ) Includes transaction IDs, time, before and after images of data record, etc. Checkpoint (synchronization point) Synchronizes transaction execution and logs Writes a checkpoint record to the log file Shadow copies (details on slide 14) DB Administration 12 David McDonald,, GSU 11-4

5 Database Recovery Media Failure Recovery Restored by the last backup copy Forward recovery if the system log is intact System Failure Recovery Using Logs Undo ( Backward Recovery ). The uncommitted changes made by a transaction to a database is undone. DB with changes Before Images Undo DB without changes Redo ( Forward Recovery ). The committed changes made by a transaction is reapplied to an earlier copy of the database. DB without changes After Images Redo DB with changes DB Administration 13 Shadow Paging Maintain two page tables during life of a transaction: current page and shadow page table. When transaction s, two pages are the same. Shadow page table is never changed thereafter and is used to restore database in event of failure. During transaction, current page table records all updates to database. When transaction completes, current page table becomes shadow page table. DB Administration 14 Concurrency Problems Deposit and withdraw funds from a bank account. Transaction Start Read account balance and store it in B; B = B + Deposit-Amount; Write B as the new balance; If ( B < Withdraw-Amount ) then Transaction rollback; else B = B - Withdraw-Amount; Write B as the new balance; Transaction commit; End. DB Administration 15 David McDonald,, GSU 11-5

6 Uncommitted Dependency Problem Assume that two transactions and both access to a same account with $200 in balance. withdraws $100. deposits $50 first, and then withdraws $500. balance = balance - $100 write balance commit balance = balance + $50 balance rollback What is the final balance? DB Administration 16 Lost Update Problem Assume that two transactions and both access to a same account with $200 in balance. reads a balance of $200 and withdraws $100. reads a $200 balance and deposits 50. balance = balance - $100 write balance balance = balance + $50 commit write balance commit What is the final balance? DB Administration 17 Inconsistent Analysis Problem Assume three accounts: Acc 1 = Balance of $40, Acc 2 = Balance of $50, Acc 3 = Balance of $30 Fetch Acc 1 (40) Sum = 40 Fetch Acc 2 (50) Sum = 90 Fetch Acc 3 (30) Fetch Acc 3 (20) Sum = 110, not 120 Update Acc > 20 Update Acc 1 (40) 40 -> 50 commit Can write back to the database and will the DB be consistent? DB Administration 18 David McDonald,, GSU 11-6

7 Concurrency Control Algorithms Serializability -- The execution of a set of transactions in some concurrent order is equivalent to the execution of those transactions in strictly sequential order. Pessimistic concurrency control Locking stamping Multiple versions Optimistic concurrency control What is A.C.I.D.? DB Administration 19 Currency Control Using Locks Two-Phase Locking Growing phase -- Acquires required locks it will ever need. Shrinking phase -- Releases acquired locks and can't request any new lock Binary Locks or Read/Write Locks Read-lock (S) a.k.a shared-lock -- Non-exclusive Write-lock (X) Exclusive Intention Locks Intent Shared (IS) Transaction intends to set S locks on individual tuples in R. Intent Exclusive (IX) Same as IS, plus transaction might update individual tuples in R and will therefore set X locks on those tuples Shared Intent Exclusive (SIX) Combines S and IX; i.e., a transaction can tolerate concurrent readers, but not concurrent updaters in R, thus the transaction might update individual tuples in R and will therefore set X locks on those tuples DB Administration 20 Granularity of Data Items Size of data items chosen as unit of protection by concurrency control protocol. Ranging from coarse to fine: The entire database. A table. An area or database space. A tuple. A attribute value of a tuple. DB Administration 21 David McDonald,, GSU 11-7

8 Granularity of Data Items Tradeoff: coarser, the lower the degree of concurrency; finer, more locking information that is needed to be stored. Best item size depends on the types of transactions. DB Administration 22 Hierarchy of Granularity Intention lock could be used to lock all ancestors of a locked node. Intention locks can be read or write. Applied top-down, released bottom-up. DB Administration 23 Example Using R/W Locks write-lock (balance) ing... balance = balance - $100 write balance commit write-lock (balance) balance = balance + $50 balance < 500 rollback (release lock) What is the final balance? DB Administration 24 David McDonald,, GSU 11-8

9 Deadlocks o Deadlocks may occur when Transactions, and, request locks in parallel: request X lock on R1 request X lock on R2 etc. DB Administration 25 request X lock on R2 request X lock on R1 etc. What Is Query Optimization? A process that translates a query into an equivalent form which is more efficient and costs less than the original one. Query optimization is responsible by the DBMS Supports associative queries Example: select from where Assume: Student.id, Student.name Student, Class Student.id = Class.id and Class.course# = '3210'; 10,000 students 100 student are taking 3210 each student takes 3 classes DB Administration 26 No Optimization Example 30,000 tuples Class 10,000 tuples Student Σ Sudent.id=Class.id Main Memory 5 Result tuples Result 1 Σ course#='3210' Read: tuples Write: tuples ID Name DB Administration 27 David McDonald,, GSU 11-9

10 Optimization Example 30,000 tuples Class 10,000 tuples Student Σ Sudent.id=Class.id Main Memory 5 (100 tuples) Σ course#='3210' 106 tuples Result 1 6 Read: 10 tuples 6 Write: 10 tuples ID Name DB Administration 28 Query Optimization Gives the DBMS more control over system performance. Two main techniques for query optimization: heuristic rules that order operations in a query; comparing different strategies based on relative costs, and selecting one that minimizes resource usage. Disk access tends to be dominant cost in query processing for centralized DBMS. DB Administration 29 Heuristical Processing Strategies Perform Projection as early as possible. Keep projection attributes on same relation together. Perform Selection operations as early as possible. Keep predicates on same relation together. Combine Cartesian product with subsequent Selection whose predicate represents join condition into a Join operation. Use associativity of binary operations to rearrange leaf nodes so leaf nodes with most restrictive Selection operations executed first. Compute common expressions once. If common expression appears more than once, and result not too large, store result and reuse it when required. Useful when querying views, as same expression is used to construct view each time. DB Administration 30 David McDonald,, GSU 11-10

11 Cost Estimation for Attribute Operations Many different ways of implementing attribute operations. Aim of Query Optimization is to choose most efficient one. Use formulae that estimate costs for a number of options, and select one with lowest cost. Consider only cost of disk access, which is usually dominant cost in Query Processing. Many estimates are based on cardinality of the relation, so need to be able to estimate this. DB Administration 31 Query Optimization (Summary) Translate query into an internal representation Relational algebra expressions Find other equivalent internal representation Several equivalent algebra expressions Determine the execution cost of each representation Determine costs based on attribute access Optimizers chooses the best solution for a query plan DB Administration 32 David McDonald,, GSU 11-11