1. Transactions Controlling Concurrent Behavior

2. Why Transactions? Database systems are normally being accessed by many users or processes at the same time. –Both queries and modifications. Unlike operating systems, which support interaction of processes, a DMBS needs to keep processes from troublesome interactions.

3. ACID Transactions ACID transactions are: –Atomic : Whole transaction or none is done. –Consistent : Database constraints preserved. –Isolated : It appears to the user as if only one process executes at a time. That is, even though actions of several transactions might be interleaved, the net effect is identical to executing all transactions one after another in some serial order. –Durable : Effects of a process survive a crash. Optional: weaker forms of transactions are often supported as well.

4. Transactions A process that reads or modifies the DB is called a transaction. It’s a unit of execution of database operations. Basic JDBC transaction pattern Connection conn =...; conn.setAutoCommit(false); try {... //JDBC statements } finally { conn.commit(); }

5. Interleaving For performance reasons, a DBMS has to interleave the actions of several transactions. However, the interleaving must be done carefully… Consider two transactions T1 and T2, each of which, when running alone preserves database consistency: –T1 transfers $100 from A to B, and –T2 increments both A and B by 1% (e.g. daily interest) –Consider the following interleaving. (1) T1 deducts $100 from A, (2) T2 increments both A and B by 1%, (3) T2 adds $100 to B. –What ’ s the problem? –B just lost $1.

6. COMMIT and ROLLBACK The SQL statement COMMIT causes a transaction to complete. –It’s database modifications are now permanent in the database. The SQL statement ROLLBACK also causes the transaction to end, but by aborting. –No effects on the database. Failures like division by 0 or a constraint violation can also cause rollback, even if the programmer does not request it.

7. Transactions and Schedules A transaction is a list of actions. The actions could be: –read, write, commit, abort A schedule is a list of actions from a set of transactions. E.g. T1T2 r(A) w(A) r(B) w(B) commit r(C) w(C) Commit

8. Anomalies: Reading Uncommitted Data T1T2 r(A) w(A) r(A) w(A) r(B) w(B) commit r(B) w(B) commit T1 transfers $100 from A to B, and T2 increments both A and B by 1% (e.g. daily interest) The problem is that the bank didn ’ t pay interest on the $100 that was being transferred. Of course there would be no problem if we executed T1 and after that T2, or vice versa.

9. Anomalies: Unrepeatable Reads Suppose that A is the number of copies available for a book. Transactions T1 and T2 both place an order for this book. First they check the availability of the book. Consider now the following: 1.T1 checks whether A is greater than 1. Suppose T1 sees (reads) value 1. 2.T2 also reads A and sees 1. 3.T2 decrements A to 0. 4.T2 commits. 5.T1 tries to decrement A, which is now 0, and gets an error because some integrity check doesn ’ t allow it. This situation can never arise in a serial execution of T1 and T2.

10. Anomalies: Overwriting uncommitted data Suppose that Larry and Harry are two employees, and their salaries must be kept the equal. T1 sets their salaries to $2000 and T2 sets their salaries to $1000. Now consider the following schedule: T1 T2 r(Larry) w(Larry) r(Harry) w(Harry) r(Harry) w(Harry) r(Larry) w(Larry) commit Unfortunately, Harry will be paid more than Larry.

11. Summarizing the Terminology A transaction (model) is a sequence of r and w actions on database elements. A schedule is a sequence of reads/writes actions performed by a collection of transactions. –r T (X) denotes T reads the DB element –w T (X) denotes T writes X –If transactions are T 1,…,T k, then we will simply use r i and w i, instead of r Ti and w Ti Serial Schedule = All actions for each transaction are consecutive. r1(A); w1(A); r1(B); w1(B); r2(A); w2(A); r2(B); w2(B); … Serializable Schedule: A schedule whose “effect” is equivalent to that of some serial schedule. We will introduce a sufficient condition for serializability.

12. To flip or not to flip… Suppose for DB elements X and Y, r i (X); r j (Y) is part of a schedule, and we flip the order of these operations. –r i (X); r j (Y) ≡ r j (Y); r i (X) –This holds always (even when X=Y) We can flip r i (X); w j (Y), as long as X≠Y That is, r i (X); w j (X)  w j (X); r i (X) – In the RHS, T i reads the value of X written by T j, whereas it is not so in the LHS. We can flip w i (X); w j (Y); provided X≠Y However, w i (X); w j (X)  w j (X); w i (X); –The final value of X may be different depending on which write occurs last.

13. Conflicts There is a conflict if one of two conditions hold. A read and a write of the same X, or Two writes of the same X In such cases the operations may not be swapped in order. All other events (reads/writes) may be swapped without changing the effect of the schedule (on the DB).

14. Precedence Graphs If by swapping pairs of non-conflicting actions in a schedule S, we end up in a serial schedule, then S is called a conflict- serializable schedule. S: r 1 (A); w 1 (A); r 2 (A); w 2 (A); r 1 (B); w 1 (B); r 2 (B); w 2 (B); Serializability/precedence graph for S Nodes: transactions {T 1,…,T k } Arcs: There is an arc from T i to T j if they have conflict access to the same database element X and T i is first; In written T i < S T j.

16. If there is a cycle in the graph –Then, there is no serial schedule which is conflict­equivalent to S. Each arc represents a requirement on the order of transactions in a conflict­ equivalent serial schedule. A cycle puts too many requirements on any linear order of transactions. If there is no cycle in the graph –Then any topological order of the graph suggests a conflict­equivalent schedule.

17. Why the Precedence-Graph Test Works? Idea: if the precedence graph is acyclic, then we can swap actions to form a serial schedule. Proof: By induction on n, number of transactions. Basis: n = 1. That is, S={T 1 }; then S is already serial. Induction: S={T 1,T 2,…,T n }. Given that the precedence graph is acyclic, there exists T i in S such that no T j in S is dependent on. –We swap all actions of T i to the front (of S). –(Actions of T i )(Actions of the other n-1 transactions) –The tail is a precedence graph that is the same as the original without T i, i.e. it has n-1 nodes.  By the induction hypothesis, we can reorder the actions of the other transactions to turn it into a serial schedule

18. Schedulers A scheduler takes requests from transactions for reads and writes, and decides if it is “OK” to allow them to operate on DB or defer them until it is safe to do so. Ideal: a scheduler forwards a request iff it cannot lead to inconsistency of DB –Too hard to decide this in real time. Real: it forwards a request if it cannot result in a violation of conflict­ serializability. We thus need to develop schedulers which ensure conflict- serializability.

19. Lock Actions Before reading or writing an element X, a transaction T i requests a lock on X from the scheduler. The scheduler can either grant the lock to T i or make T i wait for the lock. If granted, T i should eventually unlock (release) the lock on X. Shorthands: –l i (X) = “transaction T i requests a lock on X” –u i (X) = “T i unlocks/releases the lock on X”

20. The use of lock must be proper in 2 senses: –Consistency of Transactions: Read or write X only when hold a lock on X. –r i (X) or w i (X) must be preceded by some l i (X) with no intervening u i (X). If Ti locks X, T i must eventually unlock X. –Every l i (X) must be followed by u i (X). –Legality of Schedules: Two transactions may not have locked the same element X without one having first released the lock. –A schedule with l i (X) cannot have another l j (X) until u i (X) appears in between

21. Legal Schedule Doesn’t Mean Serializable Consistency constraint required for this example: A=B

22. Two Phase Locking There is a simple condition, which guarantees conflict-serializability: In every transaction, all lock requests (phase 1) precede all unlock requests (phase 2).

23. Why 2PL Works? Precisely: a legal schedule S of 2PL transactions is conflict­ serializable. Proof is an induction on n, the number of transactions. Remember, conflicts involve only read/write actions, not locks, but the legality of the transaction requires that the r/w's be consistent with the l/u's.

24. Why 2PL Works (Cont’d) Basis: if n=1, then S={T 1 }, and hence S is conflict-serializable. Induction: S={T 1,…,T n }. Find the first transaction, say T i, to perform an unlock action, say u i (X). We show that the r/w actions of T i can be moved to the front of the other transactions without conflict. Consider some action such as w i (Y). Can it be preceded by some conflicting action w j (Y) or r j (Y)? In such a case we cannot swap them. –If so, then u j (Y) and l i (Y) must intervene, as w j (Y)...u j (Y)...l i (Y)...w i (Y). –Since T i is the first to unlock, u i (X) appears before u j (Y). –But then l i (Y) appears after u i (X), contradicting 2PL. Conclusion: w i (Y) can slide forward in the schedule without conflict; similar argument for a r i (Y) action.

25. Shared/Exclusive Locks Problem: while simple locks + 2PL guarantee conflict­ serializability, they do not allow two readers of DB element X at the same time. But having multiple readers is not a problem for conflict­ serializability (since read actions commute).

26. Shared/Exclusive Locks (Cont’d) Solution: Two kinds of locks: 1. Shared lock sl i (X) allows T i to read, but not write X. –It prevents other transactions from writing X but not from reading X. 2. Exclusive lock xl i (X) allows T i to read and/or write X; no other transaction may read or write X.

27. Consistency of transaction conditions: –A read r i (X) must be preceded by sl i (X) or xl i (X), with no intervening u i (X). –A write w i (X) must be preceded by xl i (X), with no intervening u i (X). Legal schedules: –No two exclusive locks. If xl i (X) appears in a schedule, then there cannot be a xl j (X) until after a u i (X) appears. –No exclusive and shared locks. If xl i (X) appears, there can be no sl j (X) until after u i (X). If sl i (X) appears, there can be no wl j (X) until after u i (X). 2PL condition: –No transaction may have a sl(X) or xl(X) after a u(Y).

28. Scheduler Rules When there is more than one kind of lock, the scheduler needs a rule that says “if there is already a lock of type A on DB element X, can I grant a lock of type B on X?” The compatibility matrix answers the question. Compatibility Matrix for Shared/Exclusive Locks is:

29. Upgrading Locks Instead of taking an exclusive lock immediately, a transaction can take a shared lock on X, read X, and then upgrade the lock to exclusive so that it can write X. Upgrading Locks allows more concurrent operation: Had T1 asked for an exclusive lock on B before reading B, the request would have been denied, because T2 already has a shared lock on B.

30. Possibility for Deadlocks Problem: when we allow upgrades, it is easy to get into a deadlock situation. Example:T1 and T2 each reads X and later writes X.

31. Solution: Update Locks Update lock ul i (X). –Only an update lock (not shared lock) can be upgraded to exclusive lock (if there are no shared locks anymore). –A transaction that will read and later on write some element A, asks initially for an update lock on A, and then asks for an exclusive lock on A. Such transaction doesn’t ask for a shared lock on A. Legal schedules: –read action permitted when there is either a shared or update lock. –An update lock can be granted while there is a shared lock, but the scheduler will not grant a shared lock when there is an update lock. –2PL condition: No transaction may have an sl(X), ul(X) or xl(X) after a u(Y).

32. Example T1T2T3 sl(A); r(A) ul(A); r(A) sl(A) Denied xl(A) Denied u(A) xl(A); w(A) u(A) sl(A); r(A) u(A)

33. (No) Deadlock Example T 1 and T 2 each read X and later write X. Deadlock when using sl and xl locks only. Fine when using update locks.

34. What should we lock? The whole table or just single rows? What ’ s database object? What should be the locking granularity? SELECT min(year) FROM Movies; What happens if we insert a new tuple with the smallest year? Phantom problem: A transaction retrieves a collection of tuples twice and sees different results, even though it doesn ’ t modify those tuples itself.

35. Transaction support in SQL SET TRANSACTION ISOLATION LEVEL X –Where X can be SERIALIZABLE (Default) REPEATABLE READ READ COMMITED READ UNCOMMITED With a scheduler based on locks: A SERIALIZABLE transaction obtains locks before reading and writing objects, including locks on sets (e.g. table) of objects that it requires to be unchangeable and holds them until the end, according to 2PL. A REPEATABLE READ transaction sets the same locks as a SERIALIZABLE transaction, except that it doesn ’ t lock sets of objects, but only individual objects.

36. Transaction support in SQL A READ COMMITED transaction T obtains exclusive locks before writing objects and keeps them until the end. However, it obtains shared locks before reading values and then immediately releases them; their effect is to ensure that the transaction that last modified the values is complete. Thus, 1.T reads only the changes made by committed transactions. 2.No value written by T is changed by any other transaction until T is completed. 3.However, a value read by T may well be modified by another transaction (which eventually commits) while T is still in progress. 4.T is also exposed to the phantom problem. A READ UNCOMMITED transaction doesn ’ t obtain any shared lock at all. So, it can read data that is being modified. Such transactions are allowed to be READ ONLY only. So, such transaction doesn ’ t ask for any lock at all.

37. In Summary LevelReading Uncommited Data (Dirty Read) Unrepeatable Read Phantom READ UNCOMMITED Maybe READ COMMITTED NoMaybe REPEATABLE READ No Maybe SERIALIZABLENo