Chapter 6 Distributed Concurrency Control

Transcription

1 Chapter 6 Distributed Concurrency Control Table of Contents Serializability Theory Taxonomy of Concurrency Control Algorithms Locking-Based Concurrency Control Timestamp-Based Concurrency Control Optimistic Concurrency Control Deadlock Management Chapter6-1 1

2 1. Serializability Theory Serial and Serializable Schedule Schedule: A time-ordered sequence of the important actions taken by one or more transactions. Serial Schedule: No interleaving of actions or transactions Serializable Schedule: If its effect on the database is the same as that of some serial schedule. Chapter6-2 Example: Two Transactions T1 READ(A, t) t := t WRITE(A, t) READ(B, t) t := t WRITE(B, t) READ(A, s) s := s * 2 WRITE(A, s) READ(B, s) s := s * 2 WRITE(B, s) Chapter6-3 2

3 Example: Serial Schedule(1) T1 A B READ(A, t) t := t WRITE(A, t) READ(B, t) t := t WRITE(B, t) READ(A, s) s := s * 2 WRITE(A, s) READ(B, s) s := s * 2 WRITE(B, s) Chapter6-4 Example: Serial Schedule(2) T1 A B READ(A, t) t := t WRITE(A, t) READ(B, t) t := t WRITE(B, t) READ(A, s) s := s * 2 WRITE(A, s) READ(B, s) s := s * 2 WRITE(B, s) Chapter6-5 3

4 Example: Serializable Schedule T1 A B READ(A, t) t := t WRITE(A, t) READ(B, t) t := t WRITE(B, t) READ(A, s) s := s * 2 WRITE(A, s) READ(B, s) s := s * 2 WRITE(B, s) Chapter6-6 Example: Non-serializable Schedule T1 A B READ(A, t) t := t WRITE(A, t) READ(B, t) t := t WRITE(B, t) READ(A, s) s := s * 2 WRITE(A, s) READ(B, s) s := s * 2 WRITE(B, s) Chapter6-7 4

5 A Notation for Transactions and Schedules Actions r i (X): A transaction T i reads DB element X. w i (X): A transaction T i reads DB element X. Transaction T i A sequence of actions with subscript i. Schedule S of a set of transactions T A sequence of actions, in which for each transaction T i in T, the actions of T i appear in S in the same order that they appear in the definition of T i itself. : r 1 (A); w 1 (A); r 2 (A); w 2 (A); r 1 (B); w 1 (B); r 2 (B); w 2 (B) Chapter6-8 Conflict Serializability (1) Conflicts : Action, r i (X); r j (Y) r i (X); w j (Y) w i (X); r j (Y) w i (X); w j (Y) Not Conflict, even if X = Y NotConflictifX Y NotConflictifX Y NotConflictifX Y Conclusion A read and a write of the same X, or two writes of X are conflict and may not be swapped in order. All other events may be swapped without possibility that the effect of the schedule will change. Chapter6-9 5

6 Conflict Serializability (2) Schedule S S conflict equivalent S non-conflicting operation swap S Concurrent schedule S conflict serializable S conflict equivalent serial schedule Example r 1 (A); w 1 (A); r 2 (A); w 2 (A); r 1 (B); w 1 (B); r 2 (B); w 2 (B) r 1 (A); w 1 (A); r 2 (A); r 1 (B); w 2 (A); w 1 (B); r 2 (B); w 2 (B) r 1 (A); w 1 (A); r 1 (B); r 2 (A); w 2 (A); w 1 (B); r 2 (B); w 2 (B) r 1 (A); w 1 (A); r 1 (B); r 2 (A); w 1 (B); w 2 (A); r 2 (B); w 2 (B) r 1 (A); w 1 (A); r 1 (B); w 1 (B); r 2 (A); w 2 (A); r 2 (B); w 2 (B) Chapter6-10 Precedence Graphs Nodes: transactions in S Arcs: T i T j whenever p i (A), q j (A) are actions in S p i (A) < S q j (A) at least one of p i,q j is a write (p i,q j : conflicting) Test for Conflict Serializability Schedule S ó precedence graph cycle S: Conflict Serializable Chapter6-11 6

7 Examples r 1 (A); w 1 (A); r 2 (A); w 2 (A); r 1 (B); w 1 (B); r 2 (B); w 2 (B) T 1 T 2 r 1 (A); r 2 (B); w 1 (A); r 2 (A); w 2 (A); w 2 (B); r 1 (B); w 1 (B) T 1 T 2 Chapter Taxonomy of Concurrency Control Algorithms Concurrency Control ú þ (Serializability) Concurrency Control Lost Update Dirty Read Unrepeatable Read Chapter6-13 7

8 Lost Update read(a) T1 A = A + 100; write(a) read(a) A = A + 200; write(a) Chapter6-14 Dirty Read read(a) A = A + 100; write(a) rollback; T1 read(a) display A; commit; Chapter6-15 8

9 Unrepeatable Read T1 read(a) A=A+100 write(a) read(a) read(a) Chapter6-16 Concurrency Control Concurrency Control Algorithms Pessimistic Optimistic Locking Timestamp Ordering Hybrid Locking Timestamp Ordering Centralized Basic Primary Copy Multiversion Distributed Conservative Chapter6-17 9

10 3. Locking Lock? ó Lock Lock Modes Shared (S) Lock: Exclusive (X) Lock: S X S OK Not OK X Not OK Not OK Chapter6-18 Locking Rule Lock Lock (well-formed) Conflict mode Lock Lock ÿ ó Deadlock ó Starvation Lock (Unlock), Unlock Lost Update, Dirty Read, Unrepeatable Read (two-phase rule) Chapter

11 TwoPhaseLocking(2PL) Two Phase Locking Protocol : Unlock Lock Growing Phase: Lock ñ Shrinking Phase: Unlock ñ Conservative 2PL End of Transaction Lock Lock Growing Phase Shrinking Phase Chapter6-20 Distributed Locking Centralized 2PL Primary Copy 2PL Distributed 2PL Chapter

12 3.1 Centralized 2PL Allow one site to maintain all locks. Only 3 messages per granted lock/unlock. request lock grantlock unlock + extra if cannot grant Bottleneck at central site? What if the central site crashes? Primary Copy 2PL One site holds lock for each replicated element. Bottleneck can be reduced because locking responsibility is shared. Chapter Distributed 2PL ñ þ Primary 2PL ÿ Read-One Write-All (ROWA) Majority Consensus (Voting 2PL) Quorum Consensus Lazy Update Propagation, Distributed Deadlock Chapter

13 4. Timestamp Ordering Timestamps Unique value representing a time. Example: clock time, serial counter Even distributed systems can have timestamps high-order bits: clock time of local machine low-order bits: ID for that machine Chapter6-24 Serializability vs. Timestamps ä Locking overhead ñ: - Data Structures TS(T): T òþ DB element X ó RT(X): highest timestamp of a transaction to read X WT(X): highest timestamp of a transaction to write X C(X): a bit indicating whether the most recent writer of X has committed. Dirty read Chapter

14 Physically Unrealizable Behaviors(1) Read Too Late Transaction T tries to read X, but TS(T) < WT(X). T would read something that was written after T ostensibly finished. U writes X T reads X T start U start Chapter6-26 Physically Unrealizable Behaviors(2) Write Too Late Transaction T tries to write X, but RT(X) > TS(T) > WT(X). Some other transaction read a value written earlier than T, when it should have read what was written T. U reads X T writes X T start U start Chapter

15 Problems with Dirty Data T tries to read X: TS(T) > WT(X), but C(X) = false. T would be reading dirty data. U writes X T reads X U start T start U abort Solutions Wait if data might be dirty (Conservative TO) Cascading rollback using an undo log (Basic TO) Chapter6-28 Rules for Timestamp-Based Scheduling Suppose the scheduler receives a request r T (X). TS(T) WT(X): The read is physically realizable. C(X) = true: Grant, RT(X) = TS(T) if TS(T) > RT(X). C(X) = false: Delay T until C(X) becomes true. TS(T) < WT(X): Rollback T. Suppose the scheduler receives a request w T (X). TS(T) WT(X) and TS(T) RT(X): physically realizable. Write new value for X; WT(X) = TS(T), C(X) = false TS(T) RT(X) and TS(T) < WT(X): physically realizable. C(X) = true: Ignore the write. C(X) = false: Delay T until C(X) becomes true. TS(T) < RT(X): Rollback T. Chapter

16 Example T1 T3 A B C 200 r 1 (B) w 1 (B) w 1 (A) 150 r 2 (A) w 2 (C) Abort; 175 r 3 (C) w 3 (A) RT=0 WT=0 RT=150 WT=200 RT=0 WT=0 RT=200 WT=200 RT=0 WT=0 RT=175 Chapter6-30 Distributed TO Schedulers Distributed Timestamp: (local timestamp, site ID) Global clock synchronization using Lamport s clock. ñ TO Rule scheduling. Chapter

17 5. Optimistic Concurrency Control Transactions have 3 phases: Read: all DB values read writes to temporary storage no locking Validate: check if schedule so far is serializable Write: if validate ok, write to DB Key Idea Make validation atomic. If T1,, T3, is validation order, then resulting schedule will be conflict equivalent to Ss =T1 T3. Chapter6-32 Implementation To implement validation, system keeps three sets: START Set of transactions that have started, but not yet completed validation. START(T): the time at which a transaction T started VAL Set of transactions that have successfully finished phase 2 (validation) VAL(T): the time at which a transaction T validated FIN transactions that have finished phase 3 (and are all done) FIN(T): the time at which a transaction T finished Chapter

18 Example: What validation must prevent? RS()={B} WS()={B,D} RS(T3)={A,B} φ WS(T3)={C} T3 T3 start start validated validated time Chapter6-34 Example: What validation must allow? RS()={B} WS()={B,D} RS(T3)={A,B} φ WS(T3)={C} T3 T3 start start validated validated time finish phase 3 T3 start Chapter

19 Another thing validation must prevent: RS()={A} WS()={D,E} RS(T3)={A,B} WS(T3)={C,D} validated T3 validated BAD: w3(d) w2(d) finish time? Chapter6-36 Validation Rules (1) When Tj starts phase 1: ignore(tj) FIN (2) At Tj Validation: if check (Tj) then [ VAL VAL U {Tj}; do write phase; FIN FIN U {Tj} ] Chapter

20 Check (Tj): For Ti VAL - IGNORE (Tj) DO IF [ WS(Ti) RS(Tj) or Ti FIN ] THEN RETURN false; RETURN true; Is this check too restrictive? Chapter6-38 Improving Check(Tj) For Ti VAL - IGNORE (Tj) DO IF [ WS(Ti) RS(Tj) or (Ti FIN AND WS(Ti) WS(Tj) )] THEN RETURN false; RETURN true; Chapter

21 Example start validate finish U: RS(U)={B} WS(U)={D} W: RS(W)={A,D} WS(W)={A,C} T: RS(T)={A,B} WS(T)={A,C} V: RS(V)={B} WS(V)={D,E} Chapter6-40 Distributed Optimistic CC T ij sub-transactionoft i executed at site j Local Validation: Tij òÿ Validation HB set HB(Tij): ID of global transaction which precedes Tij Global Validation: local validation HB(Tij) ÿ global transaction commitþ abortÿ ó. Deadlock. (timeout ) Write Phase: 2PC Global validation, coordinator ready message. sub-transaction ready, commit. Chapter

22 6. Deadlock Management Deadlock Scenario - Example T1 Lock-X(B) read(b) B=B 50; write(b) Lock-X(A) Lock-S(A) read(a) Lock-S(B) Chapter6-42 Deadlock Deadlock Deadlock Prevention Deadlock Detection and Resolution Deadlock Prevention : Deadlock lock. ÿ.. ó lock Preemption: wait-die, wound-wait Chapter

23 Wait-Die, Wound-Wait Wait-Die Wound-Wait T old T new lock A request A T old T new lock A request A T new : Rollback T new :Wait T old T new lock A request A T old T new lock A request A T new :Wait T new : Rollback Chapter6-44 Deadlock Detection Wait-For Graph: G = (V, E) V: Set of transactions E: Set of T w T h, where T w is waiting for T h Deadlock Detection and Resolution Deadlock : WFG cycle Resolution Selection of a victim transaction Victim transaction rollback Starvation Chapter

24 Handling Distributed Deadlocks Distributed WFG Distributed WFG = Integration of Local WFG ñ Local WFG (Local Deadlock) ñ : ó, : DD1 S1 S2, DD2 S3, S4, S5 Affinity-based Clustering Chapter6-46 Distributed Deadlock Detection ó : S1(T1 ), S2 ( T1) T1 S2,or T1 S1 False deadlock ñ. ó ò WFG edge probe message Chapter