Concurrency Control II More on Two Phase Locking Time Stamp Ordering Validation Scheme 1 Learning Objectives Variations of two phase locking Dealing with Deadlock and Starvation Time Stamp Ordering Technique Validation Database Implementation – Concurrency Control Yan Huang 2 Schedules Interleaved (or non-interleaved) actions from several transactions A schedule is good if it can be transformed into one of the serial schedules by switching two consecutive nonconflict actions We’ve learned 2PL to achieve serializability It is a pessimistic scheme Database Implementation – Concurrency Control Yan Huang 3 Recall 2PL (two phase locking) Locks of Ti growing shrinking time Each transaction has to follow, no locking anymore after the first unlocking Database Implementation – Concurrency Control Yan Huang 4 Why 2PL serializability? Acyclic precedence graph = serializability Basic idea: No cycle in precedence graph Because if there is a arc from Ti to Tj in precedence graph, the first unlocking of Ti precede the first unlocking of Tj (proof details in ccI notes) If there is circle, you will have Ti < Ti Database Implementation – Concurrency Control Yan Huang 5 Who will follow 2PL in practice? Looks like it is DB application developers’ job. But, they can not be trusted and too much work Checking conformity of every transaction is costly In practice, CC subsystems of DBMS take over the responsibility Database Implementation – Concurrency Control Yan Huang 6 Variations of 2PL Basic 2PL Conservative 2PL Strict 2PL Rigorous 2PL Database Implementation – Concurrency Control Yan Huang 7 Basic 2PL 2PL with binary locks Covered in last class Database Implementation – Concurrency Control Yan Huang 8 Shared locks So far: S = ...l1(A) r1(A) u1(A) … l2(A) r2(A) u2(A) … Do not conflict Instead: S=... ls1(A) r1(A) ls2(A) r2(A) …. us1(A) us2(A) Database Implementation – Concurrency Control Yan Huang 9 Lock actions l-ti(A): lock A in t mode (t is S or X) u-ti(A): unlock t mode (t is S or X) Shorthand: ui(A): unlock whatever modes Ti has locked A Database Implementation – Concurrency Control Yan Huang 10 Well formed transactions Ti =... l-S1(A) … r1(A) … u1 (A) … Ti =... l-X1(A) … w1(A) … u1 (A) … Database Implementation – Concurrency Control Yan Huang 11 What about transactions that read and write same object? Option 1: Request exclusive lock Ti = ...l-X1(A) … r1(A) ... w1(A) ... u1(A) … Database Implementation – Concurrency Control Yan Huang 12 • What about transactions that read and write same object? Option 2: Upgrade (E.g., need to read, but don’t know if will write…) Ti=... l-S1(A) … r1(A) ... l-X1(A) …w1(A) ...u1(A)… Think of - Get 2nd lock on A, or - Drop S, get X lock Database Implementation – Concurrency Control Yan Huang 13 Compatibility matrix Comp S S true X false X false false Database Implementation – Concurrency Control Yan Huang 14 Schedule T1 l-s1(A);Read(A) A A+100;Write(A) l-x1(B); u1(A) T2 l-s2(A);Read(A) A Ax2;Write(A); l-x2(B) Read(B);B B+100 Write(B); u1(B) delayed l-x2(B); u2(A);Read(B) B Bx2;Write(B);u2(B); Database Implementation – Concurrency Control Yan Huang 15 Conservative 2PL Lock all items it needs then transaction starts execution If any locks can not be obtained, then do not lock anything Difficult but deadlock free first action starts locks growing shrinking time Database Implementation – Concurrency Control Yan Huang 16 Strict 2PL T does not release any write locks until it commits or aborts Good for recoverability Since reads or writes on what T writes Deadlock free? locks growing time shrinking Database Implementation – Concurrency Control Yan Huang First write unlock 17 T commits or aborts Rigorous 2PL T does not release any locks until it commits or aborts Easy to implement Deadlock free? T commits or aborts locks growing shrinking time Database Implementation – Concurrency Control Yan Huang 18 2PL Does basic 2PL guarantee serializability? Does conservative 2PL guarantee serializability? Does strict 2PL guarantee serializability? Does rigorous 2PL guarantee serializability? Database Implementation – Concurrency Control Yan Huang 19 Compare variations of 2PL Deadlock Only conservative 2PLis deadlock free Q: give a schedule of two transactions following 2PL but result in deadlock. Database Implementation – Concurrency Control Yan Huang 20 Exercises: S1: r1(y)r1(x)w1(x)w2(x)w2(y) S2: r1(y)r3(x)w1(x)w3(x)w2(y)w2(x) S3: r3(y)w1(x)w3(x)r1(z)w2(y)w2(x) Assuming binary lock right before read or write; and rigorous 2PL (release all locks right after last operation), are S1, S2,and S3 possible? Database Implementation – Concurrency Control Yan Huang 21 Deadlocks Detection Wait-for graph Prevention Resource ordering Timeout Wait-die Wound-wait Database Implementation – Concurrency Control Yan Huang 22 Deadlock Detection Build Wait-For graph Use lock table structures Build incrementally or periodically When cycle found, rollback victim T2 T1 T4 T3 T5 T6 Database Implementation – Concurrency Control Yan Huang T7 23 Resource Ordering Order all elements A1, A2, …, An A transaction T can lock Ai after Aj only if i > j Problem : Ordered lock requests not realistic in most cases Database Implementation – Concurrency Control Yan Huang 24 Timeout If transaction waits more than L sec., roll it back! Simple scheme Hard to select L Database Implementation – Concurrency Control Yan Huang 25 Wait-die Transactions are given a timestamp when they arrive …. ts(Ti) Ti can only wait for Tj if ts(Ti)< ts(Tj) ...else die Database Implementation – Concurrency Control Yan Huang 26 Example: T1 wait (ts =10) T2 (ts =20) wait? wait T3 (ts =25) Very high level: only older ones have the privilege to wait, younger ones die if they attempt to wait for older ones Database Implementation – Concurrency Control Yan Huang 27 Wound-wait Transactions are given a timestamp when they arrive … ts(Ti) Ti wounds Tj if ts(Ti)< ts(Tj) else Ti waits “Wound”: Tj rolls back and gives lock to Ti Database Implementation – Concurrency Control Yan Huang 28 Example: T1 wait (ts =25) T2 (ts =20) wait wait T3 (ts =10) Very high level: younger ones wait; older ones kill (wound) younger ones who hold needed locks Database Implementation – Concurrency Control Yan Huang 29 Who die? Looks like it is always the younger ones either die automatically or killed What is the reason? Will the younger ones starve? Suggestions? Database Implementation – Concurrency Control Yan Huang 30 Timestamp Ordering Key idea: Transactions access variables according to an order decided by their time stamps when they enter the system No cycles are possible in the precedence graph Database Implementation – Concurrency Control Yan Huang 31 Timestamp System time when transactions starts An increasing unique number given to each stransaction Denoted by ts(Ti) Database Implementation – Concurrency Control Yan Huang 32 The way it works Two time stamps associated with each variable x RS(x): the largest time stamp of the transactions read it WS(x): the largest time stamp of the transactions write it Protocol: ri(x) is allowed if ts(Ti) >= WS(x) wi(x) is allowed if ts(Ti) >=WS(x) and ts(Ti) >=RS(x) Disallowed ri(x) or wi(x) will kill Ti, Ti will restart Database Implementation – Concurrency Control Yan Huang 33 x Example y z RS=-1 RS=-1 RS=-1 WS=-1 WS=-1 WS=-1 Assuming: ts(T1) = 100, ts(T2) = 200, ts(T3) = 300 T1 T2 T3 R(x); W(y); R (y); W(z); R(x); W(z); R(y); W(x); Database Implementation – Concurrency Control Yan Huang 34 x Example y z RS=100 RS=-1 RS=-1 WS=-1 WS=-1 WS=-1 Assuming: ts(T1) = 100, ts(T2) = 200, ts(T3) = 300 T1 T2 T3 R(x); Database Implementation – Concurrency Control Yan Huang 35 x Example y z RS=100 RS=-1 RS=-1 WS=-1 WS=-1 WS=100 Assuming: ts(T1) = 100, ts(T2) = 200, ts(T3) = 300 T1 T2 T3 R(x); W(y); Database Implementation – Concurrency Control Yan Huang 36 x Example y z RS=100 RS=200 RS=-1 WS=-1 WS=-1 WS=100 Assuming: ts(T1) = 100, ts(T2) = 200, ts(T3) = 300 T1 T2 T3 R(x); W(y); R (y); Database Implementation – Concurrency Control Yan Huang 37 x Example y z RS=100 RS=200 RS=-1 WS=-1 WS=300 WS=100 Assuming: ts(T1) = 100, ts(T2) = 200, ts(T3) = 300 T1 T2 T3 R(x); W(y); R (y); W(z); Database Implementation – Concurrency Control Yan Huang 38 x Example y z RS=200 RS=200 RS=-1 WS=-1 WS=300 WS=100 Assuming: ts(T1) = 100, ts(T2) = 200, ts(T3) = 300 T1 T2 T3 R(x); W(y); R (y); W(z); R(x); Database Implementation – Concurrency Control Yan Huang 39 x Example y RS=200 RS=200 WS=-1 z RS=-1 WS=100 WS=300 Assuming: ts(T1) = 100, ts(T2) = 200, ts(T3) = 300 T1 T2 T3 R(x); W(y); R (y); W(z); R(x); W(z); T1 is rolled back Database Implementation – Concurrency Control Yan Huang 40 x Example y RS=200 RS=200 WS=-1 z RS=-1 WS=100 WS=300 Assuming: ts(T1) = 100, ts(T2) = 200, ts(T3) = 300 T1 T2 T3 R(x); W(y); R (y); W(z); R(x); W(z); What happen to RS and WS? T1 is rolled back Database Implementation – Concurrency Control Yan Huang 41 Net result of TO scheduling Conflict pairs of actions are taken in the order of their home transactions But the basic TO does not guarantee recoverability later Database Implementation – Concurrency Control Yan Huang 42 Validation An optimistic scheme Transactions have 3 phases: (1) Read all DB values read writes to temporary storage no locking (2) Validate check if schedule so far is serializable (3) Write if validate ok, write to DB Database Implementation – Concurrency Control Yan Huang 43 Key idea Make validation atomic If T1, T2, T3, … is validation order, then resulting schedule will be conflict equivalent to Ss = T1 T2 T3... Database Implementation – Concurrency Control Yan Huang 44 To implement validation, system keeps two sets: FIN = transactions that have finished phase 3 (and are all done) VAL = transactions that have successfully finished phase 2 (validation) Database Implementation – Concurrency Control Yan Huang 45 Example of what validation must prevent: = RS(T2)={B} RS(T3)={A,B} WS(T2)={B,D} WS(T3)={C} T2 start T3 start T2 T3 validated validated time Database Implementation – Concurrency Control Yan Huang 46 allow Example of what validation must prevent: RS(T2)={B} WS(T2)={B,D} T2 start T3 start RS(T3)={A,B} = WS(T3)={C} T2 T3 validated validated T2 finish phase 3 T3 start Database Implementation – Concurrency Control Yan Huang time 47 Another thing validation must prevent: RS(T2)={A} WS(T2)={D,E} T2 validated RS(T3)={A,B} WS(T3)={C,D} T3 validated finish BAD: w3(D) w2(D) Database Implementation – Concurrency Control Yan Huang T2 time 48 allow Another thing validation must prevent: RS(T2)={A} WS(T2)={D,E} T2 RS(T3)={A,B} WS(T3)={C,D} T3 validated validated finish T2 Database Implementation – Concurrency Control Yan Huang finish T2 time 49 Validation Rule When start validating T Check RS(T) WS(U) is empty for U that started but did not finish validation before T started Check WS(T) WS(U) is empty for any U that started but did not finish validation T start validation Database Implementation – Concurrency Control Yan Huang 50 start validate finish Exercise: U: RS(U)={B} WS(U)={D} T: RS(T)={A,B} WS(T)={A,C} W: RS(W)={A,D} WS(W)={A,C} V: RS(V)={B} WS(V)={D,E} Database Implementation – Concurrency Control Yan Huang 51