Distributed Transactions What is a transaction A sequence

Distributed Transactions What is a transaction? (A sequence of server operations that must be carried out atomically) ACID properties - what are these (Atomicity, Consistency, Isolation, Durability) What is a distributed transaction? -Involves objects managed by communicating with one another. multiple servers

Transactions Permanent Record Commit / Abort Shared variables Server operation

Concurrency control The goal of concurrency control is to guarantee that when multiple transactions are concurrently executed, the net effect should be equivalent to executing them in some serial order. This is the essence of the serializability property.

Example 1 T 1 starts (20) W(x: =1) [OK] T 2 starts(30) T 3 starts (40) R(x) [OK] W(x: =2) [OK] T 1 commits T 2 commits W(x: =3) [OK] R(x) T 3 commits This is serializable. Think of other examples too.

Example 2 T 1 starts (20) W(x: =1) [OK] T 2 starts(30) T 3 starts (40) This is not serializable. R(x) [NO] W(x: =2) [OK] T 1 aborts R(x) T 2 commits? W(x: =3) [OK] T 3 commits

Pitfalls in concurrency control • Dirty read • Lost update • Premature write

Lost update Initially, B= $1000 Amy’s transaction 1 Load B into local 4 2 Add $250 to local 3 Store local to B Bob’s transaction Load B into local 5 Add $250 to local 6 Store local to B What if the interleaving is 1 4 2 5 3 6 ? The final value of B is $1250, although it should have been $1500

Dirty read Initially B= $1000 Amy’s transaction 1 Load B into local 4 2 Add $250 to local 3 Store local to B ABORT Bob’s transaction Load B into local 5 Add $250 to local 6 Store local to B COMMIT Execute the actions in the sequence 1 2 3 4 5 6. If Amy’s transaction aborts after Bob executes step 4, then the final result is still $1500, although it should have been $1250

Premature write {Initially B = 0} Amy’s transaction 1 B: = $500 Bob’s transaction 2 B : = $1000 3 4 COMMIT ABORT B changes to 0. This could have been avoided if the second transaction postponed its commit UNTIL the first transaction commits or aborts.

Locks are commonly used to implement serrializability of concurrent transactions. Operations on shared objects are in conflict when one of them is a write operation. Each transaction must acquire the corresponding exclusive lock before executing an action. Locks can be fine grained. Note that there is no conflict between two reads.

Serializability The serialization graph is a directed graph (V, E) where V is the set of transactions, and E is the set of directed edges between transactions - a directed edge from a transaction Tj to a transaction Tk implies that Tk applied a lock only after Tj released the corresponding lock. Tj Tk

Serializability theorem For a set of concurrent transaction, the serializability property holds if and only if the corresponding serialization graph is acyclic [Proved by Bernstein, Goodman, Hadzilacos in 1987]

Two-phase locking (2 PL) Phase 1. Acquire all locks needed to execute the transaction. The locks will be acquired one after another, and this phase is called the growing phase or acquisition phase Phase 2. Release all locks acquired so far. This is called the shrinking phase or the release phase.

Two-phase locking (2 PL) acquire Growing phase release Shrinking phase

2 PL Theorem. 2 PL guarantees serializability. Proof. Suppose that theorem is not correct. Then the serialization graph must contain a cycle …Tj Tk … Tm Tj …This implies that Tj must have released a lock (that was later acquired by Tk) and then acquired a lock (that was released by Tm). However this violates the condition of two-phase locking that rules out any locking once a lock has been released.

Atomic Commit Protocols S 1 Servers may crash Network of servers S 2 S 3 The initiator of a transaction is called the coordinator, and the remianing servers are participants

Requirements of Atomic Commit Protocols Network of servers S 2 S 1 Servers may crash S 3 Termination. All non-faulty servers must eventually reach an irrevocable decision. Agreement. If any server decides to commit, then every server must have voted to commit (i. e. no one voted abort). Validity. If all servers vote commit and there is no failure, then all servers must commit (as opposed to all deciding to abort) Irreversibility. Each participant decides at most once

One-phase Commit t mi m o C client server participant coordinator server participant If a participant deadlocks or faces a local problem then the coordinator may never be able to find it. Too simplistic.

Two-phase commit (2 PC) Phase 1: The coordinator sends VOTE to the participants. and receive yes / no from them. Phase 2: if ∀server j: vote(j) = yes multicast COMMIT to all severs

Failure scenarios in 2 PC (Phase 1) Fault: Coordinator did not receive YES / NO: OR Participant did not receive VOTE: Solution: Broadcast ABORT; Abort local transactions

Failure scenarios in 2 PC (Phase 2) (Fault) A participant does not receive a COMMIT or an ABORT message from the coordinator (it may be the case that the coordinator crashed after sending ABORT or COMIT to a fraction of the servers). The participant remains undecided, until the coordinator is repaired and reinstalled into the system. This blocking is a known weakness of 2 PC.

Coping with blocking in 2 PC A non-faulty participant can ask other participants about what message (COMMIT or ABORT) did they receive from the coordinator, and take appropriate actions. But what if no non-faulty participant* received anything? Who knows if the coordinator committed or

Non-blocking Atomic Commit A blocking protocol has the potential to prevent non-faulty participants from reaching a final decision. A solution to the atomic commitment problem is called non-blocking, if in spite of server crashes, every non-faulty participant eventually decides. One solution is to impose the requirement of uniform agreement

Uniform agreement If any participant (faulty or not) delivers a message m (commit or abort) then all correct processes eventually deliver m. To implement uniform agreement, no server should deliver a COMMIT or ABORT message until it has relayed it to all other servers. If a process times out in phase 2, then it decides

Recovery: Stable storage Creates the illusion of an incorruptible storage, even if a writer or a disk crashes at any time. The implementation Uses at least two independent disks. A 0 update A 1 inspect

Stable storage To write, do the following: 1. copy on disk A 0; 2. record timestamp T 0; 3. compute checksum S 0; 4. copy on disk A 1; 5. record timestamp T 1; 6. compute checksum S 1 A 0 update A 1 Readers check four cases: Both checksums OK and T 1>T 0 {accept any} Both checksums OK and T 1<T 0 {accept A 0} between 3 & 4} Checksum on A 1 wrong {accept A 0} Checksum on A 0 wrong {accept A 0} (Which copy to accept in each case? ) inspect {failure

Checkpointing • Mechanism for (backward) error recovery. Transaction states are periodically stored on stable storages. Following a failure, the transaction rolls back to the nearest checkpoint. • Independent (unsynchronized) or coordinated (synchronized) checkpointing

Classification of checkpointing Coordinated Checkpointing takes a consistent snapshot. Has some overhead. Uncoordinated checkpointing apparently has no overhead. But it may have some efficiency problems.