Distributed Transactions

Question 1

What is a transaction?

Answer 1

Sequence of server operations guarenteed to be completed.

Question 2

How do transactions comply to ACID?

Answer 2

Atomicity: Transactions are either completed or aborted.
Consistency: 1 consistent state to another.
Isolation: Transactions happen without interference from other transactions.
Durability: Results of transactions stored in permanent storage.

Question 3

Why must objects be recoverable?

Answer 3

If a process crashes during a transactions, a new process can recover the objects from storage and continue the transaction.

Question 4

Why must operations be synchronised?

Answer 4

To ensure isolation

Question 5

When can transactions happen concurrently?

Answer 5

When concurrent execution has the same effect as serial execution.

Question 6

What is the lost update problem?

Answer 6

Transaction T reads the value of x. (suspends)
Transaction U reads the value of x.
Transaction U adds 4 to value of x.
Transaction U commits.
Transaction T adds 5 to old value of x.
Transaction T commits.
Changes made by U are overridden.

Question 7

What is the inconsistent retrieval problem?

Answer 7

Wrong retrieval of data because a transactipon accesses data before another has finished modifying it.
(T) a.withdraw(100):100
(U) total = a.get(): 100
(U) total += b.get(): 400
(T) b.deposit(100):400
(T) commit
(U) commit

Question 8

What is a conflicting operation?

Answer 8

A pair of operations that, when executed in different orders, produce different values.
(Read->Write != Write-> Read)

Question 9

What does serial equivalence require?

Answer 9

All conflicting operations to be executed in the same order for all processes.

Question 10

What is locking?

Answer 10

Using locks to ensure that only one transaction can access data at any one time.

Question 11

What can locking lead to?

Answer 11

Deadlock

Question 12

What is optimistic concurrency?

Answer 12

Executing both transactions and checking for conflicts before commiting. If conflicts are found, both transactions are aborted.

Question 13

What is timestamp ordering?

Answer 13

Using total ordering of transactions based on start.
If object O was last read/written by EARLIER transactions, O can be written to.
If O was last written to by EARLIER transactions, O can be read.

Question 14

How are dirty reads prevented?

Answer 14

Delay commit of U until all transactions from which U has observed a value have also committed.

Question 15

How can cascading aborts be prevented?

Answer 15

Only allow reads from commited transactions.

Question 16

How are premature writes prevented?

Answer 16

Delay writes until other transactions that have affecting values have commited.

Question 17

What is a distributed transaction?

Answer 17

Transactions that access objects managed by multiple servers.

Question 18

What is the two-phase commit protocol?

Answer 18

Protocol that allows for atomicty of distributed transactions by ensuring all processes have been completed successfully before committing changes.

Question 19

What is phase 1 of the 2PC protocol?

Answer 19

Each server votes for the transaction to be commited or aborted.
When a server votes commit, state = PREPARED

Question 20

What is phase 2 of the 2PC protocol?

Answer 20

If any server votes to abort, all must abort transaction.

If all servers vote commit, transaction can be committed.

Question 21

What are the timeout situations and solutions for distributed transactions? (3)

Answer 21

1. Server doesn’t receive joint decision: Request decision again (Coordinator may have crashed.)
2. Server not received abort/commit request: Abort
3. Coordinator not received votes from all servers: Can decide to abort, inform all servers.