En anglais

Replica agreement during a network split

Computing · Networks & Security · ages 20-22

Name ______________________ Date ____________

In an eventually consistent store, updates stop and communication and reconciliation succeed. What is promised?
- The replicas eventually converge to an agreed state
- The replicas retain their conflicting states forever
- The replicas erase every update accepted during the split
A write completes before a read begins. No later write occurs. Under linearizability, what must the read return?
- The earlier value stored before the completed write
- The value supplied by the completed write
- Any value chosen from the replica's stored history
Despite network partitions that may persist indefinitely, a store can guarantee linearizability and successful eventual responses to every request received by a non-failing node.
Circle one: True False
An available basket accepts additions on both sides of a split. Its policy merges item sets by union. R adds tea and S adds rice. What follows after reconciliation?
- The basket must discard tea to restore one-copy behaviour
- The basket contains tea and rice under the stated policy
- The basket must discard rice because it was added at S
Isolated replicas hold 2. R completes a write of 8 before S receives a read. No later write occurs. Which response preserves linearizability?
- S returns 2 because its local copy has not changed
- S returns 2 and repairs the answer after reconnection
- S waits or refuses rather than returning its stale 2
A basket tolerates stale views under a merge policy. A debit must never overdraw an account. How should the team choose guarantees?
- Match each feature's operations to the rules it must preserve
- Apply the basket's merge policy to every account debit too
- Let every replica debit freely and repair overdrafts afterward
A team promises that eventual consistency alone makes every replica agree within one second of reconnection. What is missing?
- A rule requiring replicas to reject every read after reconnection
- A separate bound on communication and reconciliation time
- A rule requiring clients to stop using all replica addresses
Two buyers each read a seat count of 1 before either writes 0. Every separate read and write is linearizable. What prevents both buyers reserving the seat?
- Repeat the separate reads until both buyers see the value 1
- Merge the two writes of 0 after both reservations finish
- Make checking and reserving the seat one atomic operation

Answer key

For grown-ups. Fold this page away before handing over the rest.

Replica agreement during a network split W1-mt_j6qtpYPHsr-s1

The replicas eventually converge to an agreed state · Eventual consistency promises convergence under these conditions, not agreement at every moment.
The value supplied by the completed write · Real-time order places the completed write before the later read.
False · These are CAP's consistency and availability guarantees. A store cannot guarantee both despite partitions.
The basket contains tea and rice under the stated policy · Union retains both additions. This result depends on the stated merge policy, not on eventual consistency alone.
S waits or refuses rather than returning its stale 2 · The later read must reflect the completed write of 8. S cannot return its stale value while claiming linearizability.
Match each feature's operations to the rules it must preserve · The basket permits stale views. The debit needs a design that prevents overspending instead of repairing it later.
A separate bound on communication and reconciliation time · Eventual consistency supplies no fixed convergence deadline. A one-second promise needs additional timing guarantees.
Make checking and reserving the seat one atomic operation · Linearizable individual operations still permit both reads before either write. An atomic check-and-reserve prevents that race.