Junior Consistency, CAP & Consensus Interview Questions (50 MCQs)

Question 1

What do the letters C, A, and P stand for in the CAP theorem?

Accepted Answer

Consistency, Availability, and Partition tolerance

Answer

Concurrency, Atomicity, and Persistence

Answer

Caching, Availability, and Performance

Answer

Consistency, Atomicity, and Partitioning

Question 2

According to CAP, when a network partition happens, what must a system choose between?

Accepted Answer

Between Consistency and Availability

Answer

Between Latency and Throughput

Answer

Between Durability and Atomicity

Answer

Between Partition tolerance and Durability

Question 3

How does a CP system typically behave during a network partition?

Accepted Answer

Rejects or errors some requests rather than return inconsistent data

Answer

Accepts all requests and merges conflicts later on

Answer

Automatically heals the partition through fast retries

Answer

Keeps data only in each node's local cache

Question 4

How does an AP system typically behave during a network partition?

Accepted Answer

Keeps answering requests, even if data may be stale or divergent

Answer

Takes all nodes offline until the partition is resolved

Answer

Only answers requests that manage to reach the leader

Answer

Blocks every write until the network is restored

Question 5

Why is labeling a real distributed system as "CA" usually misleading?

Accepted Answer

Because network partitions are unavoidable, so P cannot be given up

Answer

Because Consistency and Availability are never possible together

Answer

Because CA is only defined for NoSQL databases

Answer

Because CA requires atomic clocks on every server

Question 6

What is a network partition in distributed systems?

Accepted Answer

Groups of nodes losing contact with each other while still running

Answer

Splitting data across several shards by a key

Answer

Dividing a server's disk into separate partitions

Answer

Separating read and write traffic onto different servers

Question 7

What does ACID stand for in database transactions?

Accepted Answer

Atomicity, Consistency, Isolation, Durability

Answer

Availability, Concurrency, Integrity, Distribution

Answer

Atomicity, Caching, Indexing, Durability

Answer

Accuracy, Consistency, Isolation, Delivery

Question 8

What does Atomicity in ACID guarantee?

Accepted Answer

Either all operations of a transaction apply, or none do

Answer

Concurrent transactions never see each other at all

Answer

Data survives a power loss once it is committed

Answer

Each transaction executes in under one millisecond

Question 9

What does Isolation in ACID mean?

Accepted Answer

Concurrent transactions run without unwanted interference with each other

Answer

Each transaction runs on its own separate server

Answer

Each user's data is stored in an independent database

Answer

Transactions never have any access to the network

Question 10

What does Durability in ACID guarantee?

Accepted Answer

After commit, data persists even through a crash or power loss

Answer

A transaction is retried automatically until it succeeds

Answer

Data is kept forever in RAM for quick access

Answer

No two transactions can ever modify the same row

Question 11

What does the BASE model stand for in distributed databases?

Accepted Answer

Basically Available, Soft state, Eventual consistency

Answer

Batch Access, Strong state, Exact consistency

Answer

Basic Atomicity, Serial execution, Ensured durability

Answer

Balanced Availability, Sharded storage, Elastic scaling

Question 12

Which scenario usually needs ACID guarantees more than BASE?

Accepted Answer

Recording a money transfer between two bank accounts

Answer

Counting the likes on a social media post

Answer

Showing a video's view count on the homepage

Answer

Storing the user's recent search history

Question 13

What does strong consistency mean from the client's perspective?

Accepted Answer

Every read sees the latest acknowledged write, as if one copy existed

Answer

Data will eventually become identical on all replicas

Answer

Each client can only ever see its own writes

Answer

Reads are always served by the nearest replica

Question 14

What does eventual consistency guarantee?

Accepted Answer

If writes stop, all replicas eventually converge to the same value

Answer

Every read always returns the most recently written value

Answer

Writes are applied in one global order on every node

Answer

Data reaches all nodes within at most one second

Question 15

Which system is a classic example of eventual consistency?

Accepted Answer

DNS, where record changes propagate gradually based on TTL

Answer

A single-node relational database with Serializable transactions

Answer

etcd, which needs majority agreement for every write

Answer

A mutex in the memory of a single process

Question 16

What does the read-your-writes guarantee mean?

Accepted Answer

A user who wrote something always sees that change on later reads

Answer

All users see each other's writes immediately

Answer

Each user is only allowed to read their own data

Answer

Writes are only allowed after a successful read

Question 17

What problem does the monotonic reads guarantee prevent?

Accepted Answer

Seeing new data and then an older version on a later read

Answer

Writing two different values to one key concurrently

Answer

Reading data that has not been committed yet

Answer

Reads slowing down as replicas are added

Question 18

In asynchronous replication, what is the main cause of stale reads?

Accepted Answer

Replication lag; the follower has not yet received the leader's latest changes

Answer

Data being erased from the leader's disk after commits

Answer

Data being compressed while sent across the network

Answer

Using indexes on the table's columns

Question 19

What is a quorum in distributed systems?

Accepted Answer

The minimum number of nodes that must acknowledge an operation for it to count

Answer

The total number of servers installed in the cluster

Answer

The node with the most CPU and RAM available

Answer

The list of clients that are allowed to write

Question 20

With N replicas, which condition makes a read overlap with the latest write?

Accepted Answer

`R + W > N`

Answer

`R + W = N`

Answer

`R + W < N`

Answer

`R = W = 1`

Question 21

In a three-node cluster, how many nodes form a majority quorum?

Accepted Answer

2 nodes

Answer

1 node

Answer

3 nodes

Answer

4 nodes

Question 22

How many simultaneous node failures can a five-node majority-quorum cluster tolerate?

Accepted Answer

2 nodes

Answer

1 node

Answer

3 nodes

Answer

4 nodes

Question 23

Why do consensus clusters like etcd usually have an odd number of nodes?

Accepted Answer

An extra even node adds no fault tolerance but enlarges the quorum

Answer

Because consensus algorithms cannot run at all with even counts

Answer

Because odd counts make data get written to disk faster

Answer

Because the leader must always be the middle node in the list

Question 24

What is clock skew in distributed systems?

Accepted Answer

The difference between the physical clocks of different nodes

Answer

The network delay between client and server

Answer

The time a transaction spends waiting on a lock

Answer

The gap between a request's CPU time and I/O time

Question 25

Why is ordering events purely by system-clock timestamps from different nodes risky?

Accepted Answer

Clocks skew, so a later event may get a smaller timestamp

Answer

Because timestamps take a lot of storage in the database

Answer

Because timestamps are only stored with second precision

Answer

Because comparing two timestamps is a very slow operation

Question 26

What is a Lamport clock designed for?

Accepted Answer

Logical ordering consistent with happened-before, without physical clocks

Answer

Precisely syncing server clocks with an atomic clock

Answer

Measuring the network latency between two nodes

Answer

Scheduling cron job execution in a cluster

Question 27

When a node receives a message with Lamport timestamp `t`, how does it update its clock?

Accepted Answer

`max(local, t) + 1`

Answer

`min(local, t) + 1`

Answer

`local + t`

Answer

`t` unchanged

Question 28

What is the main advantage of a vector clock over a Lamport clock?

Accepted Answer

It can detect that two events were concurrent

Answer

It takes less space than a single integer

Answer

It precisely synchronizes the nodes' physical clocks

Answer

It removes the need to send messages between nodes

Question 29

What is the consensus problem in distributed systems?

Accepted Answer

Several nodes agreeing on one value even if some nodes fail

Answer

Splitting request load evenly across servers

Answer

Compressing data before sending it to replicas

Answer

Picking the fastest network path between datacenters

Question 30

What is Raft?

Accepted Answer

A consensus algorithm designed to be easier to understand than Paxos

Answer

An encryption protocol for secure node communication

Answer

A NoSQL database built on eventual consistency

Answer

A load balancing algorithm based on round-robin

Question 31

Which roles can a node have in Raft?

Accepted Answer

Leader, Follower, and Candidate

Answer

Master, Slave, and Arbiter

Answer

Proposer, Acceptor, and Learner

Answer

Primary, Secondary, and Witness

Question 32

In Raft, which node handles client writes?

Accepted Answer

The leader, which appends entries and replicates them to followers

Answer

Whichever follower receives the request first

Answer

All nodes, independently and at the same time

Answer

The node with the lowest CPU load

Question 33

What role do heartbeats play in Raft?

Accepted Answer

The leader signals it is alive so followers don't start a new election

Answer

Followers use it to report their CPU usage to the leader

Answer

Clients use it to keep their cluster connection open

Answer

It is used to sync all nodes' physical clocks

Question 34

What is a term in Raft?

Accepted Answer

A logical period with an increasing number, starting with each election

Answer

A fixed one-minute period each leader may stay active

Answer

The number of entries sent in one batch

Answer

The unique identifier of each node in the cluster

Question 35

What does Paxos refer to in distributed systems?

Accepted Answer

A family of consensus algorithms introduced by Leslie Lamport

Answer

A serialization format for messages between services

Answer

A sharding method based on consistent hashing

Answer

A protocol for synchronizing server clocks

Question 36

What is split brain in a cluster?

Accepted Answer

Two nodes simultaneously believe they are leader and accept writes

Answer

A table's data is split into two equal shards

Answer

A server has two separate CPUs installed

Answer

Reads and writes go to two different databases

Question 37

What is a distributed lock used for?

Accepted Answer

Ensuring only one process across servers does a task at a time

Answer

Encrypting data before storing it in several datacenters

Answer

Locking a user account after failed login attempts

Answer

Blocking network access to production servers

Question 38

Why do distributed locks usually have a TTL or expiry?

Accepted Answer

So the lock isn't stuck forever if its holder crashes

Answer

To reduce memory usage on the lock server

Answer

So the lock travels faster over the network

Answer

So several processes can hold the lock at once

Question 39

What is a lease in distributed systems?

Accepted Answer

A time-limited grant that must be renewed before it expires

Answer

A contract for renting servers from a cloud provider

Answer

A permanent connection between client and database

Answer

A periodic backup of the cluster's data

Question 40

What are the two phases of Two-Phase Commit (2PC)?

Accepted Answer

Prepare (voting) and Commit/Abort

Answer

Read and Write

Answer

Lock and Unlock

Answer

Election and Replication

Question 41

In 2PC, what happens if one participant votes "no" during prepare?

Accepted Answer

The coordinator aborts the transaction on all participants

Answer

The others commit while that one participant aborts

Answer

The coordinator ignores the vote and commits anyway

Answer

The transaction waits until that vote is changed

Question 42

What is the Saga pattern in microservices?

Accepted Answer

A sequence of local transactions undone by compensating actions on failure

Answer

A distributed transaction that locks all services at once

Answer

A way of caching service responses in the API gateway

Answer

A pattern for splitting a big database into shards

Question 43

In a travel booking saga, if payment fails after the hotel is booked, what is the right compensating action?

Accepted Answer

Cancelling the hotel booking with a separate operation

Answer

Automatic rollback of the hotel database through 2PC

Answer

Ignoring the failure and continuing to book the flight

Answer

Retrying the payment forever until it succeeds

Question 44

Why does idempotency matter so much in distributed systems?

Accepted Answer

Retries are inevitable, and repeating an operation must not double its effect

Answer

Because it ensures requests never time out at all

Answer

Because it cuts network bandwidth usage in half

Answer

Because it removes the need to authenticate clients

Question 45

How does an idempotency key work in a payment API?

Accepted Answer

The client sends a unique ID, and the server runs duplicate requests with it only once

Answer

The server generates a new encryption key for each request

Answer

The key makes the request travel through the fastest route

Answer

The key is used in place of the user's auth token

Question 46

With at-least-once message delivery, what must a consumer be prepared for?

Accepted Answer

Receiving duplicate messages

Answer

Some messages being permanently lost

Answer

Receiving messages only in batches

Answer

Manually decrypting every message

Question 47

Why is true "exactly-once delivery" over an unreliable network so hard?

Accepted Answer

The sender can't distinguish a lost message from a lost ack

Answer

Because messages are always encrypted on the network

Answer

Because TCP does not allow any retransmission

Answer

Because brokers only keep messages in RAM

Question 48

What are ZooKeeper and etcd typically used for?

Accepted Answer

Coordination: config storage, leader election, and service discovery

Answer

Storing large files such as videos and images

Answer

Running analytical queries on big datasets

Answer

Caching web pages at the network edge

Question 49

What is the main cost of synchronous replication compared to async?

Accepted Answer

Higher write latency and availability depending on replicas

Answer

Higher chance of losing data after the leader crashes

Answer

More stale reads on the synchronous followers

Answer

More disk space consumed on the leader

Question 50

How does the Last-Write-Wins (LWW) conflict resolution strategy work?

Accepted Answer

The write with the highest timestamp is kept and others are discarded

Answer

All conflicting writes are merged together and stored

Answer

The user is asked to choose the correct version

Answer

The first write to reach the server is kept