Junior Replication & Sharding Interview Questions (50 MCQs)

Question 1

What is the main purpose of replication in a database?

Accepted Answer

Keeping identical copies of data on several machines

Answer

Splitting data into separate parts across machines

Answer

Compressing data to reduce disk usage

Answer

Encrypting data before writing it to disk

Question 2

What does sharding (partitioning) mean in a database?

Accepted Answer

Splitting data into subsets, each stored on a different node

Answer

Copying the full dataset onto every node in the cluster

Answer

Moving old data to a cheaper storage tier

Answer

Building indexes on frequently queried columns

Question 3

What is the key difference between replication and sharding?

Accepted Answer

Replication copies data; sharding divides data

Answer

Replication divides data; sharding copies data

Answer

Both copy data, just at different speeds

Answer

Replication is NoSQL-only and sharding is SQL-only

Question 4

In a leader-follower (single-leader) setup, where are writes sent?

Accepted Answer

Only to the leader, which streams changes to followers

Answer

To whichever follower is closest to the client

Answer

To all nodes at once, sent directly by the client

Answer

To a random follower that later informs the leader

Question 5

What role do followers typically play in a leader-follower setup?

Accepted Answer

Serving reads and standing by to replace the leader

Answer

Accepting writes and forwarding them to the leader for approval

Answer

Holding only indexes without the actual data

Answer

Dividing the data among themselves by shard key

Question 6

What does adding more read replicas scale?

Accepted Answer

Read capacity, but not write capacity

Answer

Write capacity, but not read capacity

Answer

Both read and write capacity equally

Answer

Only the total storage capacity for data

Question 7

In synchronous replication, when does the leader acknowledge a write to the client?

Accepted Answer

After the synchronous follower(s) confirm the write

Answer

Immediately after writing to the leader's own disk

Answer

Before the write is even stored on the leader

Answer

After a fixed delay, regardless of followers

Question 8

What is the defining trait of asynchronous replication?

Accepted Answer

The leader acknowledges writes without waiting for followers

Answer

Followers store the write before the leader does

Answer

Writes are acknowledged only when all replicas have them

Answer

Followers never receive the leader's changes

Question 9

What is the biggest risk of async replication when the leader crashes?

Accepted Answer

Losing writes that had not yet reached the followers

Answer

All followers locking up permanently until the leader returns

Answer

Every record being duplicated on the followers

Answer

Automatic deletion of all follower data

Question 10

What is the main drawback of synchronous replication?

Accepted Answer

Slower writes, which block if the follower is unavailable

Answer

Data loss whenever the leader crashes

Answer

Being unable to read from followers at all

Answer

Requiring sharding in order to work

Question 11

What is replication lag?

Accepted Answer

The delay between a write on the leader and its application on a follower

Answer

The time a query takes to execute on the leader

Answer

The network latency between client and load balancer

Answer

The time needed to build a new replica from scratch

Question 12

A user edits their profile and after refreshing sees the old value. Which problem is this?

Accepted Answer

A read-your-writes violation from reading a lagging replica

Answer

Split brain between two leaders in the cluster

Answer

A hot spot on the user's shard

Answer

Data loss caused by a disk failure

Question 13

What is failover in a leader-follower system?

Accepted Answer

Promoting a follower to leader when the leader fails

Answer

Shutting down all followers to reduce leader load

Answer

Moving data onto a new shard

Answer

Restoring the database from last night's backup

Question 14

What is split brain in replication?

Accepted Answer

Two nodes both believe they are leader and accept writes

Answer

A table's data has been split across two shards

Answer

Leader and follower run different database versions

Answer

Clients send half their queries to a cache

Question 15

Sharding is usually what kind of partitioning?

Accepted Answer

Horizontal: splitting a table's rows across nodes

Answer

Vertical: splitting a table's columns into several tables

Answer

Functional: giving each service its own database

Answer

Temporal: keeping only today's data

Question 16

What is a shard key (partition key)?

Accepted Answer

The field that determines which shard each record goes to

Answer

An encryption key generated separately for each shard

Answer

The primary key of the table that lists all shards

Answer

An ID each replica uses to connect to the leader

Question 17

How does range partitioning divide data across shards?

Accepted Answer

Each shard holds a contiguous range of key values

Answer

Each record is sent to a random shard

Answer

The key's hash is taken modulo the shard count

Answer

All records are copied onto every shard

Question 18

What is the main advantage of hash partitioning over range partitioning?

Accepted Answer

More even distribution of data and load across shards

Answer

Faster range queries over the key

Answer

Completely removing the need for replication

Answer

Free joins across all tables

Question 19

Why is range partitioning on `created_at` problematic for log data?

Accepted Answer

All new writes land on the last shard (current range)

Answer

Timestamps cannot be used as a shard key at all

Answer

Old data gets deleted automatically

Answer

Range queries on time become impossible

Question 20

What is the main downside of hash partitioning?

Accepted Answer

Range queries on the key must hit every shard

Answer

Data always piles up on a single shard

Answer

Looking up a single key requires scanning all shards

Answer

Shards can no longer be replicated

Question 21

How does directory-based sharding work?

Accepted Answer

A lookup table stores the mapping of each key to a shard

Answer

The key's hash directly determines the shard number

Answer

Each shard owns a fixed alphabetical range

Answer

Clients pick a shard at random

Question 22

What is a hot spot in a sharded system?

Accepted Answer

A shard receiving a disproportionate share of the load

Answer

A shard whose data was just backed up

Answer

A replica whose hardware is running hot

Answer

A shard that contains no data at all

Question 23

What does the 'celebrity problem' refer to in sharding?

Accepted Answer

A very popular key that overloads its own shard

Answer

Users whose data is spread across many shards

Answer

A shard manually chosen by an administrator

Answer

A leader with more memory than its followers

Question 24

Why are queries that span multiple shards expensive?

Accepted Answer

They must go to several nodes and results be merged

Answer

The database forbids running them in any form

Answer

They always cause data loss

Answer

They can only run on leaders, never replicas

Question 25

What is rebalancing in a sharded cluster?

Accepted Answer

Moving data between nodes to even out the load

Answer

Changing the schema on all shards simultaneously

Answer

Restarting all nodes one by one

Answer

Removing extra replicas to save money

Question 26

What is the problem with using `hash(key) % N` to pick a shard?

Accepted Answer

When N changes, almost every key has to move

Answer

Data distribution across shards is always uneven

Answer

Finding a key's shard is very slow

Answer

It only works for numeric keys

Question 27

What benefit does consistent hashing provide?

Accepted Answer

Adding/removing a node moves only a small fraction of keys

Answer

It guarantees every key is stored on every node

Answer

It removes the need for a hash function entirely

Answer

It makes range queries over keys faster

Question 28

What are virtual nodes (vnodes) used for in consistent hashing?

Accepted Answer

Evening out key spread by placing each node at many ring points

Answer

Creating virtual nodes that keep data only in memory

Answer

Replacing replication with virtual copies

Answer

Hiding node addresses from clients

Question 29

Why are replicas usually placed in different availability zones?

Accepted Answer

So one zone failure does not take out all copies at once

Answer

So data transfer between zones becomes free

Answer

Because a zone can only host one database

Answer

So that replication lag drops to zero

Question 30

What does a replication factor of 3 mean?

Accepted Answer

Each piece of data is stored on 3 different nodes

Answer

Data is split into 3 different shards

Answer

Each write is repeated 3 times on one disk

Answer

The cluster can have at most 3 nodes

Question 31

What is semi-synchronous replication?

Accepted Answer

The leader waits for at least one follower; the rest are async

Answer

Half the writes are sync and the other half async

Answer

Replication only runs during off-peak hours

Answer

Followers receive only half of the data

Question 32

A user sees a new comment, refreshes, and the comment disappears. Which guarantee was violated?

Accepted Answer

Monotonic reads

Answer

Transaction atomicity

Answer

Write durability

Answer

Partition tolerance

Question 33

What does eventual consistency guarantee in a replicated system?

Accepted Answer

If writes stop, all replicas eventually converge

Answer

All replicas hold identical data at every moment

Answer

Every write reaches all replicas within one second

Answer

Reads always return the latest written value

Question 34

What is multi-leader replication?

Accepted Answer

Several nodes accept writes and replicate changes to each other

Answer

One leader manages several different shards

Answer

The leader rotates every few minutes

Answer

Every client gets its own dedicated leader

Question 35

How are writes done in leaderless (Dynamo-style) replication?

Accepted Answer

The client or a coordinator sends the write to several replicas

Answer

The write is sent only to a fixed leader

Answer

The write goes to a queue and then to the leader

Answer

Each replica accepts writes on its assigned day

Question 36

With N replicas, what does the quorum condition `W + R > N` ensure?

Accepted Answer

Read and write replica sets overlap in at least one node

Answer

All replicas always have exactly the same data

Answer

No replica ever goes down

Answer

Writes are always faster than reads

Question 37

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

Accepted Answer

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

Answer

All conflicting writes are merged together

Answer

The conflict is shown to the user to resolve manually

Answer

The write that arrived first is kept

Question 38

Two users add an item to a shared cart concurrently and the system uses LWW. What happens?

Accepted Answer

One of the two additions is silently lost

Answer

Both items are correctly recorded in the cart

Answer

The system rejects both writes

Answer

The cart is wiped out completely

Question 39

What is the benefit of placing replicas in regions close to users?

Accepted Answer

Lower read latency for users in that region

Answer

Completely eliminating replication lag

Answer

Reducing the number of writes the leader receives

Answer

Removing the need for backups

Question 40

Why is a column like `country` with few distinct values a poor shard key?

Accepted Answer

Few, skewed values produce unbalanced shards

Answer

Strings can never be used as shard keys

Answer

This column causes replication to be disabled

Answer

Querying by country becomes impossible

Question 41

In a store, why is sharding the `orders` table by `user_id` common?

Accepted Answer

All of a user's orders land on one shard, so queries hit one shard

Answer

Because `user_id` is always the smallest column

Answer

Because it puts all orders on a single shard

Answer

Because it removes the need for any index

Question 42

How is a new follower typically set up without stopping the leader?

Accepted Answer

Take a leader snapshot, then apply changes made after it

Answer

Lock the whole database and copy its files

Answer

Copy the data from the application cache

Answer

Replay all user requests from the very beginning

Question 43

What does a leader typically use to ship changes to followers?

Accepted Answer

A replication log such as the WAL or binlog

Answer

A full database copy after every write

Answer

An email or message to the system admin

Answer

The application's HTTP access logs

Question 44

What is usually tried before sharding a database?

Accepted Answer

Vertical scaling, read replicas, caching, and query tuning

Answer

Dropping all indexes to speed up writes

Answer

Disabling backups to free up resources

Answer

Moving all data into a CSV file

Question 45

Why is using each shard's local `AUTO_INCREMENT` for IDs problematic?

Accepted Answer

Different shards generate duplicate IDs

Answer

`AUTO_INCREMENT` stops working in sharded systems

Answer

IDs become too large to store

Answer

It restricts sharding to range-based only

Question 46

Why is geo-partitioning (storing each user's data in their region) usually done?

Accepted Answer

Lower latency and data-residency compliance like GDPR

Answer

Reducing the total number of users

Answer

Simplifying global queries across regions

Answer

Removing the need for replication in each region

Question 47

Why is replication not a substitute for backups?

Accepted Answer

A mistaken `DELETE` or corrupt data replicates to all replicas too

Answer

Replicas do not allow reading data

Answer

Replication copies only table structure

Answer

Replicas wipe themselves after a day

Question 48

How do automatic failover systems usually detect that the leader has failed?

Accepted Answer

Missing leader heartbeats for a set timeout

Answer

An operator manually checking logs every minute

Answer

Counting the number of online users

Answer

Comparing the leader's disk size with followers

Question 49

In a system using both sharding and replication, which statement is usually true?

Accepted Answer

Each shard has replicas, and a node can lead some shards and follow others

Answer

Every node holds all data and one node leads the whole cluster

Answer

Shards have no replicas because sharding replaces replication

Answer

Each replica stores different columns of a shard

Question 50

Why does multi-leader replication need conflict resolution while single-leader usually does not?

Accepted Answer

Because two leaders can change the same record differently at once

Answer

Because in multi-leader all followers reject writes

Answer

Because single-leader does not use a replication log at all

Answer

Because multi-leader is only for read-only data