Question 1

What is a load balancer and what goes wrong without one?

Accepted Answer

A component between clients and a pool of backend servers that routes each request to the most appropriate server according to an algorithm. Without it, all traffic hits one server: high response times, overload, degraded performance, more downtime. With it: better performance, scalability, availability and fault tolerance.

Question 2

What are the two core functions of a load balancer?

Accepted Answer

1. Choosing the server for each incoming request using a load-balancing algorithm.
2. Health checks: continuously verifying backends are healthy before routing to them.

Question 3

Round robin: how it works and its disadvantage.

Accepted Answer

Request 1 → Server 1
Request 2 → Server 2
Request 3 → Server 3
Request 4 → Server 1 …

Each server gets roughly equal requests; only a single counter is needed, no server state. Disadvantage: ignores server capacity, so a small server gets the same load as a big one.

Question 4

Weighted round robin: what does it fix?

Accepted Answer

Each server gets a weight based on hardware capacity and receives proportionally more requests. Weights 5 / 3 / 2 give Server 1 50%, Server 2 30%, Server 3 20% of traffic. Fixes plain round robin's blindness to capacity.

Question 5

Geo-based routing: goal, requirements, and its disadvantage.

Accepted Answer

Route each user to the geographically closest server to minimize network latency: India → Mumbai, Europe → Frankfurt, USA → Virginia. The balancer tracks request origin and server locations. Data must be replicated and partitioned across regions, with nearby replicas as failover. Disadvantage: VPNs and proxies mask the user's real location and misroute traffic.

Question 6

Least connection: how it works and what it misses.

Accepted Answer

Route to the server with the fewest active connections (Server 1: 120, Server 2: 45, Server 3: 70 → Server 2). Counts must be kept current. Disadvantage: only connections are considered, not response time or capacity; a server with few connections may still be doing heavy processing. Suited to long-lived sessions.

Question 7

Least time: how is it better than least connection?

Accepted Answer

Route to the server with the lowest average response time. The balancer continuously measures response time, request count and current workload per server, so it implicitly accounts for capacity (bigger servers respond faster). Ideal for latency-sensitive systems like stock trading.

Question 8

IP hash: what does it guarantee and why is it hard to scale?

Accepted Answer

Client IP → hash function → assigned server   (e.g. hashes 1–10 → S1, 11–20 → S2, 21–30 → S3)

The same client always lands on the same server: session persistence. That server must stay healthy, or session data must be replicated elsewhere. Disadvantages: adding a server means rehashing and reassigning everything; not ideal for microservices where session continuity is critical.

Question 9

Compare the six algorithms.

Accepted Answer

- Round robin: Decides by - Sequence. Strength - Simplest; identical servers. Weakness - Ignores capacity.
- Weighted round robin: Decides by - Sequence × weight. Strength - Respects capacity. Weakness - Static weights.
- Geo-based: Decides by - User location. Strength - Lowest network latency. Weakness - VPN/proxy misrouting; needs regional data.
- Least connection: Decides by - Active connections. Strength - Adapts to live load. Weakness - Blind to processing cost.
- Least time: Decides by - Avg response time. Strength - Reflects real performance. Weakness - More bookkeeping.
- IP hash: Decides by - Hash of client IP. Strength - Session persistence. Weakness - Rehash on scaling.

Question 10

What is hybrid load balancing? Give combinations.

Accepted Answer

Real systems combine algorithms: IP hash + round robin (persistence plus even spread), geo-based + least time (nearest region, then fastest server in it), weighted round robin + least connection (capacity plus current load). Optimizes several factors at once.

Question 11

What are health checks, and what do they buy you?

Accepted Answer

Mechanisms by which the load balancer continuously monitors availability, responsiveness and health of backends and routes only to healthy ones. Benefits: detect unhealthy or degraded servers automatically, stop sending them traffic, fewer user-facing errors, metrics for scaling decisions, automatic recovery when a server is healthy again.

Question 12

Passive vs active health checks.

Accepted Answer

Passive: observe real production traffic (response times, connection failures, HTTP status codes, error rates, timeouts). No extra load; reveals problems under actual workload.

Active: send dedicated probe requests at intervals regardless of user traffic, e.g. GET /health expecting 200 OK. Faster and more accurate detection at the cost of a little extra traffic.

Question 13

What are the three health-check parameters?

Accepted Answer

- Interval: Meaning - How often checks run. Example - Every 5 seconds.
- Timeout: Meaning - Max wait for a response. Example - 50 ms–5 s expected; 10 s counts as failure.
- Threshold: Meaning - Consecutive failures/successes needed to flip status. Example - 20 failures → unhealthy; 100 successes → healthy.

Question 14

Walk through the health-check workflow from failure to recovery.

Accepted Answer

1. At each interval, check all servers.
2. Response within timeout → success; none → failure.
3. Each failure increments a counter.
4. Failure threshold reached → server marked unhealthy and removed from the pool; no new traffic.
5. Checks continue on the removed server.
6. Success threshold met → server added back and receives traffic again.

Question 15

What is replication and what does it improve?

Accepted Answer

Copying and maintaining identical data on multiple databases or nodes instead of one location. Improves high availability, fault tolerance, read performance, disaster recovery and durability: the application keeps running when a database server goes down.

Question 16

Name the three replication strategies.

Accepted Answer

Single-leader, multi-leader, leaderless.

Question 17

Single-leader replication: how does it work?

Accepted Answer

One database is the leader (primary); the rest are followers (replicas). Only the leader accepts writes; followers receive replicated data from it and serve reads.

User → Application → Leader → Follower 1 / Follower 2 / Follower 3

Question 18

Asynchronous vs synchronous replication: trade-offs, and which is used in practice.

Accepted Answer

- Leader responds: Asynchronous - Immediately after its own write; followers catch up in the background. Synchronous - Only after every follower acknowledges.
- Advantage: Asynchronous - Fast response, no blocking. Synchronous - Followers always current; any follower can become leader.
- Disadvantage: Asynchronous - Consistency not guaranteed; followers may lag. Synchronous - Longer response time; resources blocked while waiting.

In practice asynchronous is preferred for most use cases; failed followers resync once they recover.

Question 19

How do you add a new follower node?

Accepted Answer

1. Take a snapshot of the data from the leader (or an existing follower).
2. Load the snapshot into the new node.
3. Connect the new node to the leader for ongoing replication.

Question 20

What happens when the leader fails?

Accepted Answer

A new leader is elected: identify the most up-to-date follower (compare timestamps), then inform every other node about the new leader. Tools such as FSImage (full metadata snapshot) and EditLogs (incremental change logs) support replication and recovery.

Scalability & Data Management

Scalability & Data Management