Why In-sync replicas (ISR) in Kafka? - Purpose & Use Cases

The Big Idea

What if your data vanished the moment your main server failed? ISR stops that nightmare.

The Scenario

Imagine you run a busy online store where orders come in every second. You keep a list of orders on one computer. If that computer crashes, you lose all recent orders and your customers get upset.

The Problem

Relying on just one computer means if it fails, data is lost. Manually copying data to other computers takes time and can miss updates. This causes delays and errors, making your store unreliable.

The Solution

In-sync replicas (ISR) automatically keep copies of data on multiple computers updated at the same time. If one fails, another ready copy takes over instantly, so no orders are lost and your store stays open.

Before vs After

✗ Before

producer.send(data)
# no guarantee data is copied to other servers immediately

✓ After

producer.send(data).get()
# data is confirmed on all in-sync replicas before continuing

What It Enables

It enables reliable, fault-tolerant systems that never lose data even if some servers fail.

Real Life Example

When you post a message on social media, ISR ensures your post is saved on multiple servers instantly, so it won't disappear if one server crashes.

Key Takeaways

Manual data copying is slow and risky.

ISR keeps multiple copies updated together automatically.

This protects data and keeps systems running smoothly.

Practice

1. What does In-sync replicas (ISR) mean in Kafka?

easy

A. Replicas that are offline

B. Replicas that never receive data

C. Replicas that have the latest data from the leader

D. Replicas that only store old data

Solution

1. Step 1: Understand the role of ISR

   ISR are replicas that keep up-to-date with the leader's data.

2. Step 2: Compare options

   Only Replicas that have the latest data from the leader correctly describes ISR as replicas with the latest data.

3. Final Answer:

   Replicas that have the latest data from the leader -> Option C

4. Quick Check:

   ISR = Latest data replicas [OK]

Hint: ISR means replicas fully caught up with leader data [OK]

Common Mistakes:

• Thinking ISR includes offline replicas
• Confusing ISR with replicas that lag behind
• Assuming ISR means replicas with old data

2. Which Kafka configuration controls the minimum number of ISR replicas required to acknowledge a write?

easy

A. replica.fetch.max.bytes

B. min.insync.replicas

C. max.insync.replicas

D. offsets.retention.minutes

Solution

1. Step 1: Identify configuration related to ISR acknowledgments

   The setting min.insync.replicas defines the minimum ISR count for write success.

2. Step 2: Eliminate unrelated options

   Other options control fetch size or retention, not ISR acknowledgments.

3. Final Answer:

   min.insync.replicas -> Option B

4. Quick Check:

   Minimum ISR replicas = min.insync.replicas [OK]

Hint: min.insync.replicas sets minimum ISR for writes [OK]

Common Mistakes:

• Confusing min.insync.replicas with max.insync.replicas
• Choosing unrelated configs like offsets.retention.minutes
• Assuming replica.fetch.max.bytes controls ISR count

3. Given a Kafka topic with 3 replicas and ISR = [replica1, replica2], what happens if replica2 falls behind and leaves ISR?

medium

A. Only replica1 remains in ISR; leader continues with one ISR

B. Leader stops producing until replica2 catches up

C. All replicas are removed from ISR

D. Replica2 becomes the new leader immediately

Solution

1. Step 1: Understand ISR behavior when a replica falls behind

   If a replica lags, it is removed from ISR, leaving others in ISR.

2. Step 2: Analyze leader and ISR state

   The leader continues with remaining ISR replicas; it does not stop or change leader immediately.

3. Final Answer:

   Only replica1 remains in ISR; leader continues with one ISR -> Option A

4. Quick Check:

   Lagging replica leaves ISR, leader continues [OK]

Hint: Lagging replica leaves ISR; leader stays with remaining ISR [OK]

Common Mistakes:

• Thinking leader stops producing when ISR shrinks
• Assuming all replicas leave ISR together
• Believing lagging replica becomes leader immediately

4. You see this error in Kafka logs: "Not enough replicas are in ISR to satisfy min.insync.replicas". What is the likely cause?

medium

A. The consumer group is not committed

B. The leader replica is offline

C. The topic has zero partitions

D. Too few replicas are currently in ISR compared to min.insync.replicas

Solution

1. Step 1: Interpret the error message

   The error means the number of replicas in ISR is less than the configured minimum.

2. Step 2: Match cause with options

   Too few replicas are currently in ISR compared to min.insync.replicas directly matches the error cause; other options are unrelated to ISR count.

3. Final Answer:

   Too few replicas are currently in ISR compared to min.insync.replicas -> Option D

4. Quick Check:

   ISR count below min.insync.replicas causes error [OK]

Hint: Error means ISR count < min.insync.replicas [OK]

Common Mistakes:

• Blaming leader offline instead of ISR count
• Confusing consumer group commit with ISR errors
• Thinking zero partitions cause this error

5. You want to ensure high availability in Kafka by allowing leader failover only if at least 2 replicas are in ISR. Which configuration and setup achieve this?

hard

A. Set min.insync.replicas=2 and have at least 3 replicas per partition

B. Set min.insync.replicas=1 and have 2 replicas per partition

C. Set min.insync.replicas=3 with 2 replicas per partition

D. Set min.insync.replicas=0 and 1 replica per partition

Solution

1. Step 1: Understand min.insync.replicas meaning

   This setting defines the minimum ISR count required for leader failover and write success.

2. Step 2: Match configuration to requirement

   To require at least 2 ISR replicas, set min.insync.replicas=2 and have at least 3 replicas to allow failover.

3. Final Answer:

   Set min.insync.replicas=2 and have at least 3 replicas per partition -> Option A

4. Quick Check:

   High availability needs min.insync.replicas=2 with 3 replicas [OK]

Hint: min.insync.replicas ≤ replica count for failover [OK]

Common Mistakes:

• Setting min.insync.replicas higher than replica count
• Using too few replicas for desired availability
• Setting min.insync.replicas to zero disables safety