Understanding Replication

1. Understanding Replication Concept

Concept Description Detail
Replication Copy partitions across brokers Fault tolerance
RF=N Survives N-1 failures Durability
Per Partition Each has own replica set Independent

Example: Topic with RF=3

kafka-topics.sh --create --topic critical --partitions 6 \
  --replication-factor 3 --bootstrap-server localhost:9092

2. Understanding Leader and Followers

Role Responsibility Detail
Leader Handles all reads/writes One per partition
Follower Replicates from leader Passive
Fetch Followers pull from leader Like consumers

Example: Identify leader

kafka-topics.sh --describe --topic critical --bootstrap-server localhost:9092 \
  | grep "Leader:"

3. Configuring Replication Factor

Config Description Default
default.replication.factor Auto-created topics RF 1
--replication-factor Per-topic at create Explicit
Constraint RF ≤ broker count Hard limit

Example: Broker default RF

default.replication.factor=3
offsets.topic.replication.factor=3

4. Understanding In-Sync Replicas

Concept Description Detail
ISR Replicas caught up to leader Election eligible
Shrink Lagging replica removed Falls behind
Expand Caught-up replica re-added Recovered

Example: Check ISR

kafka-topics.sh --describe --topic critical --bootstrap-server localhost:9092
# ... Isr: 1,2,3  (all in sync)

5. Configuring Replica Lag

Config Description Default
replica.lag.
time.max.ms		 Max follower lag before ISR removal 30000
replica.fetch.
max.bytes		 Bytes per fetch 1048576
num.replica.fetchers Fetcher threads 1

Example: Tune replica lag window

replica.lag.time.max.ms=30000
num.replica.fetchers=4

6. Understanding Leader Election

Type Description Risk
Clean New leader from ISR No data loss
Unclean Out-of-sync replica promoted Possible loss
Preferred Restore first replica Load balance

Example: Disable unclean election

unclean.leader.election.enable=false

7. Understanding Replica High Watermark

Concept Description Detail
High Watermark Highest offset replicated to all ISR HW
Consumer Visibility Reads only up to HW Committed data
Advance Moves as ISR catch up Min of LEOs

Example: HW limits reads

LEO=1050  HW=1000
Consumers can read offsets 0..999 (records up to HW)

8. Understanding Log End Offset

Concept Description Detail
LEO Offset of next record to append Per replica
Leader LEO Tip of the partition Latest
Follower LEO How far follower copied ≤ leader LEO

Example: Inspect LEO via JMX

# kafka.log:type=Log,name=LogEndOffset,topic=critical,partition=0

9. Configuring Preferred Leader

Aspect Description Detail
Preferred First replica in list Default leader
Rebalance Restore for even load Auto or manual
Tool kafka-leader-election.sh PREFERRED type

Example: Run preferred election

kafka-leader-election.sh --election-type PREFERRED \
  --all-topic-partitions --bootstrap-server localhost:9092

10. Understanding Replica Fetcher

Aspect Description Detail
ReplicaFetcherThread Pulls data from leaders Per broker pair
num.replica.fetchers Parallel fetcher count Scale with load
Throttling Limit during reassignment replica.alter.log.dirs

Example: More fetcher threads

num.replica.fetchers=4

11. Manual Replica Reassignment

Step Command Detail
Plan JSON with new replicas Per partition
Execute --execute Apply + throttle
Verify --verify Confirm complete

Example: Reassign replicas

{"version":1,"partitions":[
  {"topic":"critical","partition":0,"replicas":[2,3,4]},
  {"topic":"critical","partition":1,"replicas":[3,4,5]}
]}

Related Cheat Sheets

Kafka

Implementing Stream Joins

Kafka

Managing Offsets

Kafka

Using Kafka CLI Tools

Kafka

Implementing Multi-Datacenter Replication