Kafka

Understanding how it works

Kafka Introduction - official website
Confluent Tutorial - step-by-step tutorial

Documentation

Notes

Consumer API is not thread-safe
Be always mindful about event ordering
Evaluate early your scalability capabilities (how many partitions max your Kafka platform team allows, what are my alternative levers to scale)
Records are supposed to be lightweight payloads, the default max size is 1MB, it’s not a temporary data storage for
large payloads
Kafka is performance-oriented but can easily be deteriorated by introducing API calls in the event processing part

Tools

Confluent Platform - contains all the tools to consume, produce, and manage topics, offsets, etc…
kcat - more accessible tool to start consuming/producing Kafka records

Consume:


kcat -C -b $KAFKA_CLUSTER_URL \
    -t $TOPIC_NAME \
    -X security.protocol=SASL_SSL \
    -X sasl.mechanisms=SCRAM-SHA-512 \
    -X sasl.username=$SASL_USERNAME \
    -X sasl.password=$SASL_PASSWORD \
    -c 1 \
    -e

To consume the last 100 messages for partition 2:


    -p 2 \
    -o-100 \

To support Avro messages, add:


    -s value=avro \
    -r https://$SASL_USERNAME:$SASL_PASSWORD@$SCHEMA_REGISTRY_URL \

Produce:


kcat -P -F $HOME/path/to/producer.properties -t $TOPIC_NAME -l $HOME/path/to/data

Configuration file - producer.properties


bootstrap.servers=$KAFKA_CLUSTER_URL
sasl.mechanism=SCRAM-SHA-512
security.protocol=SASL_SSL
sasl.jaas.config=org.apache.kafka.common.security.scram.ScramLoginModule required username="$SASL_USERNAME" password="$SASL_PASSWORD";
client.id=my-test-kafka-producer
# schema.registry.url=https://schema-registry.endpoint.com
# basic.auth.credentials.source=SASL_INHERIT
# auto.register.schemas=false

Reset offset

Poison pill, skip data by moving all offsets to the latest:


kafka-consumer-groups --bootstrap-server $KAFKA_CLUSTER_URL --command-config $HOME/path/to/admin.properties --group $CONSUMER_GROUP_NAME --topic $TOPIC_NAME --reset-offsets --to-latest --execute

Move back your consumer group to a specific offset $OFFSET for partition $PARTITION_NUM


kafka-consumer-groups.sh \
    --command-config $HOME/path/to/admin.properties \
    --bootstrap-server $KAFKA_CLUSTER_URL \
    --group $CONSUMER_GROUP_NAME \
    --topic $TOPIC_NAME:$PARTITION_NUM \
    --reset-offsets --to-offset $OFFSET --execute

admin.properties is similar to producer.properties above except it doesn’t require any client ID.