Unusual(?) behaviour in mirror maker cluster

[Max Weaver <ma...@infectiousmedia.com>]

Hi,

We've deployed an EU based mirror maker cluster to consume topics from
clusters situated in the US.

We have five brokers in the mm cluster, but it seems to be that one
particular one consistently has higher throughput than all the others
(which stay around the same).

We are also having issues with lag, especially with certain topics. The
mirror maker is hosted on the target cluster and our producer / consumer
settings look like this:

producer.properties:

bootstrap.servers=localhost:9092
acks=1
retries=2147483647
client.id=mm.glb.data.producer

consumer.properties:

bootstrap.servers=us-broker-1:9092,...:9092
group.id=mm_data_production
exclude.internal.topics=true
client.id=mm.us1.data.consumer
partition.assignment.strategy=org.apache.kafka.clients.consumer.RoundRobinAssignor

The mirror maker itself is running as a service, which looks like:

[Unit]
Description=Kafka-mm
After=network.target

[Service]
User=kafka
Group=kafka
Environment=KAFKA_HEAP_OPTS=-Xmx1G
SyslogIdentifier=kafka-mm
ExecStart=/opt/kafka/bin/kafka-run-class.sh kafka.tools.MirrorMaker
--consumer.config /opt/kafka/config/consumer.properties --num.streams 24
--producer.config /opt/kafka/config/producer.properties --whitelist="*"
Restart=on-failure

[Install]
WantedBy=multi-user.target


each topic has 24 partitions and a replication factor of 2. There are 9
topics in total.

Perhaps someone could come up with an idea why this is happening?

Many thanks

Max