we have a kafka cluster with 3 brokers on GKE(m/c type - e2-standard-8 i.e. 8 vCPU, 32GB memory). Each Broker has space(SSD) of 500GB, and it is current 17% full i.e. space is not an issue.
We are getting data from external Producer, and it seems some data is not being received (i.e. Producer is not receiving ack from the Broker resulting in kafka transport error)
Is there way to determine throughput (i.e. volume of data/min or sec) that can be handled by the kafka cluster ?
Another question : On the Observability tab, I see CPU & Memory hiking up frequently(to more than 100%), can that be one reason for this ?
tia!
The Kafka Java process should not be configured to more than 6GB and hence please verify your JMX configuration.
Kafka writes will be faster only if data is kept on the OS page cache because the replication will be faster when the data is present on the page cache rather than seeking from disk.
The disk reads should be zero please verify the same from disk monitoring stats.