Why is replacement not a good idea in zookeeper and kafka?

Question

I read the instructions on

do not use swap

both on zookeeper and on kafka. I know that kafka depends on pagecaching to keep parts of sequential logs in cache in memory, even if they are written to disk.

But I can not understand how swapping can harm zk and kafka.

Answer 1

Swapping can cause performance issues as well as stability issues; in your example, you do not want the Linux kernel to "erroneously / accidentally" replace your Kafka or ZooKeeper processes.

, JVM, Kafka ZooKeeper, :

[The] JVM GC , , , , . ( , , ), . GC , , , , ; , GC !

, , vm.swappiness 0, , RHEL 6.5, 1 ( 0 ), , .

. , Hadoop Elasticsearch, , Kafka ZooKeeper:

• Hadoop: Apache Hadoop ( OOM) ; Hadoop Spotify.
• Elasticsearch: Elasticsearch.