As mentioned at the beginning of this section, compaction can consume a large amount of resources if left unchecked. A symptom of this would be an application experiencing query timeouts or consistency failures. When investigating the offending node, you may find that the CPU usage is maxed out, and several compactions (nodetool compactionstats) may be pending. A common way to remedy this problem is to lower the compaction throughput, as follows:
nodetool getcompactionthroughputCurrent compaction throughput: 256 MB/s
Woah! Allowing compaction to consume 256 Mbps (megabits per second) is a sure way to saturate your instance's resources. Let's lower that, and monitor our node's progress: ...