Higher latency for lower CPU allocation

[ghost]

A thing that has been concerning to us because of the hardware running our services, was the ultra-high CPU allocation by Redpanda.

Redpanda was hitting 125%-200% allocation as soon as the consumers were connecting to it, thus hindering the operation of our services running alongside it.

I was looking for the configuration that would allow us to sacrifice some latency which would be a zero problem for our services; after posting on redpandacommunity.slack.com @BenPope provided the missing configuration. To sacrifice 50 milliseconds of latency and have the CPU allocation going down to 2%-4% (for our hardware):
,
rpk cluster config set fetch_reads_debounce_timeout 50

Thank you @BenPope, @emaxerrno for the suppport

[emaxerrno]

cc: @r-vasquez considering setting this higher for developer_mode
cc: @twmb and @vladoschreiner