Closed wyukawa closed 6 years ago
Thanks for the report.
I didn't check yet deeply but I assume fluency(in kafka-fluentd-consumer) doesn't set size
option in the payload. It seems to cause ensure_unpacked
call in size
for buffer chunk.
I will test the performance with/without size
option in forward protocol.
I tested dummy plugin with/without size
.
With size
, CPU usage is 1%.
Without size
, CPU usage is 4 - 5%.
If the message body is large, it has more impact.
So support size in fluency, may resolve the problem.
Release kafka-fluentd-consumer v0.3.2: https://github.com/treasure-data/kafka-fluentd-consumer/releases/tag/v0.3.2 This problem should be resolved.
Check CONTRIBUTING guideline first and here is the list to help us investigate the problem.
gemfile
conf
If you have any comments, it would be nice.
data flow is here kafka -> kafka-fluentd-consumer -> this fluend -> fluent-plugin-elasticsearch -> elasticsearch
fluentd v1 stackprof
v0.12 environemnt is the following
CentOS 6.6 Ruby 2.1
gemfile
conf