Analyze event logs using Fluentd and Elasticsearch

Fluentd is a flexible and robust event log collector, but Fluentd doesn’t have own data-store and Web UI. If you want to analyze the event logs collected by Fluentd, then you can use Elasticsearch and Kibana :)

Elasticsearch is an easy to use Distributed Search Engine and Kibana is an awesome Web front-end for Elasticsearch.

Setup

I tested on Mac OS X.

Pre requirements

Java for Elasticsearch

Use Mac OS X’s Java.

% java -version
java version "1.6.0_51"
Java(TM) SE Runtime Environment (build 1.6.0_51-b11-457-11M4509)
Java HotSpot(TM) 64-Bit Server VM (build 20.51-b01-457, mixed mode)

Ruby for Fluentd

In this article, I use my rbenv’s Ruby and Fluentd gem directly.

% ruby --version
ruby 1.9.3p484 (2013-11-22 revision 43786) [x86_64-darwin12.4.0]

Note that Fluentd doesn’t work on Ruby 1.8.

Treasure Agent (td-agent)

Treasure Agent, as known as td-agent, is a stable distribution which consists of Fluentd, popular plugins and own Ruby processor. Many users use td-agent instead of Fluentd gem. See this FAQ.

Elasticsearch

Downlod and extract the latest package.

% curl -O https://download.elasticsearch.org/elasticsearch/elasticsearch/elasticsearch-0.90.11.tar.gz
% tar zxvf elasticsearch-0.90.11.tar.gz
% cd elasticsearch-0.90.11/

I use version v0.90.11 but v1.0 will be released soon!

Start Elasticsearch:

% ./bin/elasticsearch -f
[2014-02-12 17:51:32,645][INFO ][node                     ] [Powerhouse] version[0.90.11], pid[79737], build[11da1ba/2014-02-03T15:27:39Z]
[2014-02-12 17:51:32,646][INFO ][node                     ] [Powerhouse] initializing ...
[2014-02-12 17:51:32,651][INFO ][plugins                  ] [Powerhouse] loaded [], sites []
[2014-02-12 17:51:34,319][INFO ][node                     ] [Powerhouse] initialized
[2014-02-12 17:51:34,320][INFO ][node                     ] [Powerhouse] starting ...
[2014-02-12 17:51:34,395][INFO ][transport                ] [Powerhouse] bound_address { inet[/0:0:0:0:0:0:0:0%0:9300]}, publish_address { inet[/192.168.1.112:9300]}
[2014-02-12 17:51:37,448][INFO ][cluster.service          ] [Powerhouse] new_master [Powerhouse][tL1IC8xHSCudeVsFt4JFsQ][inet[/192.168.1.112:9300]], reason: zen-disco-join (elected_as_master)
[2014-02-12 17:51:37,481][INFO ][discovery                ] [Powerhouse] elasticsearch/tL1IC8xHSCudeVsFt4JFsQ
[2014-02-12 17:51:37,492][INFO ][http                     ] [Powerhouse] bound_address { inet[/0:0:0:0:0:0:0:0%0:9200]}, publish_address { inet[/192.168.1.112:9200]}
[2014-02-12 17:51:37,493][INFO ][node                     ] [Powerhouse] started
[2014-02-12 17:51:37,505][INFO ][gateway                  ] [Powerhouse] recovered [0] indices into cluster_state

Kibana

Latest Kibana, called Kibana 3, consists of only HTML and JavaScript, so setup is very easy:

% curl -O https://download.elasticsearch.org/kibana/kibana/kibana-3.0.0milestone5.tar.gz
% tar zxvf kibana-3.0.0milestone5.tar.gz
% cd kibana-3.0.0milestone5/

Open index.html in the kibana directory:

% open index.html

If you want to change Kibana configuration, please edit config.js, e.g. change elasticsearch URL.

Fluentd

Install Fluentd gem and Elasticsearch plugin:

% gem install fluentd fluent-plugin-elasticsearch

Fluentd configuration is below:

# es.conf
<source>
  type forward
</source>

<match es.**>
  type elasticsearch
  logstash_format true
  flush_interval 5s # 5s for testing. On production environment, 60s or higher is better
</match>

fluent-plugin-elasticsearch provides logstash_format option. It enables Kibana to analyze the event logs with day based indexes.

Start Fluentd:

% fluentd -c es.conf
2014-02-12 18:43:31 +0900 [info]: starting fluentd-0.10.43
2014-02-12 18:43:31 +0900 [info]: reading config file path = "es.conf"
2014-02-12 18:43:32 +0900 [info]: gem 'fluent-plugin-elasticsearch' version '0.2.0'
2014-02-12 18:43:32 +0900 [info]: gem 'fluentd' version '0.10.43'
2014-02-12 18:43:32 +0900 [info]: using configuration file: <ROOT>
  <source>
    type forward
  </source>
  <match es.**>
    type elasticsearch
    logstash_format true
    flush_interval 5s
  </match>
</ROOT>
2014-02-12 18:43:32 +0900 [info]: adding source type = "forward"
2014-02-12 18:43:32 +0900 [info]: adding match pattern = "es.**" type = "elasticsearch"
2014-02-12 18:43:32 +0900 [info]: listening fluent socket on 0.0.0.0:24224

Analyze event logs

Send some events to Fluentd. fluent-cat is an utility command to send json text to Fluentd’s in_forward plugin.

% echo '{"message":"D"}' | fluent-cat es.event # es.event is a tag. es.event matches es.** of <match>
% echo '{"message":"Ruby"}' | fluent-cat es.event
% echo '{"message":"Elasticsearch"}' | fluent-cat es.event
% echo '{"message":"msgpack"}' | fluent-cat es.event
% ...

After Fluentd flushed received events to Elasticsearch, you can analyze the event logs via Kibana! Following image is one panel example:

"Fluentd, Elasticsearch and Kibana"

Kibana has some built-in panels, so you can create own dashboard easily. See Kibana demo

Advanced tips

If your service has a high traffic, then fluent-plugin-elasticsearch sometimes get stucked. In this case, built-in out_roundrobin plugin is useful. You can distribute a write request to elasticsearch nodes for load balancing.

Of course, putting Queue with multiple fluentd nodes is an another approach.

Conclusion

This article introduced Fluentd, Elasticsearch and Kibana combination to analyze the event logs. These components are easy to setup and work fine, so you can try this framework soon! I heard many companies have already tried / deployed this setup on production :)

Enjoy Fluentd and Elasticsearch!

Go ahead!

Memoization for Everything