This article describes Fluentd's logging mechanism.

Fluentd has two log layers: global and per plugin. Different log levels can be set for global logging and plugin level logging.

Log Level

Shown below is the list of supported values, in increasing order of verbosity:

• fatal

• error

• warn

• info

• debug

• trace

The default log level is info, and Fluentd outputs info, warn, error and fatal logs by default.

Global Logs

Global logging is used by Fluentd core and plugins that don't set their own log levels. The global log level can be adjusted up or down.

By Command Line Option

Increase Verbosity Level

The -v option sets the verbosity to debug while the -vv option sets the verbosity to trace.

$ fluentd -v  ... # debug level
$ fluentd -vv ... # trace level

These options are useful for debugging purposes.

Decrease Verbosity Level

The -q option sets the verbosity to warn while the -qq option sets the verbosity to error.

$ fluentd -q  ... # warn level
$ fluentd -qq ... # error level

By Config File

You can also change the logging level with <system> section in the config file like below.

<system>
  # equal to -qq option
  log_level error
</system>

Per Plugin Log

The @log_level option sets different levels of logging for each plugin. It can be set in each plugin's configuration file.

For example, in order to debug in_tail but suppress all but fatal log messages for in_http, their respective @log_level options should be set as follows:

<source>
  @type tail
  @log_level debug
  path /var/log/data.log
  ...
</source>
<source>
  @type http
  @log_level fatal
</source>

If you don't specify the @log_level parameter, the plugin will use the global log level.

Log format

text and json are supported. The default is text. The format can be changed via <log> directive in <system>.

<system>
  <log>
    format json
    time_format %Y-%m-%d
  </log>
</system>

With this setting:

2017-07-27 06:44:54 +0900 [info]: #0 fluentd worker is now running worker=0

This text log line is changed to:

{"time":"2017-07-27","level":"info","message":"fluentd worker is now running worker=0","worker_id":0}

Output to log file

Fluentd outputs logs to STDOUT by default. To output to a file instead, please specify the -o option.

$ fluentd -o /path/to/log_file

Log rotation setting

Fluentd doesn't rotate logs by default. You can configure this behaviour via command line options:

--log-rotate-age AGE

AGE is integer or string:

• integer: Generations to keep rotated log files.

• string: frequency of rotation. daily, weekly and monthly are

  supported

--log-rotate-size BYTES

The byte size to rotate log files. This is applied when --log-rotate-age is specified with integer.

Here is an example:

$ fluentd -c fluent.conf --log-rotate-age 5 --log-rotate-size 104857600

Capture Fluentd logs

Fluentd marks its own logs with the fluent tag. You can process Fluentd logs by using <match fluent.**>(Of course, ** captures other logs) in <label @FLUENT_LOG>. If you define <label @FLUENT_LOG> in your configuration, then Fluentd will send its own logs to this label. This is useful for monitoring Fluentd logs.

For example, if you have the following configuration:

# omit other source / match
<label @FLUENT_LOG>
  <match fluent.*>
    @type stdout
  </match>
</label>

then Fluentd outputs fluent.info logs to stdout like below:

2014-02-27 00:00:00 +0900 [info]: shutting down fluentd
2014-02-27 00:00:01 +0900 fluent.info: {"message":"shutting down fluentd"} # by <match fluent.*>
2014-02-27 00:00:01 +0900 [info]: process finished code = 0

Case1: Send Fluentd logs to monitoring service

You can send Fluentd logs to a monitoring service by plugins, e.g. datadog, sentry, irc, etc.

# Add hostname for identifying the server
<label @FLUENT_LOG>
  <filter fluent.*>
    @type record_transformer
    <record>
      host "#{Socket.gethostname}"
    </record>
  </filter>
​
  <match fluent.*>
    @type monitoring_plugin
    # parameters...
  </match>
<label>

Case2: Use aggregation/monitoring server

You can use out_forward to send Fluentd logs to a monitoring server. The monitoring server can then filter and send the logs to your notification system: chat, irc, etc.

Leaf server example:

# Add hostname for identifying the server and tag to filter by log level
<label @FLUENT_LOG>
  # If you want to capture only error events, use 'fluent.error' instead.
  <filter fluent.*>
    @type record_transformer
    <record>
      host "#{Socket.gethostname}"
      original_tag ${tag}
    </record>
  </filter>
​
  <match fluent.*>
    @type forward
    <server>
      # Monitoring server parameters
    </server>
  </match>
<label>

Monitoring server example:

<source>
  @type forward
</source>
​
# Ignore trace, debug and info log. Of course, you can use strict matching like `<filter fluent.{warn,error,fatal}>` without grep filter.
<filter fluent.*>
  @type grep
  <regexp>
    key original_tag
    pattern fluent.(warn|error|fatal)
  </regexp>
</filter>
​
<match fluent.*>
  # your notification setup. This example uses irc plugin
  @type irc
  host irc.domain
  channel notify
  message notice: %s [%s] @%s %s
  out_keys original_tag,time,host,message
</match>
​
# Unlike v0.12, if `<label @FLUENT_LOG>` is defiend, `<match fluent.*>` in root is not used for log capturing.
<label @FLUENT_LOG>
  # Monitoring server's internal log
</label>

If an error occurs, you will get a notification message in your Slack notify channel.

01:01  fluentd: [11:10:24] notice: fluent.warn [2014/02/27 01:00:00] @leaf.server.domain detached forwarding server 'server.name'

You can still use v0.12 way without <label @FLUENT_LOG>.

If this article is incorrect or outdated, or omits critical information, please let us know. Fluentd is a open source project under Cloud Native Computing Foundation (CNCF). All components are available under the Apache 2 License.