tail

The in_tail Input plugin allows Fluentd to read events from the tail of text files. Its behavior is similar to the tail -F command.

It is included in Fluentd's core.

Example Configuration

<source>
  @type tail
  path /var/log/httpd-access.log
  pos_file /var/log/td-agent/httpd-access.log.pos
  tag apache.access
  <parse>
    @type apache2
  </parse>
</source>

Refer to the Configuration File article for the basic structure and syntax of the configuration file.

For <parse>, see Parse Section.

How It Works

When Fluentd is first configured with in_tail, it will start reading from the tail of that log, not the beginning. Once the log is rotated, Fluentd starts reading the new file from the beginning. It keeps track of the current inode number.

If td-agent restarts, it resumes reading from the last position before the restart. This position is recorded in the position file specified by the pos_file parameter.

Linux Capability

Since v1.12.0, in_tail handles the following Linux capabilities if Fluentd's Linux capability handling module is enabled:

  • CAP_DAC_READ_SEARCH (:dac_read_search on in_tail code.)

  • CAP_DAC_OVERRIDE (:dac_override on in_tail code.)

Plugin Helpers

See also: Linux capability

Parameters

See Common Parameters.

@type (required)

The value must be tail.

tag

type
default
version

string

required parameter

0.14.0

The tag of the event.

* can be used as a placeholder that expands to the actual file path, replacing '/' with '.'.

With the following configuration:

path /path/to/file
tag foo.*

in_tail emits the parsed events with the foo.path.to.file tag.

path

type
default
version

string

required parameter

0.14.0

The path(s) to read. Multiple paths can be specified, separated by comma ','.

* and strftime format can be included to add/remove the watch file dynamically. At the interval of refresh_interval, Fluentd refreshes the list of watch files.

path /path/to/%Y/%m/%d/*

For multiple paths:

path /path/to/a/*,/path/to/b/c.log

If the date is 20140401, Fluentd starts to watch the files in /path/to/2014/04/01 directory. See also read_from_head parameter.

Using ** for path globbing is supported.

path /path/to/**/some.log

By default, You should not use * with log rotation because it may cause the log duplication. To avoid log duplication, you need to set follow_inodes true in the configuration.

If you want to use other glob patterns such as [] and ?, you need to set up glob_policy extended as described in the glob_policy section.

path_timezone

type
default
version

string

nil

1.8.1

This parameter is for strftime formatted path like /path/to/%Y/%m/%d/.

in_tail uses system timezone by default. This parameter overrides it:

path_timezone "+00"

For timezone format, see Timezone Section.

glob_policy

type
default
available values
version

enum

backward_compatible

backward_compatible/extended/always

1.17.0

This parameter permits to extend glob patterns on path and exclude_path parameters. When specifying extended, users can use [] and ? in glob patterns. When specifying always, users can use [], ?, and additionally {} in glob patterns.

However, always option is not able to use with the default value of path_delimiter. When using the default value of path_delimiter, it will be marked as Fluent::ConfigError.

exclude_path

type
default
version

array

[] (empty)

0.14.0

The paths excluded from the watcher list.

For example, to remove the compressed files, you can use the following pattern:

path /path/to/*
exclude_path ["/path/to/*.gz", "/path/to/*.zip"]

exclude_path takes input as an array, unlike path which takes as a string.

follow_inodes

type
default
version

bool

false

1.12.0

Avoid to read rotated files duplicately. You should set true when you use * or strftime format in path.

path /path/to/*
read_from_head true
follow_inodes true  # Without this parameter, file rotation causes log duplication.

refresh_interval

type
default
version

time

60 (seconds)

0.14.0

The interval to refresh the list of watch files. This is used when the path includes *.

limit_recently_modified

type
default
version

time

nil (disabled)

0.14.13

Limits the watching files that the modification time is within the specified time range when using * in path.

skip_refresh_on_startup

type
default
version

bool

false

0.14.13

Skips the refresh of the watch list on startup. This reduces the startup time when * is used in path.

read_from_head

type
default
version

bool

false

0.14.0

Starts to read the logs from the head of the file or the last read position recorded in pos_file, not tail.

Notes:

  • in_tail tries to read a file during the startup phase when this is true. So that if the target file is too large and takes a long time to read it, other plugins are blocked to start until the reading is finished. You can avoid it by skip_refresh_on_startup.

  • For Fluentd <= v1.14.2: If you use * or strftime format as path and new files may be added into such paths while tailing, you should set this parameter to true. Otherwise some logs in newly added files may be lost. On the other hand you should guarantee that the log rotation will not occur in * directory in that case to avoid log duplication. Or you can use follow_inodes true to avoid such log duplication, which is available as of v1.12.0.

    • From Fluentd v1.14.3, in_tail reads newly added files from head automatically even if read_from_head is false. read_from_head false is affected only on start up.

encoding, from_encoding

type
default
version

string

nil (string encoding is ASCII-8BIT)

0.14.0

Specifies the encoding of reading lines.

By default, in_tail emits string value as ASCII-8BIT encoding.

These options change it:

  • If encoding is specified, in_tail changes string to encoding.

    This uses Ruby's String#force_encoding.

  • If encoding and from_encoding both are specified, in_tail tries to

    encode string from from_encoding to encoding. This uses Ruby's

    String#encode.

You can get the list of supported encodings with this command:

$ ruby -e 'p Encoding.name_list.sort'

Caution: From v0.14.12 to v1.18.x, there was a bug.

  • You need to specify both encoding and from_encoding.

  • If you specify only encoding, String#encode will be executed unintentionally. It can break the data.

  • To change only the encoding info as metadata, without transforming the string data itself (String#force_encoding), you need to specify the same encoding for both encoding and from_encoding.

read_lines_limit

type
default
version

integer

1000

0.14.0

The number of lines to read with each I/O operation.

If you see chunk bytes limit exceeds for an emitted event stream or similar log with in_tail, set a smaller value.

read_bytes_limit_per_second

type
default
version

size

-1 (unlimited)

1.13.0

The number of reading bytes per second to read with I/O operation.

This value should be equal or greater than 8192.

If you work with a big cluster with high volume of log, you can use this parameter to avoid network saturation and make it easier to calculate the max throughput per node. To restrict shipping log volumes per second, set a positive number.

max_line_size

type
default
version

size

nil

1.14.4

The maximum length of a line. Longer lines than it will be just skipped.

If you see BufferChunkOverflowError exception frequently, it means that incoming data is too long. If such a long line is unexpected incoming data and want to ignore it, then set a smaller value than chunk_limit_size in <buffer> section.

multiline_flush_interval

type
default
version

time

nil (disabled)

0.14.0

The interval of flushing the buffer for multiline format.

If you set multiline_flush_interval 5s, in_tail flushes buffered event after 5 seconds from last emit. This option is useful when you use format_firstline option.

pos_file (highly recommended)

type
default
version

string

nil

0.14.0

Fluentd will record the position it last read from this file:

pos_file /var/log/td-agent/tmp/access.log.pos

pos_file handles multiple positions in one file so no need to have multiple pos_file parameters per source.

Don't share pos_file between in_tail configurations. It causes unexpected behavior e.g. corrupt pos_file content.

in_tail removes the untracked file position at startup. It means that the content of pos_file keeps growing until a restart when you tail lots of files with the dynamic path setting. This issue can be solved by using pos_file_compaction_interval.

pos_file_compaction_interval

type
default
version

time

nil

1.9.2

The interval of doing compaction of pos file.

The targets of compaction are unwatched, unparsable, and the duplicated line. You can use this value when pos_file option is set:

pos_file /var/log/td-agent/tmp/access.log.pos
pos_file_compaction_interval 72h

<parse> Directive (required)

The format of the log.

in_tail uses the parser plugin to parse the log. See parser for more detail.

Examples:

# json
<parse>
  @type json
</parse>

# regexp
<parse>
  @type regexp
  expression ^(?<name>[^ ]*) (?<user>[^ ]*) (?<age>\d*)$
</parse>

If @type contains multiline, in_tail works in multiline mode.

format

Deprecated parameter. Use <parse> instead.

path_key

type
default
version

string

nil (no assign)

0.14.0

Adds the watching file path to the path_key field.

With this configuration:

path /path/to/access.log
path_key tailed_path

The generated events are like this:

{"tailed_path":"/path/to/access.log","k1":"v1",...,"kN":"vN"}

rotate_wait

type
default
version

time

5 (seconds)

0.14.0

in_tail actually does a bit more than tail -F itself. When rotating a file, some data may still need to be written to the old file as opposed to the new one.

in_tail takes care of this by keeping a reference to the old file (even after it has been rotated) for some time before transitioning completely to the new file. This helps prevent data designated for the old file from getting lost. By default, this time interval is 5 seconds.

The rotate_wait parameter accepts a single integer representing the number of seconds you want this time interval to be.

enable_watch_timer

type
default
version

bool

true

0.14.0

Enables the additional watch timer. Setting this parameter to false will significantly reduce CPU and I/O consumption when tailing a large number of files on systems with inotify support. The default is true which results in an additional 1 second timer being used.

in_tail (via Cool.io) uses inotify on systems which support it. Earlier versions of libev on some platforms (e.g. macOS) did not work properly; therefore, an explicit 1 second timer was used. Even on systems with inotify support, this results in additional I/O each second, for every file being tailed.

Early testing demonstrates that modern Cool.io and in_tail work properly without the additional watch timer. In the future, depending on the feedback and testing, the additional watch timer may be disabled by default.

enable_stat_watcher

type
default
version

bool

true

1.0.1

Enables the additional inotify-based watcher. Setting this parameter to false will disable the inotify events and use only timer watcher for file tailing.

This option is mainly for avoiding the stuck issue with inotify.

open_on_every_update

type
default
version

bool

false

0.14.12

Opens and closes the file on every update instead of leaving it open until it gets rotated.

emit_unmatched_lines

type
default
version

bool

false

0.14.12

Emits unmatched lines when <parse> format is not matched for incoming logs.

Emitted record is {"unmatched_line" : incoming line}, e.g. {"unmatched_line" : "Non JSON format!"}.

ignore_repeated_permission_error

type
default
version

bool

false

0.14.0

If you have to exclude the non-permission files from the watch list, set this parameter to true. It suppresses the repeated permission error logs.

@log_level

The @log_level option allows the user to set different levels of logging for each plugin. The supported log levels are: fatal, error, warn, info, debug, and trace.

Refer to the Logging for more details.

<group> Section

The in_tail plugin can assign each log file to a group, based on user defined rules. The limit parameter controls the total number of lines collected for a group within a rate_period time interval.

Example:

# group rules -- 1
<group>
  rate_period 5s

  <rule>
    match {
        "namespace": "/shopping/",
        "podname": "/frontend/",
    }
    limit 1000
  </rule>
</group>

# group rules -- 2
<group>
  <rule>
    match {
        directory: /payment/
    }
    limit 2000
  </rule>
</group>

pattern

type
default
version

regexp

/^\/var\/log\/containers\/(?<podname>[a-z0-9]([-a-z0-9]*[a-z0-9])?(\/[a-z0-9]([-a-z0-9]*[a-z0-9])?)*)_(?<namespace>[^_]+)_(?<container>.+)-(?<docker_id>[a-z0-9]{64})\.log$/

1.15

Specifies the regular expression for extracting metadata (namespace, podname) from log file path. Default value of the pattern regexp extracts information about namespace, podname, docker_id, container of the log (K8s specific).

You can also add custom named captures in pattern for custom grouping of log files. For example,

    pattern /^\/home\/logs\/(?<file>.+)\.log$/

In this example, filename will be extracted and used to form groups.

rate_period

type
default
version

time

60 (seconds)

1.15

Time period in which the group line limit is applied. in_tail resets the counter after every rate_period interval.

<rule> Section (required)

Grouping rules for log files.

match

type
default
version

hash

{"namespace": "/./", "podname": "/./"}

1.15

match parameter is used to check if a file belongs to a particular group based on hash keys (named captures from pattern) and hash values (regexp in string)

limit

type
default
version

integer

-1

1.15

Maximum number of lines allowed from a group in rate_period time interval. The default value of -1 doesn't throttle log files of that group.

Learn More

FAQ

What happens when <parse> type is not matched for logs?

in_tail prints warning message. For example, if you specify @type json in <parse> and your log line is 123,456,str,true, then you will see following message in fluentd logs:

2018-04-19 02:23:44 +0900 [warn]: #0 pattern not match: "123,456,str,true"

See also emit_unmatched_lines parameter.

in_tail doesn't start to read the log file, why?

in_tail follows tail -F command's behavior by default, so in_tail reads only the new logs. If you want to read the existing lines for the batch use case, set read_from_head true.

in_tail shows /path/to/file unreadable log message. Why?

If you see this message:

/path/to/file unreadable. It is excluded and would be examined next time.

It means that fluentd does not have read permission for /path/to/file. Check your fluentd and target files permission.

Note: When td-agent is launched by systemd, the default user of the td-agent process is the td-agent user. You must ensure that this user has read permission to the tailed /path/to/file. For instance, on Ubuntu, the default Nginx access file /var/log/nginx/access.log is mode 0640 and owned by www-data:adm. In this case, several options are available to allow read access:

  1. Add the td-agent user to the adm group, e.g. through usermod -aG, or

  2. Use the cap_dac_read_search capability to allow the invoking user to read the file without otherwise changing its permission bits or ownership.

A bug exists in Fluentd 1.13.x where it may suppress warning logs about unreadable files. (See Fluentd PR #3478.)

logrotate Setting

logrotate has the nocreate parameter and it does not create a new file if log rotation is triggered. It means in_tail cannot find the new file to tail.

This parameter does not fit the typical application log use cases, so check your logrotate setting which does not include the nocreate parameter.

What happens when in_tail receives BufferOverflowError?

in_tail stops reading the new lines and pos file updates until BufferOverflowError is resolved. After resolving BufferOverflowError, resume emitting new lines and pos file updates.

in_tail is sometimes stopped when monitor lots of files. How to avoid it?

Try to set enable_stat_watcher false in in_tail setting. We got several reports that in_tail is stopped when * is included in path, and the problem is resolved by disabling the inotify events.

Wildcard pattern in path does not work on Windows, why?

The backslash (\) with * does not work on Windows by internal limitations. To avoid this, use slash style instead:

# good
path C:/path/to/*/foo.log

# bad
path C:\\path\\to\\*\\foo.log

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

What happens when a file can be assigned to more than one group?

Example,

  <rule> ## Rule1
    match {
      namespace: /monitoring/
    }
    limit 100
  </rule>

  <rule> ## Rule2
    match {
      namespace: /monitoring/,
      podname: /logger/,
    }
    limit 2000
  </rule>

In this case, rules with more constraints, i.e., greater number of match hash keys will be given a higher priority. So a file will be assigned to Rule2 if it can be assigned to both Rule1 and Rule2.

PreviousInput PluginsNextforward

Last updated

Was this helpful?