File input plugin

edit
  • Plugin version: v4.1.18
  • Released on: 2020-04-30
  • Changelog

For other versions, see the Versioned plugin docs.

Getting Help

edit

For questions about the plugin, open a topic in the Discuss forums. For bugs or feature requests, open an issue in Github. For the list of Elastic supported plugins, please consult the Elastic Support Matrix.

Description

edit

Stream events from files, normally by tailing them in a manner similar to tail -0F but optionally reading them from the beginning.

Normally, logging will add a newline to the end of each line written. By default, each event is assumed to be one line and a line is taken to be the text before a newline character. If you would like to join multiple log lines into one event, you’ll want to use the multiline codec. The plugin loops between discovering new files and processing each discovered file. Discovered files have a lifecycle, they start off in the "watched" or "ignored" state. Other states in the lifecycle are: "active", "closed" and "unwatched"

By default, a window of 4095 files is used to limit the number of file handles in use. The processing phase has a number of stages:

  • Checks whether "closed" or "ignored" files have changed in size since last time and if so puts them in the "watched" state.
  • Selects enough "watched" files to fill the available space in the window, these files are made "active".
  • The active files are opened and read, each file is read from the last known position to the end of current content (EOF) by default.

In some cases it is useful to be able to control which files are read first, sorting, and whether files are read completely or banded/striped. Complete reading is all of file A then file B then file C and so on. Banded or striped reading is some of file A then file B then file C and so on looping around to file A again until all files are read. Banded reading is specified by changing file_chunk_count and perhaps file_chunk_size. Banding and sorting may be useful if you want some events from all files to appear in Kibana as early as possible.

The plugin has two modes of operation, Tail mode and Read mode.

Tail mode

edit

In this mode the plugin aims to track changing files and emit new content as it’s appended to each file. In this mode, files are seen as a never ending stream of content and EOF has no special significance. The plugin always assumes that there will be more content. When files are rotated, the smaller or zero size is detected, the current position is reset to zero and streaming continues. A delimiter must be seen before the accumulated characters can be emitted as a line.

Read mode

edit

In this mode the plugin treats each file as if it is content complete, that is, a finite stream of lines and now EOF is significant. A last delimiter is not needed because EOF means that the accumulated characters can be emitted as a line. Further, EOF here means that the file can be closed and put in the "unwatched" state - this automatically frees up space in the active window. This mode also makes it possible to process compressed files as they are content complete. Read mode also allows for an action to take place after processing the file completely.

In the past attempts to simulate a Read mode while still assuming infinite streams was not ideal and a dedicated Read mode is an improvement.

Tracking of current position in watched files

edit

The plugin keeps track of the current position in each file by recording it in a separate file named sincedb. This makes it possible to stop and restart Logstash and have it pick up where it left off without missing the lines that were added to the file while Logstash was stopped.

By default, the sincedb file is placed in the data directory of Logstash with a filename based on the filename patterns being watched (i.e. the path option). Thus, changing the filename patterns will result in a new sincedb file being used and any existing current position state will be lost. If you change your patterns with any frequency it might make sense to explicitly choose a sincedb path with the sincedb_path option.

A different sincedb_path must be used for each input. Using the same path will cause issues. The read checkpoints for each input must be stored in a different path so the information does not override.

Files are tracked via an identifier. This identifier is made up of the inode, major device number and minor device number. In windows, a different identifier is taken from a kernel32 API call.

Sincedb records can now be expired meaning that read positions of older files will not be remembered after a certain time period. File systems may need to reuse inodes for new content. Ideally, we would not use the read position of old content, but we have no reliable way to detect that inode reuse has occurred. This is more relevant to Read mode where a great many files are tracked in the sincedb. Bear in mind though, if a record has expired, a previously seen file will be read again.

Sincedb files are text files with four (< v5.0.0), five or six columns:

  1. The inode number (or equivalent).
  2. The major device number of the file system (or equivalent).
  3. The minor device number of the file system (or equivalent).
  4. The current byte offset within the file.
  5. The last active timestamp (a floating point number)
  6. The last known path that this record was matched to (for old sincedb records converted to the new format, this is blank.

On non-Windows systems you can obtain the inode number of a file with e.g. ls -li.

Reading from remote network volumes

edit

The file input is not thoroughly tested on remote filesystems such as NFS, Samba, s3fs-fuse, etc, however NFS is occasionally tested. The file size as given by the remote FS client is used to govern how much data to read at any given time to prevent reading into allocated but yet unfilled memory. As we use the device major and minor in the identifier to track "last read" positions of files and on remount the device major and minor can change, the sincedb records may not match across remounts. Read mode might not be suitable for remote filesystems as the file size at discovery on the client side may not be the same as the file size on the remote side due to latency in the remote to client copy process.

File rotation in Tail mode

edit

File rotation is detected and handled by this input, regardless of whether the file is rotated via a rename or a copy operation. To support programs that write to the rotated file for some time after the rotation has taken place, include both the original filename and the rotated filename (e.g. /var/log/syslog and /var/log/syslog.1) in the filename patterns to watch (the path option). For a rename, the inode will be detected as having moved from /var/log/syslog to /var/log/syslog.1 and so the "state" is moved internally too, the old content will not be reread but any new content on the renamed file will be read. For copy/truncate the copied content into a new file path, if discovered, will be treated as a new discovery and be read from the beginning. The copied file paths should therefore not be in the filename patterns to watch (the path option). The truncation will be detected and the "last read" position updated to zero.

File Input Configuration Options

edit

This plugin supports the following configuration options plus the Common Options described later.

Duration settings can be specified in text form e.g. "250 ms", this string will be converted into decimal seconds. There are quite a few supported natural and abbreviated durations, see string_duration for the details.

Also see Common Options for a list of options supported by all input plugins.

 

check_archive_validity

edit
  • Value type is boolean
  • The default is false.

When set to true, this setting verifies that a compressed file is valid before processing it. There are two passes through the file—​one pass to verify that the file is valid, and another pass to process the file.

Validating a compressed file requires more processing time, but can prevent a corrupt archive from causing looping.

close_older

edit

The file input closes any files that were last read the specified duration (seconds if a number is specified) ago. This has different implications depending on if a file is being tailed or read. If tailing, and there is a large time gap in incoming data the file can be closed (allowing other files to be opened) but will be queued for reopening when new data is detected. If reading, the file will be closed after closed_older seconds from when the last bytes were read. This setting is retained for backward compatibility if you upgrade the plugin to 4.1.0+, are reading not tailing and do not switch to using Read mode.

delimiter

edit
  • Value type is string
  • Default value is "\n"

set the new line delimiter, defaults to "\n". Note that when reading compressed files this setting is not used, instead the standard Windows or Unix line endings are used.

discover_interval

edit
  • Value type is number
  • Default value is 15

How often we expand the filename patterns in the path option to discover new files to watch. This value is a multiple to stat_interval, e.g. if stat_interval is "500 ms" then new files files could be discovered every 15 X 500 milliseconds - 7.5 seconds. In practice, this will be the best case because the time taken to read new content needs to be factored in.

exclude

edit
  • Value type is array
  • There is no default value for this setting.

Exclusions (matched against the filename, not full path). Filename patterns are valid here, too. For example, if you have

    path => "/var/log/*"

In Tail mode, you might want to exclude gzipped files:

    exclude => "*.gz"

exit_after_read

edit
  • Value type is boolean
  • Default value is false

This option can be used in read mode to enforce closing all watchers when file gets read. Can be used in situation when content of the file is static and won’t change during execution. When set to true it also disables active discovery of the files - only files that were in the directories when process was started will be read. It supports sincedb entries. When file was processed once, then modified - next run will only read newly added entries.

file_chunk_count

edit
  • Value type is number
  • Default value is 4611686018427387903

When combined with the file_chunk_size, this option sets how many chunks (bands or stripes) are read from each file before moving to the next active file. For example, a file_chunk_count of 32 and a file_chunk_size 32KB will process the next 1MB from each active file. As the default is very large, the file is effectively read to EOF before moving to the next active file.

file_chunk_size

edit
  • Value type is number
  • Default value is 32768 (32KB)

File content is read off disk in blocks or chunks and lines are extracted from the chunk. See file_chunk_count to see why and when to change this setting from the default.

file_completed_action

edit
  • Value can be any of: delete, log, log_and_delete
  • The default is delete.

When in read mode, what action should be carried out when a file is done with. If delete is specified then the file will be deleted. If log is specified then the full path of the file is logged to the file specified in the file_completed_log_path setting. If log_and_delete is specified then both above actions take place.

file_completed_log_path

edit
  • Value type is string
  • There is no default value for this setting.

Which file should the completely read file paths be appended to. Only specify this path to a file when file_completed_action is log or log_and_delete. IMPORTANT: this file is appended to only - it could become very large. You are responsible for file rotation.

file_sort_by

edit
  • Value can be any of: last_modified, path
  • The default is last_modified.

Which attribute of a "watched" file should be used to sort them by. Files can be sorted by modified date or full path alphabetic. Previously the processing order of the discovered and therefore "watched" files was OS dependent.

file_sort_direction

edit
  • Value can be any of: asc, desc
  • The default is asc.

Select between ascending and descending order when sorting "watched" files. If oldest data first is important then the defaults of last_modified + asc are good. If newest data first is more important then opt for last_modified + desc. If you use special naming conventions for the file full paths then perhaps path + asc will help to control the order of file processing.

ignore_older

edit

When the file input discovers a file that was last modified before the specified duration (seconds if a number is specified), the file is ignored. After it’s discovery, if an ignored file is modified it is no longer ignored and any new data is read. By default, this option is disabled. Note this unit is in seconds.

max_open_files

edit
  • Value type is number
  • There is no default value for this setting.

What is the maximum number of file_handles that this input consumes at any one time. Use close_older to close some files if you need to process more files than this number. This should not be set to the maximum the OS can do because file handles are needed for other LS plugins and OS processes. A default of 4095 is set in internally.

mode

edit
  • Value can be either tail or read.
  • The default value is tail.

What mode do you want the file input to operate in. Tail a few files or read many content-complete files. Read mode now supports gzip file processing.

If read is specified, these settings can be used:

  • ignore_older (older files are not processed)
  • file_completed_action (what action should be taken when the file is processed)
  • file_completed_log_path (which file should the completed file path be logged to)

If read is specified, these settings are ignored:

  • start_position (files are always read from the beginning)
  • close_older (files are automatically closed when EOF is reached)

path

edit
  • This is a required setting.
  • Value type is array
  • There is no default value for this setting.

The path(s) to the file(s) to use as an input. You can use filename patterns here, such as /var/log/*.log. If you use a pattern like /var/log/**/*.log, a recursive search of /var/log will be done for all *.log files. Paths must be absolute and cannot be relative.

You may also configure multiple paths. See an example on the Logstash configuration page.

sincedb_clean_after

edit
  • Value type is number or string_duration
  • The default value for this setting is "2 weeks".
  • If a number is specified then it is interpreted as days and can be decimal e.g. 0.5 is 12 hours.

The sincedb record now has a last active timestamp associated with it. If no changes are detected in a tracked file in the last N days its sincedb tracking record expires and will not be persisted. This option helps protect against the inode recycling problem. Filebeat has a FAQ about inode recycling.

sincedb_path

edit
  • Value type is string
  • There is no default value for this setting.

Path of the sincedb database file (keeps track of the current position of monitored log files) that will be written to disk. The default will write sincedb files to <path.data>/plugins/inputs/file NOTE: it must be a file path and not a directory path

sincedb_write_interval

edit

How often (in seconds) to write a since database with the current position of monitored log files.

start_position

edit
  • Value can be any of: beginning, end
  • Default value is "end"

Choose where Logstash starts initially reading files: at the beginning or at the end. The default behavior treats files like live streams and thus starts at the end. If you have old data you want to import, set this to beginning.

This option only modifies "first contact" situations where a file is new and not seen before, i.e. files that don’t have a current position recorded in a sincedb file read by Logstash. If a file has already been seen before, this option has no effect and the position recorded in the sincedb file will be used.

stat_interval

edit

How often (in seconds) we stat files to see if they have been modified. Increasing this interval will decrease the number of system calls we make, but increase the time to detect new log lines.

Discovering new files and checking whether they have grown/or shrunk occurs in a loop. This loop will sleep for stat_interval seconds before looping again. However, if files have grown, the new content is read and lines are enqueued. Reading and enqueuing across all grown files can take time, especially if the pipeline is congested. So the overall loop time is a combination of the stat_interval and the file read time.

Common Options

edit

The following configuration options are supported by all input plugins:

Setting Input type Required

add_field

hash

No

codec

codec

No

enable_metric

boolean

No

id

string

No

tags

array

No

type

string

No

Details

edit

 

add_field

edit
  • Value type is hash
  • Default value is {}

Add a field to an event

codec

edit
  • Value type is codec
  • Default value is "plain"

The codec used for input data. Input codecs are a convenient method for decoding your data before it enters the input, without needing a separate filter in your Logstash pipeline.

enable_metric

edit
  • Value type is boolean
  • Default value is true

Disable or enable metric logging for this specific plugin instance by default we record all the metrics we can, but you can disable metrics collection for a specific plugin.

  • Value type is string
  • There is no default value for this setting.

Add a unique ID to the plugin configuration. If no ID is specified, Logstash will generate one. It is strongly recommended to set this ID in your configuration. This is particularly useful when you have two or more plugins of the same type, for example, if you have 2 file inputs. Adding a named ID in this case will help in monitoring Logstash when using the monitoring APIs.

input {
  file {
    id => "my_plugin_id"
  }
}

Variable substitution in the id field only supports environment variables and does not support the use of values from the secret store.

tags

edit
  • Value type is array
  • There is no default value for this setting.

Add any number of arbitrary tags to your event.

This can help with processing later.

type

edit
  • Value type is string
  • There is no default value for this setting.

Add a type field to all events handled by this input.

Types are used mainly for filter activation.

The type is stored as part of the event itself, so you can also use the type to search for it in Kibana.

If you try to set a type on an event that already has one (for example when you send an event from a shipper to an indexer) then a new input will not override the existing type. A type set at the shipper stays with that event for its life even when sent to another Logstash server.

String Durations

edit

Format is number string and the space between these is optional. So "45s" and "45 s" are both valid.

Use the most suitable duration, for example, "3 days" rather than "72 hours".

Weeks

edit

Supported values: w week weeks, e.g. "2 w", "1 week", "4 weeks".

Days

edit

Supported values: d day days, e.g. "2 d", "1 day", "2.5 days".

Hours

edit

Supported values: h hour hours, e.g. "4 h", "1 hour", "0.5 hours".

Minutes

edit

Supported values: m min minute minutes, e.g. "45 m", "35 min", "1 minute", "6 minutes".

Seconds

edit

Supported values: s sec second seconds, e.g. "45 s", "15 sec", "1 second", "2.5 seconds".

Milliseconds

edit

Supported values: ms msec msecs, e.g. "500 ms", "750 msec", "50 msecs

milli millis and milliseconds are not supported

Microseconds

edit

Supported values: us usec usecs, e.g. "600 us", "800 usec", "900 usecs"

micro micros and microseconds are not supported