imfile: Text File Input Module¶
|Author:||Rainer Gerhards <firstname.lastname@example.org>|
This module provides the ability to convert any standard text file into a syslog message. A standard text file is a file consisting of printable characters with lines being delimited by LF.
The file is read line-by-line and any line read is passed to rsyslog’s rule engine. The rule engine applies filter conditions and selects which actions needs to be carried out. Empty lines are not processed, as they would result in empty syslog records. They are simply ignored.
As new lines are written they are taken from the file and processed. Depending on the selected mode, this happens via inotify or based on a polling interval. Especially in polling mode, file reading doesn’t happen immediately. But there are also slight delays (due to process scheduling and internal processing) in inotify mode.
The file monitor supports file rotation. To fully work, rsyslogd must run while the file is rotated. Then, any remaining lines from the old file are read and processed and when done with that, the new file is being processed from the beginning. If rsyslogd is stopped during rotation, the new file is read, but any not-yet-reported lines from the previous file can no longer be obtained.
When rsyslogd is stopped while monitoring a text file, it records the last processed location and continues to work from there upon restart. So no data is lost during a restart (except, as noted above, if the file is rotated just in this very moment).
Parameter names are case-insensitive.
New in version 8.1.5.
This specifies if imfile is shall run in inotify (“inotify”) or polling (“polling”) mode. Traditionally, imfile used polling mode, which is much more resource-intense (and slower) than inotify mode. It is suggested that users turn on “polling” mode only if they experience strange problems in inotify mode. In theory, there should never be a reason to enable “polling” mode and later versions will most probably remove it.
Note: if a legacy “$ModLoad” statement is used, the default is polling. This default was kept to prevent problems with old configurations. It might change in the future.
New in version 8.32.0.
On Solaris, the FEN API is used instead of INOTIFY. You can set the mode to fen or inotify (which is automatically mapped to fen on Solaris OS). Please note that the FEN is limited compared to INOTIFY. Deep wildcard matches may not work because of the API limits for now.
Default: 0 (no timeout)
New in version 8.23.0.
This sets the default value for input timeout parameters. See there for exact meaning. Parameter value is the number of seconds.
New in version 8.23.0.
This sets the interval in which multi-line-read timeouts are checked. The interval is specified in seconds. Note that this establishes a lower limit on the length of the timeout. For example, if a timeoutGranularity of 60 seconds is selected and a readTimeout value of 10 seconds is used, the timeout is nevertheless only checked every 60 seconds (if there is no other activity in imfile). This means that the readTimeout is also only checked every 60 seconds, which in turn means a timeout can occur only after 60 seconds.
Note that timeGranularity has some performance implication. The more frequently timeout processing is triggered, the more processing time is needed. This effect should be negligible, except if a very large number of files is being monitored.
New in version 8.32.0.
If this parameter is set to on, the files will be processed in sorted order, else not. However, due to the inherent asynchronicity of the whole operations involved in tracking files, it is not possible to guarantee this sorted order, as it also depends on operation mode and OS timing.
This setting specifies how often files are to be polled for new data. For obvious reasons, it has effect only if imfile is running in polling mode. The time specified is in seconds. During each polling interval, all files are processed in a round-robin fashion.
A short poll interval provides more rapid message forwarding, but requires more system resources. While it is possible, we strongly recommend not to set the polling interval to 0 seconds. That will make rsyslogd become a CPU hog, taking up considerable resources. It is supported, however, for the few very unusual situations where this level may be needed. Even if you need quick response, 1 seconds should be well enough. Please note that imfile keeps reading files as long as there is any data in them. So a “polling sleep” will only happen when nothing is left to be processed.
We recommend to use inotify mode.
The file being monitored. So far, this must be an absolute name (no macros or templates). Note that wildcards are supported at the file name level (see WildCards below for more details).
The tag to be assigned to messages read from this file. If you would like to
see the colon after the tag, you need to include it when you assign a tag
value, like so:
|integer or string (preferred)||local0||no||
The syslog facility to be assigned to messages read from this file. Can be
specified in textual form (e.g.
local1, …) or as numbers (e.g.
local0). Textual form is suggested. Default is
|integer or string (preferred)||notice||no||
The syslog severity to be assigned to lines read. Can be specified
in textual form (e.g.
warning, …) or as numbers (e.g. 6
info). Textual form is suggested. Default is
Specifies how often the state file shall be written when processing the input file. The default value is 0, which means a new state file is only written when the monitored files is being closed (end of rsyslogd execution). Any other value n means that the state file is written every time n file lines have been processed. This setting can be used to guard against message duplication due to fatal errors (like power fail). Note that this setting affects imfile performance, especially when set to a low value. Frequently writing the state file is very time consuming.
Note: If this parameter is not set, state files are not created.
New in version 8.10.0.
This permits the processing of multi-line messages. When set, a
messages is terminated when the next one begins, and
startmsg.regex contains the regex that identifies the start
of a message. As this parameter is using regular expressions, it
is more flexible than
readMode but at the cost of lower
endmsg.regex cannot all be
defined for the same input.
New in version 8.38.0.
This permits the processing of multi-line messages. When set, a message is
endmsg.regex matches the line that
identifies the end of a message. As this parameter is using regular
expressions, it is more flexible than
readMode but at the cost of lower
endmsg.regex cannot all be
defined for the same input.
The primary use case for this is multiline container log files which look like
date stdout P start of message date stdout P middle of message date stdout F end of message
The F means this is the line which contains the final part of the message. The fully assembled message should be start of message middle of message end of message. endmsg.regex=”^[^ ]+ stdout F “ will match.
New in version 8.23.0.
This can be used with startmsg.regex (but not readMode). If specified, partial multi-line reads are timed out after the specified timeout interval. That means the current message fragment is being processed and the next message fragment arriving is treated as a completely new message. The typical use case for this parameter is a file that is infrequently being written. In such cases, the next message arrives relatively late, maybe hours later. Specifying a readTimeout will ensure that those “last messages” are emitted in a timely manner. In this use case, the “partial” messages being processed are actually full messages, so everything is fully correct.
To guard against accidential too-early emission of a (partial) message, the timeout should be sufficiently large (5 to 10 seconds or more recommended). Specifying a value of zero turns off timeout processing. Also note the relationship to the timeoutGranularity global parameter, which sets the lower bound of readTimeout.
Setting timeout vaues slightly increases processing time requirements; the effect should only be visible of a very large number of files is being monitored.
This provides support for processing some standard types of multiline
messages. It is less flexible than
offers higher performance than regex processing. Note that
endmsg.regex cannot all be defined for the same
The value can range from 0-2 and determines the multiline detection method.
0 - (default) line based (each line is a new message)
1 - paragraph (There is a blank line between log messages)
2 - indented (new log messages start at the beginning of a line. If a line starts with a space or tab “t” it is part of the log message before it)
This is only meaningful if multi-line messages are to be processed. LF characters embedded into syslog messages cause a lot of trouble, as most tools and even the legacy syslog TCP protocol do not expect these. If set to “on”, this option avoid this trouble by properly escaping LF characters to the 4-byte sequence “#012”. This is consistent with other rsyslog control character escaping. By default, escaping is turned on. If you turn it off, make sure you test very carefully with all associated tools. Please note that if you intend to use plain TCP syslog with embedded LF characters, you need to enable octet-counted framing. For more details, see Rainer’s blog posting on imfile LF escaping.
This is a legacy setting that only is supported in polling mode. In inotify mode, it is fixed at 0 and all attempts to configure a different value will be ignored, but will generate an error message.
Please note that future versions of imfile may not support this parameter at all. So it is suggested to not use it.
In polling mode, if set to 0, each file will be fully processed and then processing switches to the next file. If it is set to any other value, a maximum of [number] lines is processed in sequence for each file, and then the file is switched. This provides a kind of mutiplexing the load of multiple files and probably leads to a more natural distribution of events when multiple busy files are monitored. For polling mode, the default is 10240.
This is an expert option. It can be used to set the maximum input batch size that imfile can generate. The default is 1024, which is suitable for a wide range of applications. Be sure to understand rsyslog message batch processing before you modify this option. If you do not know what this doc here talks about, this is a good indication that you should NOT modify the default.
This parameter controls if state files are deleted if their associated main file is deleted. Usually, this is a good idea, because otherwise problems would occur if a new file with the same name is created. In that case, imfile would pick up reading from the last position in the deleted file, which usually is not what you want.
However, there is one situation where not deleting associated state file makes sense: this is the case if a monitored file is modified with an editor (like vi or gedit). Most editors write out modifications by deleting the old file and creating a new now. If the state file would be deleted in that case, all of the file would be reprocessed, something that’s probably not intended in most case. As a side-note, it is strongly suggested not to modify monitored files with editors. In any case, in such a situation, it makes sense to disable state file deletion. That also applies to similar use cases.
In general, this parameter should only by set if the users knows exactly why this is required.
Binds the listener to a specific ruleset.
Default: see intro section on Metadata
This is used to turn on or off the addition of metadata to the message object.
This is used to turn on or off the addition of the “@cee:” cookie to the message object.
This is an experimental feature that tells rsyslog to reopen input file when it was truncated (inode unchanged but file size on disk is less than current offset in memory).
This is used to tell rsyslog to truncate the line which length is greater than specified bytes. If it is positive number, rsyslog truncate the line at specified bytes. Default value of ‘trimLineOverBytes’ is 0, means never truncate line.
This option can be used when
readMode is 0 or 2.
This is used to tell rsyslog to seek to the end/tail of input files (discard old logs) at its first start(freshStart) and process only new log messages.
When deploy rsyslog to a large number of servers, we may only care about new log messages generated after the deployment. set freshstartTail to on will discard old logs. Otherwise, there may be vast useless message burst on the remote central log receiver
This parameter only applies to files that are already existing during rsyslog’s initial processing of the file monitors.
Depending on the number and location of existing files, this initial startup processing may take some time as well. If another process creates a new file at exactly the time of startup processing and writes data to it, rsyslog might detect this file and it’s data as prexisting and may skip it. This race is inevitable. So when freshStartTail is used, some risk of data loss exists. The same holds true if between the last shutdown of rsyslog and its restart log file content has been added. As such, the rsyslog team advises against activating the freshStartTail option.
When messages are too long they are truncated and the following part is processed as a new message. When this parameter is turned on the truncated part is not processed but discarded.
Upon truncation an error is given. When this parameter is turned off, no error will be shown upon truncation.
The imfile module supports message metadata. It supports the following data items:
Name of the file where the message originated from. This is most useful when using wildcards inside file monitors, because it then is the only way to know which file the message originated from. The value can be accessed using the %$!metadata!filename% property. Note: For symlink-ed files this does not contain name of the actual file (source of the data) but name of the symlink (file which matched configured input).
Offset of the file in bytes at the time the message was read. The offset reported is from the start of the line. This information can be useful when recreating multi-line files that may have been accessed or transmitted non-sequentially. The value can be accessed using the %$!metadata!fileoffset% property.
Metadata is only present if enabled. By default it is enabled for input() statements that contain wildcards. For all others, it is disabled by default. It can explicitly be turned on or off via the addMetadata input() parameter, which always overrides the default.
Rsyslog must keep track of which parts of the monitored file
are already processed. This is done in so-called “state files” that
are created in the rsyslog working directory and are read on startup to
resume monitoring after a shutdown. The location of the rsyslog
working directory is configurable via the
advanced format parameter.
PersistStateInterval parameter must be set, otherwise state
files will NOT be created.
To avoid problems with duplicate state files, rsyslog automatically generates state file names according to the following scheme:
- the string “imfile-state:” is added before the actual file name, which includes the full path
- the full name is prepended after that string, but all occurrences of “/” are replaced by “-” to facilitate handling of these files
As a concrete example, consider file
being monitored. The corresponding state file will be named
Note that it is possible to set a fixed state file name via the
stateFile parameter. It is suggested to avoid this, as
the user must take care of name clashes. Most importantly, if
“stateFile” is set for file monitors with wildcards, the same
state file is used for all occurrences of these files. In short,
this will usually not work and cause confusion. Upon startup,
rsyslog tries to detect these cases and emit warning messages.
However, the detection simply checks for the presence of “*”
and as such it will not cover more complex cases.
Note that when the
parameter is not set or set to a non-writable location, the state file
will not be generated. In those cases, the file content will always
be completely re-sent by imfile, because the module does not know that it
already processed parts of that file.
- Before Version: 8.25.0
- Wildcards are only supported in the filename part, not in directory names.
- /var/log/*.log works. *
- /var/log/*/syslog.log does not work. *
- Since Version: 8.25.0
- Wildcards are supported in filename and paths which means these samples will work:
/var/log/*.log works. *
/var/log/*/syslog.log works. *
/var/log/*/*.log works. *
All matching files in all matching subfolders will work. Note that this may decrease performance in imfile depending on how many directories and files are being watched dynamically.
- currently, wildcards are only supported in inotify mode
- read modes other than “0” currently seem to have issues in inotify mode
The following sample monitors two files. If you need just one, remove the second one. If you need more, add them according to the sample ;). This code must be placed in /etc/rsyslog.conf (or wherever your distro puts rsyslog’s config files). Note that only commands actually needed need to be specified. The second file uses less commands and uses defaults instead.
module(load="imfile" PollingInterval="10") #needs to be done just once # File 1 input(type="imfile" File="/path/to/file1" Tag="tag1" Severity="error" Facility="local7") # File 2 input(type="imfile" File="/path/to/file2" Tag="tag2") # ... and so on ... #
Note: While these parameters are still accepted, they should no longer be used for newly created configurations.
This is the name of this file’s state file. This parameter should usually not be used. Check the section on “State Files” above for more details.
Help with configuring/using