README
¶
Plugin list
Inputs
dmesg
It reads kernel events from /dev/kmsg
fake
It provides an API to test pipelines and other plugins.
file
It watches for files in the provided directory and reads them line by line.
Each line should contain only one event. It also correctly handles rotations (rename/truncate) and symlinks.
From time to time, it instantly releases and reopens descriptors of the completely processed files.
Such behavior allows files to be deleted by a third party software even though file.d is still working (in this case the reopening will fail).
A watcher is trying to use the file system events to detect file creation and updates.
But update events don't work with symlinks, so watcher also periodically manually fstat all tracking files to detect changes.
⚠ It supports the commitment mechanism. But "least once delivery" is guaranteed only if files aren't being truncated. However,
file.dcorrectly handles file truncation, there is a little chance of data loss. It isn't afile.dissue. The data may have been written just before the file truncation. In this case, you may miss to read some events. If you care about the delivery, you should also know that thelogrotatemanual clearly states that copy/truncate may cause data loss even on a rotating stage. So use copy/truncate or similar actions only if your data isn't critical. In order to reduce potential harm of truncation, you can turn on notifications of file changes. By default the plugin is notified only on file creations. Note that following for changes is more CPU intensive.
Reading docker container log files:
pipelines:
example_docker_pipeline:
input:
type: file
watching_dir: /var/lib/docker/containers
offsets_file: /data/offsets.yaml
filename_pattern: "*-json.log"
persistence_mode: async
http
Reads events from HTTP requests with the body delimited by a new line.
Also, it emulates some protocols to allow receiving events from a wide range of software that use HTTP to transmit data.
E.g. file.d may pretend to be Elasticsearch allows clients to send events using Elasticsearch protocol.
So you can use Elasticsearch filebeat output plugin to send data to file.d.
⚠ Currently event commitment mechanism isn't implemented for this plugin. Plugin answers with HTTP code
OK 200right after it has read all the request body. It doesn't wait until events are committed.
journalctl
Reads journalctl output.
k8s
It reads Kubernetes logs and also adds pod meta-information. Also, it joins split logs into a single event.
Source log file should be named in the following format:
[pod-name]_[namespace]_[container-name]-[container-id].log
E.g. my_pod-1566485760-trtrq_my-namespace_my-container-4e0301b633eaa2bfdcafdeba59ba0c72a3815911a6a820bf273534b0f32d98e0.log
An information which plugin adds:
k8s_node– node name where pod is running;k8s_node_label_*– node labels;k8s_pod– pod name;k8s_namespace– pod namespace name;k8s_container– pod container name;k8s_label_*– pod labels.
Example:
pipelines:
example_k8s_pipeline:
input:
type: k8s
offsets_file: /data/offsets.yaml
file_config: // customize file plugin
persistence_mode: sync
read_buffer_size: 2048
kafka
It reads events from multiple Kafka topics using sarama library.
It guarantees at "at-least-once delivery" due to the commitment mechanism.
Actions
add_host
It adds field containing hostname to an event.
convert_date
It converts field date/time data to different format.
debug
It logs event to stdout. Useful for debugging.
discard
It drops an event. It is used in a combination with match_fields/match_mode parameters to filter out the events.
An example for discarding informational and debug logs:
pipelines:
example_pipeline:
...
actions:
- type: discard
match_fields:
level: /info|debug/
...
flatten
It extracts the object keys and adds them into the root with some prefix. If the provided field isn't an object, an event will be skipped.
Example:
pipelines:
example_pipeline:
...
actions:
- type: flatten
field: animal
prefix: pet_
...
It transforms {"animal":{"type":"cat","paws":4}} into {"pet_type":"b","pet_paws":"4"}.
join
It makes one big event from the sequence of the events. It is useful for assembling back together "exceptions" or "panics" if they were written line by line. Also known as "multiline".
⚠ Parsing the whole event flow could be very CPU intensive because the plugin uses regular expressions. Consider
match_fieldsparameter to process only particular events. Check out an example for details.
Example of joining Go panics:
pipelines:
example_pipeline:
...
actions:
- type: join
field: log
start: '/^(panic:)|(http: panic serving)/'
continue: '/(^\s*$)|(goroutine [0-9]+ \[)|(\([0-9]+x[0-9,a-f]+)|(\.go:[0-9]+ \+[0-9]x)|(\/.*\.go:[0-9]+)|(\(...\))|(main\.main\(\))|(created by .*\/.*\.)|(^\[signal)|(panic.+[0-9]x[0-9,a-f]+)|(panic:)/'
match_fields:
stream: stderr // apply only for events which was written to stderr to save CPU time
...
json_decode
It decodes a JSON string from the event field and merges the result with the event root. If the decoded JSON isn't an object, the event will be skipped.
keep_fields
It keeps the list of the event fields and removes others.
mask
Mask plugin matches event with regular expression and substitutions successfully matched symbols via asterix symbol. You could set regular expressions and submatch groups.
Example:
pipelines:
example_pipeline:
...
actions:
- type: mask
masks:
- mask:
re: "\b(\d{1,4})\D?(\d{1,4})\D?(\d{1,4})\D?(\d{1,4})\b"
groups: [1,2,3]
...
modify
It modifies the content for a field. It works only with strings.
You can provide an unlimited number of config parameters. Each parameter handled as cfg.FieldSelector:cfg.Substitution.
Example:
pipelines:
example_pipeline:
...
actions:
- type: modify
my_object.field.subfield: value is ${another_object.value}.
...
The resulting event could look like:
{
"my_object": {
"field": {
"subfield":"value is 666."
}
},
"another_object": {
"value": 666
}
parse_es
It parses HTTP input using Elasticsearch /_bulk API format. It converts sources defining create/index actions to the events. Update/delete actions are ignored.
Check out the details in Elastic Bulk API.
parse_re2
It parses string from the event field using re2 expression with named subgroups and merges the result with the event root.
remove_fields
It removes the list of the event fields and keeps others.
rename
It renames the fields of the event. You can provide an unlimited number of config parameters. Each parameter handled as cfg.FieldSelector:string.
When override is set to false, the field won't be renamed in the case of field name collision.
Sequence of rename operations isn't guaranteed. Use different actions for prioritization.
Example:
pipelines:
example_pipeline:
...
actions:
- type: rename
override: false
my_object.field.subfield: new_sub_field
...
The resulting event could look like:
{
"my_object": {
"field": {
"new_sub_field":"value"
}
},
throttle
It discards the events if pipeline throughput gets higher than a configured threshold.
Outputs
devnull
It provides an API to test pipelines and other plugins.
elasticsearch
It sends events into Elasticsearch. It uses _bulk API to send events in batches.
If a network error occurs, the batch will infinitely try to be delivered to the random endpoint.
gelf
It sends event batches to the GELF endpoint. Transport level protocol TCP or UDP is configurable.
It doesn't support UDP chunking. So don't use UDP if event size may be greater than 8192.
GELF messages are separated by null byte. Each message is a JSON with the following fields:
versionstring=1.1hoststringshort_messagestringfull_messagestringtimestampnumberlevelnumber_extra_field_1string_extra_field_2string_extra_field_3string
Every field with an underscore prefix _ will be treated as an extra field.
Allowed characters in field names are letters, numbers, underscores, dashes, and dots.
kafka
It sends the event batches to kafka brokers using sarama lib.
splunk
It sends events to splunk.
stdout
It writes events to stdout(also known as console).
Generated using insane-doc
Directories