How to use custom ingest pipelines with docker autodiscover - docker

Question

I'm having a hard time using custom Elasticsearch ingest pipelines with Filebeat's Docker autodiscovery. I've started out with custom processors in my filebeat.yml file, however I would prefer to shift this to custom ingest pipelines I've created.
Firstly, here is my configuration using custom processors that works to provide custom grok-like processing for my Servarr app Docker containers (identified by applying a label to them in my docker-compose.yml file). The processor copies the 'message' field to 'log.original', uses dissect to extract 'log.level', 'log.logger' and overwrite 'message'. The final processor is a JavaScript function used to convert the log.level to lowercase (overkill perhaps, but humour me).
Filebeat configuration:
filebeat.config:
modules:
path: ${path.config}/modules.d/*.yml
reload.enabled: true
reload.period: 60s
filebeat.autodiscover:
providers:
- type: docker
hints.enabled: true
json.keys_under_root: true
processors:
- if:
equals:
docker.container.labels.co_elastic_logs/custom_processor: servarr
then:
- copy_fields:
fields:
- from: message
to: log.original
fail_on_error: false
ignore_missing: true
- dissect:
tokenizer: "[%{log.level}] %{log.logger}: %{message}"
field: message
target_prefix: ""
overwrite_keys: true
ignore_failure: true
- script:
lang: javascript
id: lowercase
source: >
function process(event) {
var level = event.Get("log.level");
if(level != null) {
event.Put("log.level", level.toString().toLowerCase());
}
}
output.elasticsearch:
hosts: 'elasticsearch:9200'
username: 'elastic'
password: '*************'
setup.kibana.host: 'kibana:5601'
logging.json: true
logging.metrics.enabled: false
Excerpt from docker-compose.yml file...
lidarr:
image: ghcr.io/linuxserver/lidarr:latest
container_name: lidarr
labels:
co.elastic.logs/custom_processor: "servarr"
And an example log line (in json):
{"log":"[Info] DownloadDecisionMaker: Processing 100 releases \n","stream":"stdout","time":"2021-08-07T10:10:49.125702754Z"}
This works well, and achieves my aims of extracting fields, but ideally I'd like to use Elasticsearch's (more powerful) ingest pipelines instead, and live with a cleaner filebeat.yml, so I created a working ingest pipeline "filebeat-7.13.4-servarr-stdout-pipeline" like so (ignore the fact that for now, this only does the grokking):
[
{
"grok": {
"field": "message",
"patterns": [
"\\[%{LOGLEVEL:log.level}\\] %{WORD:log.logger}: %{GREEDYDATA:message}"
],
"trace_match": true,
"ignore_missing": true
}
}
]
I tested the pipeline against existing documents (not ones that have had my custom processing applied, I should note). The pipeline worked against all the documents I tested it against in the Kibana interface.
So now I come to shift my Filebeat config to use this pipeline for containers with my custom_processor label. This is the filebeat.yml I came up with, which is apparently valid and works for the most part, but doesn't apply the grokking:
filebeat.config:
modules:
path: ${path.config}/modules.d/*.yml
reload.enabled: true
reload.period: 60s
filebeat.autodiscover:
providers:
- type: docker
hints.enabled: true
json.keys_under_root: true
appenders:
- type: config
condition.equals:
docker.container.labels.co_elastic_logs/custom_processor: servarr
config:
pipeline: filebeat-7.13.4-servarr-stdout-pipeline
output.elasticsearch:
hosts: 'elasticsearch:9200'
username: 'elastic'
password: '*************'
setup.kibana.host: 'kibana:5601'
logging.json: true
logging.metrics.enabled: false
If I use Filebeat's inbuilt modules for my other containers such as nginx, by using a label such as in this example below, the inbuild module pipelines are used:
nginx-repo:
image: nginx:latest
container_name: nginx-repo
mem_limit: 2048m
environment:
- VIRTUAL_HOST=repo.***.***.***,repo
- VIRTUAL_PORT=80
- HTTPS_METHOD=noredirect
networks:
- default
- proxy
labels:
co.elastic.logs/module: "nginx"
co.elastic.logs/fileset.stdout: "access"
co.elastic.logs/fileset.stderr: "error"
What am I doing wrong here? The logs still end up in Elasticsearch and Kibana, and are processed, but my grok isn't applied, new fields aren't created, and the 'message' field is unchanged.
EDIT: In response to one of the comments linking to a post on the elastic forums, which suggested both the path(s) and the pipeline need to be made explicit, I tried the following filebeat.yml autodiscovery excerpt, which also fails to work (but is apparently valid config):
filebeat.autodiscover:
providers:
- type: docker
hints.enabled: true
json.keys_under_root: true
appenders:
- type: config
condition:
equals:
docker.container.labels.co_elastic_logs/custom_processor: "servarr"
config:
- type: docker
containers:
ids:
- "${data.docker.container.id}"
stream: all
paths:
- /var/lib/docker/containers/${data.docker.container.id}/${data.docker.container.id}-json.log
pipeline: filebeat-7.13.4-servarr-stdout-pipeline
I tried with the docker.container.labels.co_elastic_logs/custom_processor value both quoted and unquoted. I have the same behaviour where the logs end up in Elasticsearch / Kibana, but they are processed as if they skipped my ingest pipeline.

Answer 1

We're using Kubernetes instead of Docker with Filebeat but maybe our config might still help you out.
We have autodiscover enabled and have all pod logs sent to a common ingest pipeline except for logs from any Redis pod which use the Redis module and send their logs to Elasticsearch via one of two custom ingest pipelines depending on whether they're normal Redis logs or slowlog Redis logs, this is configured in the following block:
All other detected pod logs get sent in to a common ingest pipeline using the following catch-all configuration in the "output" section:
Something else that we do is add the name of the ingest pipeline to ingested documents using the "set" processor:
This has proven to be really helpful when diagnosing whether or not a pipeline was actually executed when viewing an event document in Kibana.