K
Q

Dynamic tagging for Fluentd td-agent source plugin

2/27/2019

I'm trying to implement a Streaming Sidecar Container logging architecture in Kubernetes using Fluentd.

In a single pod I have:

  • emptyDir Volume (as log storage)
  • Application container
  • Fluent log-forwarder container

Basically, the Application container logs are stored in the shared emptyDir volume. Fluentd log-forwarder container tails this log file in the shared emptyDir volume and forwards it an external log-aggregator.

The Fluentd log-forwarder container uses the following config in td-agent.conf:

<source>
  @type tail
  tag "#{ENV['TAG_VALUE']}"
  path (path to log file in volume)
  pos_file /var/log/td-agent/tmp/access.log.pos
  format json
  time_key time
  time_format %iso8601
  keep_time_key true
</source>

<match *.*>
  @type forward
  @id forward_tail
  heartbeat_type tcp
  <server>
    host (server-host-address)
  </server>
</match>

I'm using an environment variable to set the tag value so I can change it dynamically e.g. when I have to use this container side-by-side with a different Application container, I don't have to modify this config and rebuild this image again.

Now, I set the environment variable value during pod creation in Kubernetes:

    .
    .
    spec:
      containers:
      - name: application-pod
        image: application-image:1.0
        ports:
        - containerPort: 1234
        volumeMounts:
        - name: logvolume
          mountPath: /var/log/app
      - name: log-forwarder
        image: log-forwarder-image:1.0
        env:
        - name: "TAG_VALUE"
          value: "app.service01"
        volumeMounts:
        - name: logvolume
          mountPath: /var/log/app
      volumes:
      - name: logvolume
        emptyDir: {}

After deploying the pod, I found that the tag value in the Fluentd log-forwarder container comes out empty (expected value: "app.service01"). I imagine it's because Fluentd's td-agent initializes first before the TAG_VALUE environment variable gets assigned.

So, the main question is...
How can I dynamically set the td-agent's tag value?

But really, what I'm wondering is:
Is it possible to assign an environment variable before a container's initialization in Kubernetes?

-- libzz
fluentd
kubernetes
td-agent

Similar Questions

Am I suppose to run Kubernetes master (and Dashbaord) locally?
Cannot access pod via pod IP from within the cluster, causing liveness / readiness probe failure
spinnaker mount volume from configmap in read/write mode (0666)
K8s doesn't mount files on Persistent Volume

2 Answers

3/20/2019

As an answer to your first question (How can I dynamically set the td-agent's tag value?), this seems the best way that you are doing which is defining tag "#{ENV['TAG_VALUE']}" inside fluentd config file.

For your second question, environment variable is assigned before a container's initialization.

So it means it should work and I tested with below sample yaml, and it just worked fine.

apiVersion: v1
kind: ConfigMap
metadata:
  name: fluentd-conf
data:
  fluentd.conf.template: |
    <source>
      @type tail
      tag "#{ENV['TAG_VALUE']}"
      path /var/log/nginx/access.log
      format nginx
    </source>
    <match *.*>
      @type stdout
    </match>
---
apiVersion: v1
kind: Pod
metadata:
  name: log-forwarder
  labels:
    purpose: test-fluentd
spec:
  containers:
    - name: nginx
      image: nginx:latest
      volumeMounts:
        - name: logvolume
          mountPath: /var/log/nginx
    - name: fluentd
      image: fluent/fluentd
      env:
        - name: "TAG_VALUE"
          value: "test.nginx"
        - name: "FLUENTD_CONF"
          value: "fluentd.conf"
      volumeMounts:
        - name: fluentd-conf
          mountPath: /fluentd/etc
        - name: logvolume
          mountPath: /var/log/nginx
  volumes:
    - name: fluentd-conf
      configMap:
        name: fluentd-conf
        items:
          - key: fluentd.conf.template
            path: fluentd.conf
    - name: logvolume
      emptyDir: {}
  restartPolicy: Never

And when I curl nginx pod, I see this output on fluentd containers stdout.

kubectl logs -f log-forwarder fluentd

2019-03-20 09:50:54.000000000 +0000 test.nginx: {"remote":"10.20.14.1","host":"-","user":"-","method":"GET","path":"/","code":"200","size":"612","referer":"-","agent":"curl/7.60.0","http_x_forwarded_for":"-"}
2019-03-20 09:50:55.000000000 +0000 test.nginx: {"remote":"10.20.14.1","host":"-","user":"-","method":"GET","path":"/","code":"200","size":"612","referer":"-","agent":"curl/7.60.0","http_x_forwarded_for":"-"}
2019-03-20 09:50:56.000000000 +0000 test.nginx: {"remote":"10.128.0.26","host":"-","user":"-","method":"GET","path":"/","code":"200","size":"612","referer":"-","agent":"curl/7.60.0","http_x_forwarded_for":"-"}

As you can see, my environment variable TAG_VALUE=test.nginx has applied to log entries.

I hope it will be useful.

-- coolinuxoid
Source: StackOverflow
2/27/2019

You can use the combo fluent-plugin-kubernetes_metadata_filter and fluent-plugin-rewrite-tag-filter to set container name or something to the tag.

-- okkez
Source: StackOverflow