EFK Stack on Kubernetes (Part 2)

This is the final part of our Kubernetes logging series. In case you missed part 1 you can find it here. In this tutorial, we will learn about configuring Filebeat to run as a DaemonSet in our Kubernetes cluster in order to ship logs to the Elasticsearch backend. We are using Filebeat instead of FluentD or FluentBit because it is an extremely lightweight utility and has a first-class support for Kubernetes. It is best for production-level setups.

1. Deployment Architecture

Filebeat will run as a DaemonSet in our Kubernetes cluster. It will be:

• Deployed in a separate namespace called Logging.
• Pods will be scheduled on both Master nodes and Worker Nodes.
• Master Node pods will forward api-server logs for audit and cluster administration purposes.
• Client Node pods will forward workload related logs for application observability.

2.1 Creating Filebeat ServiceAccount and ClusterRole

Deploy the following manifest to create the required permissions for Filebeat pods.

apiVersion: v1
kind: Namespace
metadata:
  name: logging
---
apiVersion: v1
kind: ServiceAccount
metadata:
  name: filebeat
  namespace: logging
  labels:
    k8s-app: filebeat
---
apiVersion: rbac.authorization.k8s.io/v1beta1
kind: ClusterRole
metadata:
  name: filebeat
  namespace: logging
  labels:
    k8s-app: filebeat
rules:
- apiGroups: [""] # "" indicates the core API group
  resources:
  - namespaces
  - pods
  verbs:
  - get
  - watch
  - list
---
apiVersion: rbac.authorization.k8s.io/v1beta1
kind: ClusterRoleBinding
metadata:
  name: filebeat
  namespace: logging
subjects:
- kind: ServiceAccount
  name: filebeat
  namespace: kube-system
roleRef:
  kind: ClusterRole
  name: filebeat
  apiGroup: rbac.authorization.k8s.io

We should make sure that ClusterRole permissions are as limited as possible from a security point of view. If either of the pod associated with this service account gets compromised then the attacker would not be able to gain access entire cluster or applications running in it.

2.2 Creating Filebeat ConfigMap

Use the following manifest to create a ConfigMap which will be used by Filebeat pods.

apiVersion: v1
kind: Namespace
metadata:
  name: logging
---
apiVersion: v1
kind: ConfigMap
metadata:
  name: filebeat-config
  namespace: logging
  labels:
    k8s-app: filebeat
    kubernetes.io/cluster-service: "true"
data:
  filebeat.yml: |-
    filebeat.config:
    #  inputs:
    #    path: ${path.config}/inputs.d/*.yml
    #    reload.enabled: true
      modules:
        path: ${path.config}/modules.d/*.yml
        reload.enabled: true

    filebeat.autodiscover:
      providers:
        - type: kubernetes
          hints.enabled: true
          include_annotations: ["artifact.spinnaker.io/name","ad.datadoghq.com/tags"]
          include_labels: ["app.kubernetes.io/name"]
          labels.dedot: true
          annotations.dedot: true
          templates:
            - condition:
                equals:
                  kubernetes.namespace: myapp   #Set the namespace in which your app is running, can add multiple conditions in case of more than 1 namespace.
              config:
                - type: docker
                  containers.ids:
                    - "${data.kubernetes.container.id}"
                  multiline:
                    pattern: '^[A-Za-z ]+[0-9]{2} (?:[01]\d|2[0123]):(?:[012345]\d):(?:[012345]\d)'.   #Timestamp regex for the app logs. Change it as per format. 
                    negate: true
                    match: after
            - condition:
                equals:
                  kubernetes.namespace: elasticsearch
              config:
                - type: docker
                  containers.ids:
                    - "${data.kubernetes.container.id}"
                  multiline:
                    pattern: '^\[[0-9]{4}-[0-9]{2}-[0-9]{2}|^[0-9]{4}-[0-9]{2}-[0-9]{2}T'
                    negate: true
                    match: after

    processors:
      - add_cloud_metadata: ~
      - drop_fields:
          when:
            has_fields: ['kubernetes.labels.app']
          fields:
            - 'kubernetes.labels.app'

    output.elasticsearch:
      hosts: ['${ELASTICSEARCH_HOST:elasticsearch}:${ELASTICSEARCH_PORT:9200}']

Important concepts for the Filebeat ConfigMap:

• hints.enabled: This activates Filebeat’s hints module for Kubernetes. By using this we can use pod annotations to pass config directly to Filebeat pod. We can specify different multiline patterns and various other types of config. More about this can be read here.
• include_annotations: Setting this to true enables Filebeat to retain any pod annotation for a particular log entry. These annotations can be later used to filter logs in the Kibana console.
• include_labels: Setting this to true enables Filebeat to retain any pod labels for a particular log entry. These labels can be later used to filter logs in the Kibana console.
• We can also filter logs for a particular namespace and then can process the log entries accordingly. Here docker log processor is used. We can also use different multiline patterns for different namespaces.
• The output is set to Elasticsearch because we are using Elasticsearch as the storage backend. Alternatively, this can also point to Redis, Logstash, Kafka or even a File. More this can be read here.
• Cloud metadata processor includes some host-specific fields in the log entry. This is helpful when we try to filter logs specific to a particular worker node.

2.3 Deploying Filebeat DaemonSet

Use the manifest below to deploy the Filebeat DaemonSet.

apiVersion: v1
kind: Namespace
metadata:
  name: logging
---
apiVersion: extensions/v1beta1
kind: DaemonSet
metadata:
  name: filebeat
  namespace: logging
  labels:
    k8s-app: filebeat
spec:
  template:
    metadata:
      labels:
        k8s-app: filebeat
    spec:
      serviceAccountName: filebeat
      terminationGracePeriodSeconds: 30
      tolerations:
      - effect: NoSchedule
        key: node-role.kubernetes.io/master
      containers:
      - name: filebeat
        image: elastic/filebeat:6.5.4
        args: [
          "-c", "/usr/share/filebeat/filebeat.yml",
          "-e",
        ]
        env:
        - name: ELASTICSEARCH_HOST
          value: elasticsearch.elasticsearch
        - name: ELASTICSEARCH_PORT
          value: "9200"
        securityContext:
          runAsUser: 0
          # If using Red Hat OpenShift uncomment this:
          #privileged: true
        resources:
          limits:
            memory: 200Mi
          requests:
            cpu: 100m
            memory: 100Mi
        volumeMounts:
        - name: config
          mountPath: /usr/share/filebeat/filebeat.yml
          readOnly: true
          subPath: filebeat.yml
        - name: inputs
          mountPath: /usr/share/filebeat/inputs.d
          readOnly: true
        - name: data
          mountPath: /usr/share/filebeat/data
        - name: varlibdockercontainers
          mountPath: /var/lib/docker/containers
          readOnly: true
      volumes:
      - name: config
        configMap:
          defaultMode: 0600
          name: filebeat-config
      - name: varlibdockercontainers
        hostPath:
          path: /var/lib/docker/containers
      - name: inputs
        configMap:
          defaultMode: 0600
          name: filebeat-inputs
      # data folder stores a registry of read status for all files, so we don't send everything again on a Filebeat pod restart
      - name: data
        hostPath:
          path: /var/lib/filebeat-data
          type: DirectoryOrCreate
---

Let’s see what is going on here:

• Logs for each pod are written to /var/log/docker/containers. We are mounting this directory from the host to the Filebeat pod and then Filebeat processes the logs according to the provided configuration.
• We have set the env var ELASTICSEARCH_HOST to elasticsearch.elasticsearch to refer to the Elasticsearch client service which was created in part 1 of this article. In case you already have an Elasticsearch cluster running, the env var should be set to point to it.

Please note the following setting in the manifest:

...
      tolerations:
      - effect: NoSchedule
        key: node-role.kubernetes.io/master
...

This makes sure that our Filebeat DaemonSet schedules a pod on the master node as well. Once the Filebeat DaemonSet is deployed we can check if our pods get scheduled properly

root$ kubectl -n logging get pods  -o wide
NAME             READY   STATUS    RESTARTS   AGE   IP            NODE                                         NOMINATED NODE   READINESS GATES
filebeat-4kchs   1/1     Running   0          6d    100.96.8.2    ip-10-10-30-206.us-east-2.compute.internal   <none>           <none>
filebeat-6nrpc   1/1     Running   0          6d    100.96.7.6    ip-10-10-29-252.us-east-2.compute.internal   <none>           <none>
filebeat-7qs2s   1/1     Running   0          6d    100.96.1.6    ip-10-10-30-161.us-east-2.compute.internal   <none>           <none>
filebeat-j5xz6   1/1     Running   0          6d    100.96.5.3    ip-10-10-28-186.us-east-2.compute.internal   <none>           <none>
filebeat-pskg5   1/1     Running   0          6d    100.96.64.4   ip-10-10-29-142.us-east-2.compute.internal   <none>           <none>
filebeat-vjdtg   1/1     Running   0          6d    100.96.65.3   ip-10-10-30-118.us-east-2.compute.internal   <none>           <none>
filebeat-wm24j   1/1     Running   0          6d    100.96.0.4    ip-10-10-28-162.us-east-2.compute.internal   <none>           <none>

root$ kubectl -get nodes -o wide
NAME                                         STATUS   ROLES    AGE   VERSION   INTERNAL-IP    EXTERNAL-IP   OS-IMAGE                       KERNEL-VERSION   CONTAINER-RUNTIME
ip-10-10-28-162.us-east-2.compute.internal   Ready    master   6d    v1.14.8   10.10.28.162   <none>        Debian GNU/Linux 9 (stretch)   4.9.0-9-amd64    docker://18.6.3
ip-10-10-28-186.us-east-2.compute.internal   Ready    node     6d    v1.14.8   10.10.28.186   <none>        Debian GNU/Linux 9 (stretch)   4.9.0-9-amd64    docker://18.6.3
ip-10-10-29-142.us-east-2.compute.internal   Ready    master   6d    v1.14.8   10.10.29.142   <none>        Debian GNU/Linux 9 (stretch)   4.9.0-9-amd64    docker://18.6.3
ip-10-10-29-252.us-east-2.compute.internal   Ready    node     6d    v1.14.8   10.10.29.252   <none>        Debian GNU/Linux 9 (stretch)   4.9.0-9-amd64    docker://18.6.3
ip-10-10-30-118.us-east-2.compute.internal   Ready    master   6d    v1.14.8   10.10.30.118   <none>        Debian GNU/Linux 9 (stretch)   4.9.0-9-amd64    docker://18.6.3
ip-10-10-30-161.us-east-2.compute.internal   Ready    node     6d    v1.14.8   10.10.30.161   <none>        Debian GNU/Linux 9 (stretch)   4.9.0-9-amd64    docker://18.6.3
ip-10-10-30-206.us-east-2.compute.internal   Ready    node     6d    v1.14.8   10.10.30.206   <none>        Debian GNU/Linux 9 (stretch)   4.9.0-9-amd64    docker://18.6.3

If we tail the logs for one of the pods we can clearly see that it connected to Elasticsearch and has started harvester for the files. The snippet below shows this:

2019-11-19T06:22:03.435Z    INFO    log/input.go:138    Configured paths: [/var/lib/docker/containers/c2b29f5e06eb8affb2cce7cf2501f6f824a2fd83418d09823faf4e74a5a51eb7/*.log]
2019-11-19T06:22:03.435Z    INFO    input/input.go:114  Starting input of type: docker; ID: 4134444498769889169 
2019-11-19T06:22:04.786Z    INFO    input/input.go:149  input ticker stopped
2019-11-19T06:22:04.786Z    INFO    input/input.go:167  Stopping Input: 4134444498769889169
2019-11-19T06:22:19.295Z    INFO    [monitoring]    log/log.go:144  Non-zero metrics in the last 30s    {"monitoring": {"metrics": {"beat":{"cpu":{"system":{"ticks":641680,"time":{"ms":16}},"total":{"ticks":2471920,"time":{"ms":180},"value":2471920},"user":{"ticks":1830240,"time":{"ms":164}}},"handles":{"limit":{"hard":1048576,"soft":1048576},"open":20},"info":{"ephemeral_id":"007e8090-7c62-4b44-97fb-e74e8177dc54","uptime":{"ms":549390018}},"memstats":{"gc_next":47281968,"memory_alloc":29021760,"memory_total":156062982472}},"filebeat":{"events":{"added":111,"done":111},"harvester":{"closed":2,"open_files":15,"running":13}},"libbeat":{"config":{"module":{"running":0}},"output":{"events":{"acked":108,"batches":15,"total":108},"read":{"bytes":69},"write":{"bytes":123536}},"pipeline":{"clients":1847,"events":{"active":0,"filtered":3,"published":108,"total":111},"queue":{"acked":108}}},"registrar":{"states":{"current":87,"update":111},"writes":{"success":18,"total":18}},"system":{"load":{"1":0.98,"15":1.71,"5":1.59,"norm":{"1":0.0613,"15":0.1069,"5":0.0994}}}}}}

2019-11-19T06:22:49.295Z    INFO    [monitoring]    log/log.go:144  Non-zero metrics in the last 30s    {"monitoring": {"metrics": {"beat":{"cpu":{"system":{"ticks":641720,"time":{"ms":44}},"total":{"ticks":2472030,"time":{"ms":116},"value":2472030},"user":{"ticks":1830310,"time":{"ms":72}}},"handles":{"limit":{"hard":1048576,"soft":1048576},"open":20},"info":{"ephemeral_id":"007e8090-7c62-4b44-97fb-e74e8177dc54","uptime":{"ms":549420018}},"memstats":{"gc_next":47281968,"memory_alloc":38715472,"memory_total":156072676184}},"filebeat":{"events":{"active":12,"added":218,"done":206},"harvester":{"open_files":15,"running":13}},"libbeat":{"config":{"module":{"running":0}},"output":{"events":{"acked":206,"batches":24,"total":206},"read":{"bytes":102},"write":{"bytes":269666}},"pipeline":{"clients":1847,"events":{"active":12,"published":218,"total":218},"queue":{"acked":206}}},"registrar":{"states":{"current":87,"update":206},"writes":{"success":24,"total":24}},"system":{"load":{"1":1.22,"15":1.7,"5":1.58,"norm":{"1":0.0763,"15":0.1063,"5":0.0988}}}}}}

2019-11-19T06:23:19.295Z    INFO    [monitoring]    log/log.go:144  Non-zero metrics in the last 30s    {"monitoring": {"metrics": {"beat":{"cpu":{"system":{"ticks":641750,"time":{"ms":28}},"total":{"ticks":2472110,"time":{"ms":72},"value":2472110},"user":{"ticks":1830360,"time":{"ms":44}}},"handles":{"limit":{"hard":1048576,"soft":1048576},"open":20},"info":{"ephemeral_id":"007e8090-7c62-4b44-97fb-e74e8177dc54","uptime":{"ms":549450017}},"memstats":{"gc_next":47281968,"memory_alloc":43140256,"memory_total":156077100968}},"filebeat":{"events":{"active":-12,"added":43,"done":55},"harvester":{"open_files":15,"running":13}},"libbeat":{"config":{"module":{"running":0}},"output":{"events":{"acked":55,"batches":12,"total":55},"read":{"bytes":51},"write":{"bytes":70798}},"pipeline":{"clients":1847,"events":{"active":0,"published":43,"total":43},"queue":{"acked":55}}},"registrar":{"states":{"current":87,"update":55},"writes":{"success":12,"total":12}},"system":{"load":{"1":0.99,"15":1.67,"5":1.49,"norm":{"1":0.0619,"15":0.1044,"5":0.0931}}}}}}

2019-11-19T06:23:25.261Z    INFO    log/harvester.go:255    Harvester started for file: /var/lib/docker/containers/ccb7dc75ecc755734f6befc4965b9fdae74d59810914101eadf63daa69eb62e2/ccb7dc75ecc755734f6befc4965b9fdae74d59810914101eadf63daa69eb62e2-json.log

2019-11-19T06:23:49.295Z    INFO    [monitoring]    log/log.go:144  Non-zero metrics in the last 30s    {"monitoring": {"metrics": {"beat":{"cpu":{"system":{"ticks":641780,"time":{"ms":28}},"total":{"ticks":2472310,"time":{"ms":196},"value":2472310},"user":{"ticks":1830530,"time":{"ms":168}}},"handles":{"limit":{"hard":1048576,"soft":1048576},"open":21},"info":{"ephemeral_id":"007e8090-7c62-4b44-97fb-e74e8177dc54","uptime":{"ms":549480018}},"memstats":{"gc_next":47789200,"memory_alloc":31372376,"memory_total":156086697176,"rss":-1064960}},"filebeat":{"events":{"active":16,"added":170,"done":154},"harvester":{"open_files":16,"running":14,"started":1}},"libbeat":{"config":{"module":{"running":0}},"output":{"events":{"acked":153,"batches":24,"total":153},"read":{"bytes":115},"write":{"bytes":207569}},"pipeline":{"clients":1847,"events":{"active":16,"filtered":1,"published":169,"total":170},"queue":{"acked":153}}},"registrar":{"states":{"current":87,"update":154},"writes":{"success":25,"total":25}},"system":{"load":{"1":0.87,"15":1.63,"5":1.41,"norm":{"1":0.0544,"15":0.1019,"5":0.0881}}}}}}

Once we have all our pods running, then we can create an index pattern of the type filebeat-* in Kibana. Filebeat indexes are generally timestamped. As soon as we create the index pattern, all the searchable available fields can be seen and should be imported.Lastly, we can search through our application logs and create dashboards if needed. It is highly recommended to have JSON logger in our applications because it makes log processing extremely easy and messages can be parsed easily.

3. Conclusion

This concludes our logging set-up. All of the provided configuration files have been tested in production environments and are readily deployable. Feel free to reach out should you have any questions around it.