Sumo Logic has ended development of this project, which has been replaced by the officially supported Sumo Logic Kubernetes Collection project. If you are planning to set up collection from a new cluster to Sumo Logic, we highly recommend that you use the supported Sumo Logic Kubernetes Collection project rather than this one. We will continue to support bug fixes for this project, but we expect to end support in summer 2020. Please consult the Kubernetes Collection project for instructions on how to migrate existing instances of this project to our new, fully supported solution.

This page describes the Sumo Kubernetes Fluentd plugin.

The code in this repository has been developed in collaboration with the Sumo Logic community and is not supported via standard Sumo Logic Support channels. For any issues or questions please submit an issue within the GitHub repository. The maintainers of this project will work directly with the community to answer any questions, address bugs, or review any requests for new features.

The plugin runs as a Kubernetes DaemonSet; it runs an instance of the plugin on each host in a cluster. Each plugin instance pulls system, kubelet, docker daemon, and container logs from the host and sends them, in JSON or text format, to an HTTP endpoint on a hosted collector in the Sumo service. Note the plugin with default configuration requires Kubernetes >=1.8. See the section below on running this on Kubernetes <1.8

• Step 1 Create hosted collector and HTTP source in Sumo
• Step 2 Create a Kubernetes secret
• Step 3 Install the Sumo Kubernetes FluentD plugin
  • Option A Install plugin using kubectl
  • Option B Helm chart
• Environment variables
  • Override environment variables using annotations
  • Exclude data using annotations
  • Include excluded using annotations
• Step 4 Set up Heapster for metric collection
  • Kubernetes ConfigMap
  • Kubernetes Service
  • Kubernetes Deployment
• Log data
  • Docker
  • Kubelet
  • Containers
• Taints and Tolerations
• Running On OpenShift

In this step you create, on the Sumo service, an HTTP endpoint to receive your logs. This process involves creating an HTTP source on a hosted collector in Sumo. In Sumo, collectors use sources to receive data.

1. If you don’t already have a Sumo account, you can create one by clicking the Free Trial button on https://www.sumologic.com/.
2. Create a hosted collector, following the instructions on Configure a Hosted Collector in Sumo help. (If you already have a Sumo hosted collector that you want to use, skip this step.)
3. Create an HTTP source on the collector you created in the previous step. For instructions, see HTTP Logs and Metrics Source in Sumo help.
4. When you have configured the HTTP source, Sumo will display the URL of the HTTP endpoint. Make a note of the URL. You will use it when you configure the Kubernetes service to send data to Sumo.

Create a secret in Kubernetes with the HTTP source URL. If you want to change the secret name, you must modify the Kubernetes manifest accordingly.

kubectl create secret generic sumologic --from-literal=collector-url=INSERT_HTTP_URL

You should see the confirmation message

secret "sumologic" created.

Follow the instructions in Option A below to install the plugin using kubectl. If you prefer to use a Helm chart, see Option B.

Before you start, see Environment variables for information about settings you can customize, and how to use annotations to override selected environment variables and exclude data from being sent to Sumo.

See the sample Kubernetes DaemonSet and Role in fluentd.yaml.

1. Clone the GitHub repo.

2. In fluentd-kubernetes-sumologic, install the chart using kubectl.

Which .yaml file you should use depends on whether or not you are running RBAC for authorization. RBAC is enabled by default as of Kubernetes 1.6. Note the plugin with default configuration requires Kubernetes >=1.8. See the section below on running this on Kubernetes <1.8

Non-RBAC (Kubernetes 1.5 and below)

kubectl create -f /daemonset/nonrbac/fluentd.yaml

RBAC (Kubernetes 1.6 and above)

kubectl create -f /daemonset/rbac/fluentd.yaml

Note if you modified the command in Step 2 to use a different name, update the .yaml file to use the correct secret.

Logs should begin flowing into Sumo within a few minutes of plugin installation.

If you use Helm to manage your Kubernetes resources, there is a Helm chart for the plugin at https://github.com/kubernetes/charts/tree/master/stable/sumologic-fluentd.

The following table show which environment variables affect which Fluentd sources.

When dealing with large volumes of data (TB's from what we have seen), FluentD may stop processing logs, but continue to run. This issue seems to be caused by the scalability of the inotify process that is packaged with the FluentD in_tail plugin. If you encounter this situation, setting the ENABLE_STAT_WATCHER to false should resolve this issue.

You can use the KUBERNETES_META_REDUCE environment variable (global) or the sumologic.com/kubernetes_meta_reduce annotation (per pod) to reduce the amount of Kubernetes metadata included with each log line under the kubernetes field.

When set, FluentD will remove the following properties:

• pod_id
• container_id
• namespace_id
• master_url
• labels
• annotations

Logs will still include:

• pod_name
• container_name
• namespace_name
• host

These fields still allow you to uniquely identify a pod and look up additional details with the Kubernetes API.

apiVersion: v1
kind: ReplicationController
metadata:
  name: nginx
spec:
  replicas: 1
  selector:
    app: mywebsite
  template:
    metadata:
      name: nginx
      labels:
        app: mywebsite
      annotations:
        sumologic.com/kubernetes_meta_reduce: "true"
    spec:
      containers:
      - name: nginx
        image: nginx
        ports:
        - containerPort: 80

You can override the LOG_FORMAT, KUBERNETES_META_REDUCE, SOURCE_CATEGORY and SOURCE_NAME environment variables, per pod, using Kubernetes annotations. For example:

apiVersion: v1
kind: ReplicationController
metadata:
  name: nginx
spec:
  replicas: 1
  selector:
    app: mywebsite
  template:
    metadata:
      name: nginx
      labels:
        app: mywebsite
      annotations:
        sumologic.com/format: "text"
        sumologic.com/kubernetes_meta_reduce: "true"
        sumologic.com/sourceCategory: "mywebsite/nginx"
        sumologic.com/sourceName: "mywebsite_nginx"
    spec:
      containers:
      - name: nginx
        image: nginx
        ports:
        - containerPort: 80

You can also use the sumologic.com/exclude annotation to exclude data from Sumo. This data is sent to FluentD, but not to Sumo.

apiVersion: v1
kind: ReplicationController
metadata:
  name: nginx
spec:
  replicas: 1
  selector:
    app: mywebsite
  template:
    metadata:
      name: nginx
      labels:
        app: mywebsite
      annotations:
        sumologic.com/format: "text"
        sumologic.com/sourceCategory: "mywebsite/nginx"
        sumologic.com/sourceName: "mywebsite_nginx"
        sumologic.com/exclude: "true"
    spec:
      containers:
      - name: nginx
        image: nginx
        ports:
        - containerPort: 80

If you excluded a whole namespace, but still need one or few pods to be still included for shipping to Sumologic, you can use the sumologic.com/include annotation to include data to Sumo. It takes precedence over the exclusion described above.

apiVersion: v1
kind: ReplicationController
metadata:
  name: nginx
spec:
  replicas: 1
  selector:
    app: mywebsite
  template:
    metadata:
      name: nginx
      labels:
        app: mywebsite
      annotations:
        sumologic.com/format: "text"
        sumologic.com/sourceCategory: "mywebsite/nginx"
        sumologic.com/sourceName: "mywebsite_nginx"
        sumologic.com/include: "true"
    spec:
      containers:
      - name: nginx
        image: nginx
        ports:
        - containerPort: 80

The recommended way to collect metrics from Kubernetes clusters is to use Heapster and a Sumo collector with a Graphite source.

Heapster aggregates metrics across a Kubenetes cluster. Heapster runs as a pod in the cluster, and discovers all nodes in the cluster and queries usage information from each node's kubelet—the on-machine Kubernetes agent.

Heapster provides metrics at the cluster, node and pod level.

1. Install Heapster in your Kubernetes cluster and configure a Graphite Sink to send the data in Graphite format to Sumo. For instructions, see https://github.com/kubernetes/heapster/blob/master/docs/sink-configuration.md#graphitecarbon. Assuming you have used the below YAML files to configure your system, then the sink option in graphite would be --sink=graphite:tcp://sumo-graphite.kube-system.svc:2003. You may need to change this depending on the namespace you run the deployment in, the name of the service or the port number for your Graphite source.

2. Use the Sumo Docker container. For instructions, see https://hub.docker.com/r/sumologic/collector/.

3. The following sections contain an example configmap, which contains the sources.json configuration, an example service, and an example deployment. Create these manifests in Kubernetes using kubectl.

kind: ConfigMap
apiVersion: v1
metadata:
  name: "sumo-sources"
data:
  sources.json: |-
    {
      "api.version": "v1",
      "sources": [
        {
          "name": "SOURCE_NAME",
          "category": "SOURCE_CATEGORY",
          "automaticDateParsing": true,
          "contentType": "Graphite",
          "timeZone": "UTC",
          "encoding": "UTF-8",
          "protocol": "TCP",
          "port": 2003,
          "sourceType": "Graphite"
        }
      ]
    }

apiVersion: v1
kind: Service
metadata:
  name: sumo-graphite
spec:
  ports:
    - port: 2003
  selector:
    app: sumo-graphite

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  labels:
    app: sumo-graphite
  name: sumo-graphite
spec:
  replicas: 2
  template:
    metadata:
      labels:
        app: sumo-graphite
    spec:
      volumes:
      - name: sumo-sources
        configMap:
          name: sumo-sources
          items:
          - key: sources.json
            path: sources.json
      containers:
      - name: sumo-graphite
        image: sumologic/collector:latest
        ports:
        - containerPort: 2003
        volumeMounts:
        - mountPath: /sumo
          name: sumo-sources
        env:
        - name: SUMO_ACCESS_ID
          value: <SUMO_ACCESS_ID>
        - name: SUMO_ACCESS_KEY
          value: <SUMO_ACCESS_KEY>
        - name: SUMO_SOURCES_JSON
          value: /sumo/sources.json

The following Kubernetes metadata is available for string templating:

Unlike the other templates, labels are not guaranteed to exist, so missing labels interpolate as "undefined".

For example, if you have only the label app: travel but you define SOURCE_NAME="%{label:app}@%{label:version}", the source name will appear as travel@undefined.

After performing the configuration described above, your logs should start streaming to SumoLogic in json or text format with the appropriate metadata. If you are using json format you can auto extract fields, for example _sourceCategory=some/app | json auto.

Note that Kubelet logs are only collected if you are using systemd. Kubernetes no longer outputs the kubelet logs to a file.

By default, the fluentd pods will schedule on, and therefore collect logs from, any worker nodes that do not have a taint and any master node that does not have a taint beyond the default master taint. If you would like to schedule pods on all nodes, regardless of taints, uncomment the following line from fluentd.yaml before applying it.

tolerations:
           #- operator: "Exists"

This daemonset setting mounts /var/log as service account FluentD so you need to run containers as privileged container. Here is command example:

oc adm policy add-scc-to-user privileged system:serviceaccount:logging:fluentd
oc adm policy add-cluster-role-to-user cluster-reader system:serviceaccount:logging:fluentd
oc label node —all logging-sumologic-fluentd=true
oc patch ds fluentd-sumologic -p "spec:
  template:
    spec:
      containers:
      - image: sumologic/fluentd-kubernetes-sumologic:latest
        name: fluentd
        securityContext:
          privileged: true"
oc delete pod -l name = fluentd-sumologic

In order to run this plugin on Kubernetes <1.8 you will need to make some changes the yaml file prior to deploying it.

Replace:

      - name: pos-files
        hostPath:
          path: /var/run/fluentd-pos
          type: ""

With:

      - name: pos-files
        emptyDir: {}

If you need to also send data to S3 (i.e. as a secondary backup/audit trail) the image includes the fluent-plugin-s3 plugin. In order to send the logs from FluentD to multiple outputs, you must use the copy plugin. This image comes with an OOB configuration to output the logs to Sumo Logic. In order to output to multiple destinations, you need to modify that existing configuration.

Example: Send all logs to S3 and Sumo:

<match **>
  @type copy
  <store>
    @type sumologic
    log_key log
    endpoint "#{ENV['COLLECTOR_URL']}"
    verify_ssl "#{ENV['VERIFY_SSL']}"
    log_format "#{ENV['LOG_FORMAT']}"
    flush_interval "#{ENV['FLUSH_INTERVAL']}"
    num_threads "#{ENV['NUM_THREADS']}"
    open_timeout 60
    add_timestamp "#{ENV['ADD_TIMESTAMP']}"
    proxy_uri "#{ENV['PROXY_URI']}"
  </store>
  <store>
    @type s3
  
    aws_key_id YOUR_AWS_KEY_ID
    aws_sec_key YOUR_AWS_SECRET_KEY
    s3_bucket YOUR_S3_BUCKET_NAME
    s3_region us-west-1
    path logs/
    buffer_path /var/log/fluent/s3
  
    time_slice_format %Y%m%d%H
    time_slice_wait 10m
    utc
  
    buffer_chunk_limit 256m
  </store>
</match>

You can replace the OOB configuration by creating a new Docker image from our image or by using a configmap to inject the new configuration to the pod.

More details about the S3 plugin can be found in the docs.

In version 2.0.0, some legacy FluentD configuration has been removed that could lead to duplicate logs being ingested into Sumo Logic. These logs were control plane components. This version was done as a major release as it breaks the current version of the Kubernetes App you may have installed in Sumo Logic.

After upgrading to this version, you will need to reinstall the Kubernetes App in Sumo Logic. If you do not some of the panels in the dashboards will not render properly.

If you have other content outside the app (Partitions, Scheduled Views, Field Extraction Rules or Scheduled Searches and Alerts), these may need to be updated after upgrading to v2.0.0. The logs, while the same content, have a different format and the same parsing logic and metadata may not apply.

The previous log format that is removed in v2.0.0:

{
   "timestamp": 1538776281387,
   "severity": "I",
   "pid": "1",
   "source": "wrap.go:42",
   "message": "GET /api/v1/namespaces/kube-system/endpoints/kube-scheduler: (3.514372ms) 200 [[kube-scheduler/v1.10.5 (linux/amd64) kubernetes/32ac1c9/leader-election] 127.0.0.1:46290]"
}

Is replaced by the following version. It is the same log line in a different format enriched with the same metadata the plugin applies to all pod logs.

{
   "timestamp": 1538776282152,
   "log": "I1005 21:51:21.387204       1 wrap.go:42] GET /api/v1/namespaces/kube-system/endpoints/kube-scheduler: (3.514372ms) 200 [[kube-scheduler/v1.10.5 (linux/amd64) kubernetes/32ac1c9/leader-election] 127.0.0.1:46290]",
   "stream": "stdout",
   "time": "2018-10-05T21:51:21.387477546Z",
   "docker": {
      "container_id": "a442fd2982dfdc09ab6235941f8d661a0a5c8df5e1d21f23ff48a9923ac14739"
   },
   "kubernetes": {
      "container_name": "kube-apiserver",
      "namespace_name": "kube-system",
      "pod_name": "kube-apiserver-ip-172-20-122-71.us-west-2.compute.internal",
      "pod_id": "80fa5e13-c8b9-11e8-a456-0a8c1424d0d4",
      "labels": {
         "k8s-app": "kube-apiserver"
      },
      "host": "ip-172-20-122-71.us-west-2.compute.internal",
      "master_url": "https://100.64.0.1:443/api",
      "namespace_id": "9b9b75b7-aa16-11e8-9d62-06df85b5d3bc"
   }
}