Configuring Logging in AWS EKS Using Fluent Bit and CloudWatch

Daniel German Rivera - Oct 6 - - Dev Community

This article is part of a personal project called Smart-cash. Previous posts covered the deployment of AWS and Kubernetes resources and how to install FluxCD to implement GitOps practices.

Observability is essential for any application, and the Smart-cash project is no exception. Previously Prometheus was integrated for monitoring.

Project Source Code

The full project code can be found here, the project is still under development but you can find the terraform code to create AWS resources and also the Kubernetes manifest.

In this link you can find the files used in this post

Option 1. Export logs directly to Cloudwatch Logs(No Cloudwatch add-on)

The simplest configuration involves using Fluent-Bit's Tail Input, which reads the logs in the host /var/log/containers/*.log and sends them to Cloudwatch. This approach could be enough if you want to centralize the logs in CloudWatch or maybe another platform.

Fluent-Bit installation

The Fluent-Bit Helm chart will be used in combination with FluxCD.

Adding the FluxCD source:

kind: HelmRepository
metadata:
  name: fluent-bit
  namespace: flux-system
spec:
  interval: 10m0s
  url: https://fluent.github.io/helm-charts
Enter fullscreen mode Exit fullscreen mode

Adding the Helm chart

The supported values for the Fluent-Bit Helm chart can be found here.

To use Fluent Bit with AWS, the following requirements must be met:

  • IAM Roles for Service Accounts (IRSA): You must set up an IAM role with permissions to create CloudWatch Logs streams and write logs. This role should be associated with the service account that Fluent Bit uses. AWS EKS Pod identity is also an option.
  • CloudWatch Log Group: You can either create the CloudWatch Log group in advance or allow Fluent Bit to handle the log group creation.

The main configuration is shown below (some lines have been omitted for brevity). The full configuration file can be found link

Fluent-Bit will run as a DaemonSet. The Helm chart will create the RBAC and the Service account, named fluent-bit, which will be annotated with the AWS IAM role with the appropriate CloudWatch permissions.

apiVersion: helm.toolkit.fluxcd.io/v2beta1
kind: HelmRelease
###ommited-lines
spec:
  chart:
    spec:
      chart: fluent-bit
      version: 0.47.9
      ###ommited-lines
  values:
    kind: DaemonSet
    ###ommited-lines
    serviceAccount:
      create: true
      annotations: 
        eks.amazonaws.com/role-arn: arn:aws:iam::${ACCOUNT_NUMBER}:role/role-fluent-bit-${ENVIRONMENT}
    rbac:
      create: true
    ###ommited-lines
Enter fullscreen mode Exit fullscreen mode

A volume is needed for Filesystem buffering, which helps to manage backpressure and overall memory control, also this is used to store the position (or offsets) of the log files being read, allowing Fluent Bit to track its progress and resume from the correct position if needed.

extraVolumes:
      - name: fluentbit-status
        hostPath:
          path: /var/fluent-bit/state
    extraVolumeMounts:
      - name: fluentbit-status
        mountPath: /var/fluent-bit/state
Enter fullscreen mode Exit fullscreen mode

Configs section defines the Inputs, Filters, and Outputs for collecting and processing data. For this scenario an Input of Tail type is configured to read the content of the files located at /var/log/containers/*.log. Let's break down the configuration details:

Note: AWS has some advanced configurations for inputs, filters and output, you can check this link

The service section defines the global properties for Fluent-Bit, for this case:

  • Flush Interval: Set to 1 second, meaning Fluent Bit will send the collected logs to the configured output destinations every second.

  • Log Level: Set to Info, which includes informational messages as well as warnings and errors.

  • The storage path for Filesystem buffering is the volume mounted in previous configurations with a backlog memory limit of 5M, which means that if Fluent-bit service reaches this limit, it stops loading any more backlog chunks from the storage path into memory.

config:
  service: |
    [SERVICE]
      Daemon Off
      Flush  1
      Log_Level  info
      Parsers_File /fluent-bit/etc/parsers.conf
      Parsers_File /fluent-bit/etc/conf/custom_parsers.conf
      HTTP_Server On
      HTTP_Listen 0.0.0.0
      HTTP_Port  \{\{ .Values.metricsPort \}\}
      Health_Check On
      storage.path  /var/fluent-bit/state/flb-storage/
      storage.sync              normal
      storage.checksum          off
      storage.backlog.mem_limit 5M
Enter fullscreen mode Exit fullscreen mode

Inputs define the data sources that Fluent Bit will collect logs from. In this scenario, the Tail input is used, which allows Fluent Bit to monitor one or more text files. Key points for this configuration include:

  • In common Kubernetes environments, container runtimes store logs in the /var/log/pod/ and /var/log/containers/ (containers directory has symlinks to pod directory). Each log file follows a naming convention that includes key information like the pod name, namespace, container name, and container ID, for this case entries for fluent-bit, cloudwath-agent,kube-proxy, and aws-node will be ignored
  • Some log entries may span multiple lines. Fluent Bit handles multi-line logs with built-in modes, and for this scenario, the Docker or CRI modes are used to process them correctly.
  • To track the last line read from each log file, Fluent Bit uses a database to store this position. The database is saved in the previously mounted volume, ensuring that Fluent Bit can resume reading from the correct location. 
[INPUT]
   Name                tail
   Tag                 applications.*
   Exclude_Path        /var/log/containers/cloudwatch-agent*, /var/log/containers/fluent-bit*, /var/log/containers/aws-node*, /var/log/containers/kube-proxy*
   Path                /var/log/containers/*.log
   multiline.parser    docker, cri
   DB                  /var/fluent-bit/state/flb_container.db
   Mem_Buf_Limit       50MB
   Skip_Long_Lines     On
   Refresh_Interval    10
   storage.type        filesystem
   Rotate_Wait         30
Enter fullscreen mode Exit fullscreen mode

Outputs define where the collected data is sent, and Fluent-Bit provides a plugin to send logs to CloudWatch.

  • If you check the Input configurations there is a tag defined, applications.*. this helps to assign a label to the logs collected for that Input, in this case, it ensures that logs with this tag are routed to the specified output destination.
  • CloudWatch log groups can be created by Fluent Bit, but in this scenario, the creation is disabled (set to off) since Terraform is used to manage log groups.
  • The log_stream_prefix sets a prefix for the log streams created in CloudWatch, helping organize and identify the log entries within the stream. 
[OUTPUT]
   Name cloudwatch_logs
   Match applications.*
   region ${AWS_REGION} 
   log_group_name /aws/eks/${CLUSTER_NAME}/workloads
   log_stream_prefix from-k8-fluent-bit-
   auto_create_group off
Enter fullscreen mode Exit fullscreen mode

Once you deploy the Helm chart you can check the CloudWatch service, if everything is working you should see some stream created, in this case out prefix is from-k8-fluent-bit-

Log-Stream

and the log entry

log-entry

Adding a filter

Filters in Fluent Bit allow you to enrich the data being collected. For instance, the Kubernetes Filter adds valuable metadata to log entries, such as namespace, pod_name, host, and more.

Here are some key points about the filter configuration:

  • The Tag from the input configuration is reused here to extract information like pod_name, namespace, and other relevant metadata.

  • The Kube_URL points to the Kubernetes API server, which Fluent Bit queries to obtain metadata about the pods involved in the logs. The path for the token and certificate is specified in Kube_CA_File and Kube_Token_File.

  • You can configure the filter to include annotations and labels from the pods in the log entries.

Note: Be cautious about Fluent Bit querying the API server for metadata. In clusters with a high number of resources, this can put an additional load on the API server. One optimization is to retrieve pod metadata from the node’s kubelet instead of the kube-apiserver, but this requires enabling hostNetwork in the DaemonSet

[FILTER]
   Name   kubernetes
   Match  applications.*
   Kube_URL      https://kubernetes.default.svc:443
   Kube_CA_File       /var/run/secrets/kubernetes.io/serviceaccount/ca.crt
   Kube_Token_File 
     /var/run/secrets/kubernetes.io/serviceaccount/token
   Kube_Tag_Prefix     application.var.log.containers.
   Merge_Log           On
   Merge_Log_Key       log_processed
   K8S-Logging.Parser  On
   K8S-Logging.Exclude Off
   Labels              On
   Annotations         Off
   Buffer_Size         0
Enter fullscreen mode Exit fullscreen mode

After applying this filter the logs should have pods metadata

after-filter

Option 2. Use the amazon-cloudwatch-observability add-on

Container Insights can be used to collect, aggregate, and summarize both metrics and logs. If you plan to enable this in your EKS cluster, the Amazon CloudWatch Observability add-on installs the necessary resources to achieve this.

At a high level, the add-on installs two key components:

Both components are deployed as DaemonSets.

The add-on can be installed via Terraform or by using a Helm-chart, Regardless of the method, you'll need to create an IAM role for the service account cloudwatch-agent in the amazon-cloudwatch namespace.

resource "aws_eks_addon" "cloudwatch" {
  cluster_name                = aws_eks_cluster.kube_cluster.name
  addon_name                  = "amazon-cloudwatch-observability"
  addon_version               = ""v2.1.2-eksbuild.1""
  service_account_role_arn    = aws_iam_role.cloudwatch_role.arn 
  resolve_conflicts_on_update = "OVERWRITE"
}
Enter fullscreen mode Exit fullscreen mode

The add-on creates several resources in the cluster, some of which you may not need. For example, if you list the DaemonSets in the amazon-cloudwatch namespace, you'll notice seven DaemonSets, some of which might have 0 replicas. While these resources may not be actively used, they still exist in your cluster and can create some noise.

addon-installed

You can customize the add-on configurations to suit your needs. For example, you can disable Fluent Bit logs for Accelerated Compute monitoring or skip collecting NVIDIA GPU metrics.

resource "aws_eks_addon" "cloudwatch" {
  cluster_name                = aws_eks_cluster.kube_cluster.name
  addon_name                  = "amazon-cloudwatch-observability"
  addon_version               = ""v2.1.2-eksbuild.1""
  service_account_role_arn    = aws_iam_role.cloudwatch_role.arn 
  resolve_conflicts_on_update = "OVERWRITE"
  configuration_values = jsonencode({
    containerLogs = {
        enabled = true
    },    
    agent = {
      config = {
        logs = {
          metrics_collected = {
            application_signals = {},
            kubernetes = {
              "enhanced_container_insights": true
              "accelerated_compute_metrics": false
            }}}}}
  })
}
Enter fullscreen mode Exit fullscreen mode

By default, the add-on creates four CloudWatch log groups. The following image, taken from the AWS documentation, explains the naming structure of the log groups and the type of data each group stores.

AWS-Logs-group

To change expiration days and names for the groups is better to use the Helm chart instead of the Terraform code to install the add-on. You can do this by modifying the fluent-bit outputs.

The last log group is named performance and stores metrics collected by the CloudWatch agent, such as the number of running pods, CPU usage, and memory metrics.

Bonus: Cluster dashboard

As mentioned earlier, the CloudWatch add-on collects, aggregates, and summarizes metrics. Once the add-on is installed, AWS automatically generates a dashboard that provides useful insights and metrics for your cluster.

Cluster-dashboard

You can watch metric per pod

metric-per-pod

It also generates a visual map that organizes Kubernetes resources by namespace.

Cluster-resource-mapping

Image
10 Tips For Asbestos Settlement That Are Unexpected
Image
URL Parts and Parameters in Javascript
Image
A Provocative Remark About Asbestos Compensation Payouts
Image
Mesothelioma Asbestos Claim: The Secret Life Of Mesothelioma Asbestos Claim
Image
Unveiling Spread Out AI: Automating Social Media Audience Targeting for Your Ads

ai, socialmedia, ads, webdev
Image
How to Stand Out in Your AWS Community Builders Application (And Re-Application)

aws, cloud, community, tips
Image
Introduction to Microsoft Entra ID

entraid, activedirectory, microsoft
Image
15 Secretly Funny People Work In How To Get Spare Car Key
Image
Fixing File Renaming Issues in Git: Handling Case Sensitivity and core.ignorecase

git, github, beginners, webdev
Image
Stylowe Meble do Sypialni: Jak Stworzyć Idealną Przestrzeń do Relaksu
Image
sedikit kehebohan bermain judi slots online
Image
What is a chicken? An introduction to distroless containers

dotnet, distroless, docker
Image
What To Look For To Determine If You're Ready To Electric Fireplace That Hangs On Wall
Image
Why We Do We Love Gas Boiler Engineer Near Me (And You Should Also!)
Image
Which are unique gifts for men for all the occasions?
Image
The History Of Asbestos Claims Payout
Image
Azure Network Security Groups (NSGs): Securing Your Azure Virtual Network

azure, activedirectory
Image
Cat Flap Installation - How to Choose and Install a Cat Flap
Image
Think You're Cut Out For Doing Lawsuit Asbestos? Check This Quiz
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .