Kfuse-Knight for Instrumentation-less Tracing

Kfuse Knight collects and aggregates dynamic data on nodes, using eBPF. It runs as a daemonset on a Kubernetes cluster. It can automatically trace protocol-specific network traffic to/from the containers to generate RED metrics. By default, http protocol tracing is enabled. Please contact us, for enabling collection of other protocols.

The RED metrics are exposed as a Prometheus metrics endpoint from the Kfuse Knight pod and can be discovered/scraped by Prometheus and/or Datadog agent.

Install Instructions

# Create a namespace. kubectl create namespace kfuse-knight # Get the container registry key from kloudfuse <token>.json -- (uploaded to the page) # Make sure to create a k8s secret to use this registry access token which helm will use to upgrade/install # On ubuntu (execute lines 9 through 13) cat token.json | docker login -u _json_key --password-stdin https://us-east1-docker.pkg.dev kubectl create secret docker-registry kfuse-image-pull-credentials \ --namespace='kfuse-knight' --docker-server 'us.gcr.io' --docker-username _json_key \ --docker-email 'container-registry@mvp-demo-301906.iam.gserviceaccount.com' \ --docker-password=''"$(cat token.json)"'' # On windows (execute lines 16 through 18) type token.json | helm registry login -u _json_key --password-stdin https://us-east1-docker.pkg.dev $service_json = (Get-Content token.json).replace("\n", "").replace('"', '\"') kubectl create secret docker-registry kfuse-image-pull-credentials --namespace=kfuse-knight --docker-server 'us.gcr.io' --docker-username _json_key --docker-email 'container-registry@mvp-demo-301906.iam.gserviceaccount.com' --docker-password=""$service_json"" # Get the latest knight chart version helm upgrade --install knight oci://us-east1-docker.pkg.dev/mvp-demo-301906/kfuse-helm/kfuse-knight --version=0.1.0-f381c69 -n kfuse-knight

 

Customizing the metrics to emit

The Kfuse-Knight helm values can be customized to control protocols traced by knight and any information extracted and tagged into the generated metrics. The config section of the helm values contains a section of what gets extracted by default. It can be adjusted accordingly.

 

config: http: metrics: # The tags to include/exclude when emitting logs # Exact field names can be specified. # Regex pattern can be specified. # If both includes and excludes are given, then a field is only included # if it matches an includes rule and does not match an exclude rule.. # If a field is json formatted, then subfields can be extracted using # the following format: field_name$JSON$<json pointer> tags_include: - request_type - response_code - request_headers$JSON$/Accept - request_headers$JSON$/Host tags_exclude:

 

Scraping of Metrics

Knight exposes a prometheus endpoint for its collected metrics. However, it is recommended to use Datadog agent to scrape the metrics, in order to automatically enrich the collected metrics with cloud and Kubernetes labels. In addition to the the configurations specified in https://kloudfuse.atlassian.net/wiki/spaces/EX/pages/753991796, add the following configuration in the Datadog agent yaml file.

 

datadog: kubeStateMetricsEnabled: true kubeStateMetricsCore: enabled: true ignoreLegacyKSMCheck: false prometheusScrape: enabled: true version: 1 additionalConfigs: - configurations: - send_monotonic_counter: false send_distribution_counts_as_monotonic: false send_distribution_sums_as_monotonic: false send_histograms_buckets: true max_returned_metrics: 999999 min_collection_interval: 15 autodiscovery: kubernetes_annotations: exclude: app: knight prometheus.io/scrape: "false" - configurations: - send_monotonic_counter: false send_distribution_counts_as_monotonic: false send_distribution_sums_as_monotonic: false send_histograms_buckets: true max_returned_metrics: 999999 min_collection_interval: 60 autodiscovery: kubernetes_container_names: - knight