eck-diagnostics

Diagnostic tooling for ECK installations

Installation

Go to the releases page and download the version matching your architecture. Unpack the gzip'ed tar archive and put the binary included in the archive somewhere in your PATH.

Running

Just execute the binary. On macOS versions with Gatekeeper enabled you have to explicitly allow the execution the first time round, as described in this support article (in the section "If you want to open an app that hasn’t been notarized or is from an unidentified developer").

To run diagnostics, you need to specify at least the workload resource namespace(s). The operator namespace is set by default to the elastic-system namespace, where the ECK operator typically resides.

For example, to run diagnostics for resource namespaces a and b:

eck-diagnostics -r a,b

A full list of available flags is reproduced here and is also printed when calling the eck-diagnostics binary with the --help or -h flag:

Usage:
  eck-diagnostics [flags]

Flags:
      --diagnostic-image string              Diagnostic image to be used for stack diagnostics, see run-stack-diagnostics (default "docker.elastic.co/eck-dev/support-diagnostics:8.4.0")
      --eck-version string                   ECK version in use, will try to autodetect if not specified
  -f, --filters strings                      Comma-separated list of filters in format "type=name". ex: elasticsearch=my-cluster (Supported types [agent apm beat elasticsearch enterprisesearch kibana maps])
  -h, --help                                 help for eck-diagnostics
      --kubeconfig string                    optional path to kube config, defaults to $HOME/.kube/config
  -o, --operator-namespaces strings          Comma-separated list of namespace(s) in which operator(s) are running (default [elastic-system])
      --output-directory string              Path where to output diagnostic results
  -r, --resources-namespaces strings         Comma-separated list of namespace(s) in which resources are managed
      --run-agent-diagnostics                Run diagnostics on deployed Elastic Agents. Warning: credentials will not be redacted and appear as plain text in the archive
      --run-stack-diagnostics                Run diagnostics on deployed Elasticsearch clusters and Kibana instances, requires deploying diagnostic Pods into the cluster (default true)
      --stack-diagnostics-timeout duration   Maximum time to wait for Elaticsearch and Kibana diagnostics to complete (default 5m0s)
      --verbose                              Verbose mode

Information collected by eck-diagnostics

The eck-diagnostics retrieves Kubernetes API server resources and log files and, unless disabled, it runs Elastic support-diagnostics on Elasticsearch and Kibana instances installed in the namespaces indicated by the -r, --resources-namespaces flag. In addition it can optionally run the diagnostic subcommand on all running Elastic Agents. Please note that credentials are not redacted in the Elastic Agent diagnostics and may appear in plain text inside the diagnostics archive. Elastic Agent diagnostics are therefore disabled by default.

The following Kubernetes resources are retrieved from the cluster being diagnosed:

Global resources

• Kubernetes server version
• Kubernetes nodes
• PodSecurityPolicy
• ClusterRole
• ClusterRoleBindings
• StorageClass

In all operator and workload resource namespaces

• StatefulSet
• Pod
• Service
• ConfigMap
• Event
• NetworkPolicy
• ControllerRevision
• Secret (metadata only)
• ServiceAccount

In the workload resources namespaces

In addition to the resources mentioned above, the following Kubernetes resources are retrieved:

• ReplicaSet
• Deployment
• DaemonSet
• PersistentVolume
• PersistentVolumeClaim
• Endpoint

The ECK related custom resources are included in those namespaces as well:

• Agent
• ApmServer
• Beat
• ElasticMapsServer
• Elasticsearch
• EnterpriseSearch
• Kibana

Logs

In the operator namespaces (-o, --operator-namespaces) all logs are collected, while in the workload resource namespaces only logs from Pods managed by ECK are collected.

Filtering collected resources

The resources in the specified namespaces that are collected by eck-diagnostics can be filtered with the -f, --filters flag.

Usage Example

The following example will run the diagnostics for Elastic resources in namespace a, and will only return resources associated with either an Elasticsearch cluster named mycluster or a Kibana instance named my-kb.

eck-diagnostics -r a -f "elasticsearch=mycluster" -f "kibana=my-kb"

Filtered resources

Only a certain number of Kubernetes resources support filtering when filtering by an Elastic custom resource. Along with the named Elastic custom resource type, the following resources will be returned that are associated:

• ConfigMap
• ControllerRevision
• Deployment
• DaemonSet
• Endpoint
• Pod
• PersistentVolumeClaim
• Replicaset
• Service
• StatefulSet

The following resources are returned unfiltered:

• Event
• NetworkPolicy
• PersistentVolume
• ServiceAccount
• Secret (metadata only)