Istio
editIstio
editThe instructions in this section describe how to connect the operator and managed resources to the Istio service mesh and assume that Istio is already installed and configured on your Kubernetes cluster. To know more about Istio and how to install it, see the product documentation.
These instructions have been tested with Istio 1.6.1. Older or newer versions of Istio might require additional configuration steps not documented here.
Some Elastic Stack features such as Kibana alerting and actions rely on the Elasticsearch API keys feature which requires TLS to be enabled at the application level. If you want to use these features, you should not disable the self-signed certificate on the Elasticsearch resource and enable PERMISSIVE
mode for the Elasticsearch service through a DestinationRule
or PeerAuthentication
resource. Strict mTLS mode is currently not compatible with Elastic Stack features requiring TLS to be enabled for the Elasticsearch HTTP layer.
If you use a Kubernetes distribution like Minikube, which does not have support for issuing third-party security tokens, you should explicitly set automountServiceAccountToken
field to true
in the Pod templates to allow Istio to fallback to default service account tokens. Refer to Istio security best practices for more information.
Connect the operator to the Istio service mesh
editThe operator itself must be connected to the service mesh to deploy and manage Elastic Stack resources that you wish to connect to the service mesh. This is achieved by injecting an Istio sidecar to the ECK operator Pods. The following instructions assume that automatic sidecar injection is enabled on your cluster through a mutating admissions webhook. Refer to Istio injection documentation if you prefer a different method of injection.
-
Create the
elastic-system
namespace and enable sidecar injection:kubectl create namespace elastic-system kubectl label namespace elastic-system istio-injection=enabled
-
Install ECK:
kubectl apply -f https://download.elastic.co/downloads/eck/1.6.0/all-in-one.yaml
-
Check the configuration and make sure the installation has been successful:
kubectl get pod elastic-operator-0 -n elastic-system -o=jsonpath='{range .spec.containers[*]}{.name}{"\n"}'
If the output of the above command contains both manager
and istio-proxy
, ECK was successfully installed with the Istio sidecar injected.
To make the validating webhook work under Istio, you need to exclude the inbound port 9443 from being proxied. This can be done by editing the template definition of the elastic-operator
StatefulSet to add the following annotations to the operator Pod:
[...] spec: template: metadata: annotations: traffic.sidecar.istio.io/excludeInboundPorts: "9443" traffic.sidecar.istio.io/includeInboundPorts: '*' [...]
As the default failurePolicy
of the webhook is Ignore
, the operator continues to function even if the above annotations are not present. The downside is that you are still able to submit an invalid manifest using kubectl
without receiving any immediate feedback.
ECK has a fallback validation mechanism that reports validation failures as events associated with the relevant resource (Elasticsearch, Kibana, APM Server, Enterprise Search, Beats, Elastic Agent and Elastic Maps Service) that must be manually discovered by running kubectl describe
. For example, to find the validation errors in an Elasticsearch resource named quickstart
, you can run kubectl describe elasticsearch quickstart
.
Connect Elastic Stack applications to the Istio service mesh
editThis section assumes that you are deploying ECK custom resources to a namespace that has automatic sidecar injection enabled.
If you have configured Istio in permissive mode, examples defined elsewhere in the ECK documentation will continue to work without requiring any modifications. However, if you have enabled strict mutual TLS authentication between services either via global (MeshPolicy
) or namespace-level (Policy
) configuration, the following modifications to the resource manifests are necessary for correct operation.
Elasticsearch
editapiVersion: elasticsearch.k8s.elastic.co/v1 kind: Elasticsearch metadata: name: elastic-istio spec: version: 8.16.1 http: tls: selfSignedCertificate: disabled: true nodeSets: - name: default count: 3 podTemplate: metadata: annotations: traffic.sidecar.istio.io/includeInboundPorts: "*" traffic.sidecar.istio.io/excludeOutboundPorts: "9300" traffic.sidecar.istio.io/excludeInboundPorts: "9300" spec: automountServiceAccountToken: true
Disable the default self-signed certificate generated by the operator and allow TLS to be managed by Istio. Disabling the self-signed certificate might interfere with some features such as Kibana Alerting and Actions. |
|
Exclude the transport port (port 9300) from being proxied. Currently ECK does not support switching off X-Pack security and TLS for the Elasticsearch transport port. If Istio is allowed to proxy the transport port, the traffic is encrypted twice and communication between Elasticsearch nodes is disrupted. |
|
Optional. Only set |
If you do not have automatic mutual TLS enabled, you may need to create a Destination Rule to allow the operator to communicate with the Elasticsearch cluster. A communication issue between the operator and the managed Elasticsearch cluster can be detected by looking at the operator logs to see if there are any errors reported with the text 503 Service Unavailable
.
kubectl logs -f -n elastic-system -c manager statefulset.apps/elastic-operator
If the operator logs indicate a communications problem, create a DestinationRule
to enable mutual TLS between the operator and the affected Elasticsearch cluster. For example, the following rule enables mutual TLS for a specific Elasticsearch cluster named elastic-istio
deployed to the default
namespace.
apiVersion: networking.istio.io/v1alpha3 kind: DestinationRule metadata: name: elastic-istio spec: host: "elastic-istio-es-http.default.svc.cluster.local" trafficPolicy: tls: mode: ISTIO_MUTUAL
Refer to the Istio documentation for more information about other configuration options affecting authentication between services.
Using init containers with Istio CNI
editThere are known issues with init containers when Istio CNI is configured. If you use init containers to install Elasticsearch plugins or perform other initialization tasks that require network access, they may fail due to outbound traffic being blocked by the CNI plugin. To work around this issue, explicitly allow the external ports used by the init containers.
To install plugins using an init container, use a manifest similar to the following:
apiVersion: elasticsearch.k8s.elastic.co/v1 kind: Elasticsearch metadata: name: elastic-istio spec: version: 8.16.1 http: tls: selfSignedCertificate: disabled: true nodeSets: - name: default count: 3 podTemplate: metadata: annotations: traffic.sidecar.istio.io/includeInboundPorts: "*" traffic.sidecar.istio.io/excludeOutboundPorts: "9300,443" traffic.sidecar.istio.io/excludeInboundPorts: "9300" spec: automountServiceAccountToken: true initContainers: - name: install-plugins command: - sh - -c - | bin/elasticsearch-plugin install --batch repository-gcs
Kibana
editAPM Server
editapiVersion: apm.k8s.elastic.co/v1 kind: ApmServer metadata: name: elastic-istio spec: version: 8.16.1 count: 1 elasticsearchRef: name: elastic-istio http: tls: selfSignedCertificate: disabled: true podTemplate: metadata: annotations: sidecar.istio.io/rewriteAppHTTPProbers: "true" spec: automountServiceAccountToken: true
Disable the default self-signed certificate generated by the operator and allow TLS to be managed by Istio. |
|
Automatically re-write the health checks to go through the proxy. |
|
Optional. Only set |