Application Requirements

Istio provides a great deal of functionality to applications with little or no impact on the application code itself. Many Kubernetes applications can be deployed in an Istio-enabled cluster without any changes at all. However, there are some implications of Istio’s sidecar model that may need special consideration when deploying an Istio-enabled application. This document describes these application considerations and specific requirements of Istio enablement.

Pod requirements

To be part of a mesh, Kubernetes pods must satisfy the following requirements:

  • Service association: A pod must belong to at least one Kubernetes service even if the pod does NOT expose any port. If a pod belongs to multiple Kubernetes services, the services cannot use the same port number for different protocols, for instance HTTP and TCP.

  • Application UIDs: Ensure your pods do not run applications as a user with the user ID (UID) value of 1337 because 1337 is reserved for the sidecar proxy.

  • NET_ADMIN and NET_RAW capabilities: If pod security policies are enforced in your cluster and unless you use the Istio CNI Plugin, your pods must have the NET_ADMIN and NET_RAW capabilities allowed. The initialization containers of the Envoy proxies require these capabilities.

    To check if the NET_ADMIN and NET_RAW capabilities are allowed for your pods, you need to check if their service account can use a pod security policy that allows the NET_ADMIN and NET_RAW capabilities. If you haven’t specified a service account in your pods’ deployment, the pods run using the default service account in their deployment’s namespace.

    To list the capabilities for a service account, replace <your namespace> and <your service account> with your values in the following command:

    $ for psp in $(kubectl get psp -o jsonpath="{range .items[*]}{@.metadata.name}{'\n'}{end}"); do if [ $(kubectl auth can-i use psp/$psp --as=system:serviceaccount:<your namespace>:<your service account>) = yes ]; then kubectl get psp/$psp --no-headers -o=custom-columns=NAME:.metadata.name,CAPS:.spec.allowedCapabilities; fi; done
    

    For example, to check for the default service account in the default namespace, run the following command:

    $ for psp in $(kubectl get psp -o jsonpath="{range .items[*]}{@.metadata.name}{'\n'}{end}"); do if [ $(kubectl auth can-i use psp/$psp --as=system:serviceaccount:default:default) = yes ]; then kubectl get psp/$psp --no-headers -o=custom-columns=NAME:.metadata.name,CAPS:.spec.allowedCapabilities; fi; done
    

    If you see NET_ADMIN and NET_RAW or * in the list of capabilities of one of the allowed policies for your service account, your pods have permission to run the Istio init containers. Otherwise, you will need to provide the permission.

  • Pods with app and version labels: We recommend adding an explicit app label and version label to the specification of the pods deployed using a Kubernetes Deployment. The app and version labels add contextual information to the metrics and telemetry that Istio collects.

    • The app label: Each deployment should have a distinct app label with a meaningful value. The app label is used to add contextual information in distributed tracing.

    • The version label: This label indicates the version of the application corresponding to the particular deployment.

  • Named service ports: Service ports may optionally be named to explicitly specify a protocol. See Protocol Selection for more details.

Ports used by Istio

The following ports and protocols are used by the Istio sidecar proxy (Envoy).

PortProtocolDescriptionPod-internal only
15000TCPEnvoy admin port (commands/diagnostics)Yes
15001TCPEnvoy outboundNo
15006TCPEnvoy inboundNo
15008TCPEnvoy tunnel port (inbound)No
15020HTTPMerged Prometheus telemetry from Istio agent, Envoy, and applicationNo
15021HTTPHealth checksNo
15090HTTPEnvoy Prometheus telemetryNo

The following ports and protocols are used by the Istio control plane (istiod).

PortProtocolDescriptionLocal host only
15010GRPCXDS and CA services (Plaintext)No
15012GRPCXDS and CA services (TLS, recommended for production use)No
8080HTTPDebug interface (deprecated)No
443HTTPSWebhooksNo
15014HTTPControl plane monitoringNo

Server First Protocols

Some protocols are “Server First” protocols, which means the server will send the first bytes. This may have an impact on PERMISSIVE mTLS and Automatic protocol selection.

Both of these features work by inspecting the initial bytes of a connection to determine the protocol, which is incompatible with server first protocols.

In order to support these cases, follow the Explicit protocol selection steps to declare the protocol of the application as TCP.

The following ports are known to commonly carry server first protocols, and are automatically assumed to be TCP:

ProtocolPort
SMTP25
DNS53
MySQL3306
MongoDB27017

Because TLS communication is not server first, TLS encrypted server first traffic will work with automatic protocol detection as long as you make sure that all traffic subjected to TLS sniffing is encrypted:

  1. Configure mTLS mode STRICT for the server. This will enforce TLS encryption for all requests.
  2. Configure mTLS mode DISABLE for the server. This will disable the TLS sniffing, allowing server first protocols to be used.
  3. Configure all clients to send TLS traffic, generally through a DestinationRule or by relying on auto mTLS.
  4. Configure your application to send TLS traffic directly.

Outbound traffic

In order to support Istio’s traffic routing capabilities, traffic leaving a pod may be routed differently than when a sidecar is not deployed.

For HTTP based traffic, traffic is routed based on the Host header. This may lead to unexpected behavior if the destination IP and Host header are not aligned. For example, a request like curl 1.2.3.4 -H "Host: httpbin.default" will be routed to the httpbin service, rather than 1.2.3.4.

For Non-HTTP based traffic (including HTTPS), Istio does not have access to an Host header, so routing decisions are based on the Service IP address.

One implication of this is that direct calls to pods (for example, curl <POD_IP>), rather than Services, will not be matched. While the traffic may be passed through, it will not get the full Istio functionality including mTLS encryption, traffic routing, and telemetry.

Was this information useful?
Do you have any suggestions for improvement?

Thanks for your feedback!