Skip to content

Aembit built its official Aembit container images to streamline the deployment process. Aembit provides a Helm chart for Kubernetes and a Terraform module for ECS that ease deployment in containerized environments.

If these are incompatible with your deployment environment, you may run into issues as you hand craft a Kubernetes configuration or an ECS task definition.

The details on this page help you as you follow your own path.

Some container images declare a specific user ID that the containerized application expects to run as. The following table lists Aembit container images and their expected user IDs:

Container ImageUser ID
aembit_agent_proxy65534
aembit_sidecar_init26248

You shouldn’t need to specify these user IDs unless you define a pod-level securityContext/runAsUser attribute in a Kubernetes deployment or extend the container image in a way that changes the default user ID. If you’ve specified the wrong user for either the aembit-agent-proxy container or the aembit-init-iptable container, you’ll see a log error message such as:

sudo: you do not exist in the passwd database

Since the v1.22 release of the Aembit Helm chart, the injected container definitions include securityContext/runAsUser attributes that will override any such pod-level attribute. Since the v1.22 release of the Aembit Agent Injector, you will see a warning message:

The injected container (...) is unlikely to run correctly because it will run as UID ... where UID ... is expected."

If you see this warning, you must make sure to specify the securityContext/runAsUser attribute for each of the Aembit containers that you are injecting into any Client Workload pods that specify a securityContext/runAsUser attribute at the pod-level.

Transparent Steering relies on the user ID of the process initiating a network connection to exempt the Agent Proxy outbound connections. Therefore any Client Workload that runs under the 65534 UID (commonly named nobody) will also be exempt from Transparent Steering.

The aembit_agent_proxy container image depends on being able to write to the root filesystem to download your tenant’s CA certificate and add it to the trusted certificate bundle. If you disable writing to the root filesystem, Agent Proxy logs an error message similar to the following:

Error when fetching token. Will attempt to refresh in 16 seconds. Error: error sending request ... invalid peer certificate: UnknownIssuer

ECS and Kubernetes use slightly different spelling, using a different letter casing, for the same setting:

  • readonlyRootFilesystem on ECS
  • readOnlyRootFilesystem on Kubernetes