dhi.io/kubeflow-pipelines-inverse-proxy-agent
The Kubeflow Pipelines inverse proxy agent connects the Pipelines UI in a Kubernetes cluster to a Google Cloud inverting proxy, so the UI can be reached without exposing a public endpoint.
All examples in this guide use the public image. If you've mirrored the repository for your own use (for example, to your Docker Hub namespace), update your commands to reference the mirrored image instead of the public one.
For example:
dhi.io/<repository>:<tag><your-namespace>/dhi-<repository>:<tag>For the examples, you must first use docker login dhi.io to authenticate to the registry to pull the images.
This Docker Hardened kubeflow-pipelines-inverse-proxy-agent image includes:
proxy-forwarding-agent — The inverting-proxy agent, which polls the proxy for requests and forwards them to the
Kubeflow Pipelines UI service.gcloud — Google Cloud CLI, used by the launcher to activate a service account and mint access tokens.kubectl — Kubernetes CLI, used by the launcher to read and update the inverse-proxy-config ConfigMap.The default command runs the upstream launcher, /opt/proxy/attempt-register-vm-on-proxy.sh. The launcher is a Bash
script, so unlike most hardened runtime images, the runtime variants include bash, sh, curl, jq, and python3.
They don't include a package manager.
The launcher only works inside a Kubernetes pod running on Google Cloud. It needs the Kubernetes API, the Google Compute Engine metadata server, and the Kubeflow Pipelines UI Service. Outside such a pod, the default command exits with an error. To check the image locally, run the bundled tools directly:
docker run --rm --entrypoint /opt/bin/proxy-forwarding-agent \
dhi.io/kubeflow-pipelines-inverse-proxy-agent:<tag> --help
docker run --rm --entrypoint kubectl \
dhi.io/kubeflow-pipelines-inverse-proxy-agent:<tag> version --client
docker run --rm --entrypoint gcloud \
dhi.io/kubeflow-pipelines-inverse-proxy-agent:<tag> --version
The launcher works with the inverse-proxy-config ConfigMap in the pod's own namespace:
Hostname key, the launcher logs Reuse existing hostname, reads ProxyUrl and
BackendId from the ConfigMap, and starts the agent. The UI keeps the same URL across pod restarts and image
changes.metadata.google.internal) and gets an access token with gcloud. It then writes Hostname,
ProxyUrl, and BackendId to the ConfigMap and starts the agent.In both cases, the agent forwards requests to ML_PIPELINE_UI_SERVICE_HOST:ML_PIPELINE_UI_SERVICE_PORT. Kubernetes sets
these variables in pods that start after the ml-pipeline-ui Service exists in the same namespace. The agent
authenticates to the proxy with Google Application Default Credentials: the node's metadata server credentials, or the
key file in GOOGLE_APPLICATION_CREDENTIALS.
When PROXY_URL isn't set, registration first downloads the list of proxy servers from
https://storage.googleapis.com/ml-pipeline/proxy-agent-config.json. At the time of writing, that URL returns HTTP 403,
so a new registration only succeeds when you set PROXY_URL. This is an upstream condition and affects the upstream
image in the same way. Installations that are already registered aren't affected, because they reuse the ConfigMap.
| Variable | Description | Default | Required |
|---|---|---|---|
PROXY_URL | Inverting proxy URL to register with. Skips the proxy list download. Ignored when the ConfigMap already has a Hostname. | unset | No |
GOOGLE_APPLICATION_CREDENTIALS | Path to a service account key file. The launcher activates it with gcloud auth activate-service-account, and the agent uses it to authenticate. | unset | No |
ML_PIPELINE_UI_SERVICE_HOST, ML_PIPELINE_UI_SERVICE_PORT | Backend address of the Pipelines UI. Kubernetes sets these from the ml-pipeline-ui Service. | set by Kubernetes | Yes |
DEBUG | Enables agent debug logging. | false | No |
PROXY_TIMEOUT | Timeout for agent requests to the proxy. | 60s | No |
HEALTH_CHECK_PATH | Path on the UI backend for agent health checks. | / | No |
HEALTH_CHECK_INTERVAL_SECONDS | Seconds between health checks. 0 disables them. | 0 | No |
HEALTH_CHECK_UNHEALTHY_THRESHOLD | Consecutive failed checks before the backend is marked unhealthy. | 2 | No |
To pull the image in Kubernetes, first follow the
authentication instructions for DHI in Kubernetes. The
examples assume that Kubeflow Pipelines runs in the kubeflow namespace.
The upstream Google Cloud manifests (manifests/kustomize/env/gcp) deploy the agent as the proxy-agent Deployment
with the proxy-agent-runner ServiceAccount. Add your pull secret to the ServiceAccount and replace the image. Use a
tag that matches your Kubeflow Pipelines version.
kubectl -n kubeflow patch serviceaccount proxy-agent-runner \
-p '{"imagePullSecrets": [{"name": "<secret name>"}]}'
kubectl -n kubeflow set image deployment/proxy-agent \
proxy-agent=dhi.io/kubeflow-pipelines-inverse-proxy-agent:<tag>
The upstream Deployment doesn't set a security context, so the agent runs as the image's nonroot user (UID 65532). The new pod reuses the existing registration, and the UI URL doesn't change:
kubectl -n kubeflow logs deployment/proxy-agent | grep 'Reuse existing hostname'
kubectl -n kubeflow get configmap inverse-proxy-config -o jsonpath='{.data.Hostname}'
The following manifest deploys the agent next to an existing ml-pipeline-ui Service. It keeps the upstream
hostNetwork: true setting, and sets the security context to the image's numeric user. The Role allows only get and
patch on the inverse-proxy-config ConfigMap, which is all the launcher uses. Upstream grants every verb on every
ConfigMap in the namespace.
apiVersion: v1
kind: ServiceAccount
metadata:
name: proxy-agent-runner
namespace: kubeflow
imagePullSecrets:
- name: <secret name>
---
apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
name: proxy-agent-runner
namespace: kubeflow
rules:
- apiGroups: [""]
resources: ["configmaps"]
resourceNames: ["inverse-proxy-config"]
verbs: ["get", "patch"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
name: proxy-agent-runner
namespace: kubeflow
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: Role
name: proxy-agent-runner
subjects:
- kind: ServiceAccount
name: proxy-agent-runner
namespace: kubeflow
---
apiVersion: v1
kind: ConfigMap
metadata:
name: inverse-proxy-config
namespace: kubeflow
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: proxy-agent
namespace: kubeflow
labels:
app: proxy-agent
spec:
selector:
matchLabels:
app: proxy-agent
template:
metadata:
labels:
app: proxy-agent
spec:
hostNetwork: true
serviceAccountName: proxy-agent-runner
securityContext:
runAsNonRoot: true
runAsUser: 65532
runAsGroup: 65532
containers:
- name: proxy-agent
image: dhi.io/kubeflow-pipelines-inverse-proxy-agent:<tag>
securityContext:
allowPrivilegeEscalation: false
capabilities:
drop: ["ALL"]
The container needs no command or args. The image's default command runs the launcher.
Because the default proxy list download fails, set PROXY_URL before the first registration:
kubectl -n kubeflow set env deployment/proxy-agent PROXY_URL=<inverting proxy URL>
The launcher ignores PROXY_URL while the ConfigMap holds a Hostname. To register again, clear the ConfigMap and
restart the agent. The new registration can return a different hostname, which changes the UI URL.
kubectl -n kubeflow patch configmap inverse-proxy-config --type json \
-p '[{"op": "remove", "path": "/data"}]'
kubectl -n kubeflow rollout restart deployment/proxy-agent
nonroot (UID 65532). A key file mounted for
GOOGLE_APPLICATION_CREDENTIALS must be readable by that user. gcloud keeps its state in
/home/nonroot/.config/gcloud instead of /root/.config/gcloud./opt/proxy is owned by root. The launcher's only write there is the proxy list download to
/opt/proxy/proxy-agent-config.json, which the nonroot user can write.linux/amd64 only. This image is published for linux/amd64 and
linux/arm64.In the -fips and -fips-dev variants, gcloud, curl, and the launcher's Python helper use the system OpenSSL with
its FIPS provider. proxy-forwarding-agent and kubectl are built with the Go Cryptographic Module (FIPS 140-3). One
exception: with a service account key file in GOOGLE_APPLICATION_CREDENTIALS, gcloud auth activate-service-account
signs with gcloud's bundled pure-Python rsa library, which isn't a FIPS-validated module. In FIPS-bound environments,
use the node's metadata server credentials instead of a key file.
Docker Hardened Images come in different variants depending on their intended use. Image variants are identified by their tag.
Runtime variants are designed to run your application in production. These images are intended to be used either directly or as the FROM image in the final stage of a multi-stage build. These images typically:
Build-time variants typically include dev in the tag name and are intended for use in the first stage of a
multi-stage Dockerfile. These images typically:
FIPS variants include fips in the variant name and tag. They come in both runtime and build-time variants. These
variants use cryptographic modules that have been validated under FIPS 140, a U.S. government standard for secure
cryptographic operations. For example, usage of MD5 fails in FIPS variants.
To view the image variants and get more information about them, select the Tags tab for this repository, and then select a tag.
To migrate your application to a Docker Hardened Image, you must update your Dockerfile. At minimum, you must update the base image in your existing Dockerfile to a Docker Hardened Image. This and a few other common changes are listed in the following table of migration notes.
| Item | Migration note |
|---|---|
| Base image | Replace your base images in your Dockerfile with a Docker Hardened Image. |
| Package management | Non-dev images, intended for runtime, don't contain package managers. Use package managers only in images with a dev tag. |
| Non-root user | By default, non-dev images, intended for runtime, run as the nonroot user. Ensure that necessary files and directories are accessible to the nonroot user. |
| Multi-stage build | Utilize images with a dev tag for build stages and non-dev images for runtime. For binary executables, use a static image for runtime. |
| TLS certificates | Docker Hardened Images contain standard TLS certificates by default. There is no need to install TLS certificates. |
| Ports | Non-dev hardened images run as a nonroot user by default. As a result, applications in these images can't bind to privileged ports (below 1024) when running in Kubernetes or in Docker Engine versions older than 20.10. To avoid issues, configure your application to listen on port 1025 or higher inside the container. |
| Entry point | Docker Hardened Images may have different entry points than images such as Docker Official Images. Inspect entry points for Docker Hardened Images and update your Dockerfile if necessary. |
| No shell | By default, non-dev images, intended for runtime, don't contain a shell. Use dev images in build stages to run shell commands and then copy artifacts to the runtime stage. |
The following steps outline the general migration process.
Find hardened images for your app.
A hardened image may have several variants. Inspect the image tags and find the image variant that meets your needs.
Update the base image in your Dockerfile.
Update the base image in your application's Dockerfile to the hardened image you found in the previous step. For
framework images, this is typically going to be an image tagged as dev because it has the tools needed to install
packages and dependencies.
For multi-stage Dockerfiles, update the runtime image in your Dockerfile.
To ensure that your final image is as minimal as possible, you should use a multi-stage build. All stages in your
Dockerfile should use a hardened image. While intermediary stages will typically use images tagged as dev, your
final runtime stage should use a non-dev image variant.
Install additional packages
Docker Hardened Images contain minimal packages in order to reduce the potential attack surface. You may need to install additional packages in your Dockerfile. Inspect the image variants to identify which packages are already installed.
Only images tagged as dev typically have package managers. You should use a multi-stage Dockerfile to install the
packages. Install the packages in the build stage that uses a dev image. Then, if needed, copy any necessary
artifacts to the runtime stage that uses a non-dev image.
For Alpine-based images, you can use apk to install packages. For Debian-based images, you can use apt-get to
install packages.
The following are common issues that you may encounter during migration.
The hardened images intended for runtime don't contain a shell nor any tools for debugging. The recommended method for debugging applications built with Docker Hardened Images is to use Docker Debug to attach to these containers. Docker Debug provides a shell, common debugging tools, and lets you install other tools in an ephemeral, writable layer that only exists during the debugging session.
By default image variants intended for runtime, run as the nonroot user. Ensure that necessary files and directories are accessible to the nonroot user. You may need to copy files to different directories or change permissions so your application running as the nonroot user can access them.
Non-dev hardened images run as a nonroot user by default. As a result, applications in these images can't bind to
privileged ports (below 1024) when running in Kubernetes or in Docker Engine versions older than 20.10. To avoid issues,
configure your application to listen on port 1025 or higher inside the container, even if you map it to a lower port on
the host. For example, docker run -p 80:8080 my-image will work because the port inside the container is 8080, and
docker run -p 80:81 my-image won't work because the port inside the container is 81.
By default, image variants intended for runtime don't contain a shell. Use dev images in build stages to run shell
commands and then copy any necessary artifacts into the runtime stage. In addition, use Docker Debug to debug containers
with no shell.
Docker Hardened Images may have different entry points than images such as Docker Official Images. Use docker inspect
to inspect entry points for Docker Hardened Images and update your Dockerfile if necessary.