Pulp fails to start on hosts that have ipv6 disabled

[plnordquist]

Version
Pulp Operator: v1.0.0-beta.3
Pulp Image: quay.io/pulp/pulp-minimal:3.44.0

Describe the bug
With a Kubernetes cluster that has hosts that include the ipv6.disable=1 option on the Kernel command line, the pulp-content and pulp-api pods fail to start as they attempt to bind to [::]. The pod logs include the following content:

[2024-01-17 00:31:53 +0000] [1] [INFO] Starting gunicorn 21.2.0
[2024-01-17 00:31:53 +0000] [1] [ERROR] Retrying in 1 second.
[2024-01-17 00:31:54 +0000] [1] [ERROR] Retrying in 1 second.
[2024-01-17 00:31:55 +0000] [1] [ERROR] Retrying in 1 second.
[2024-01-17 00:31:56 +0000] [1] [ERROR] Retrying in 1 second.
[2024-01-17 00:31:57 +0000] [1] [ERROR] Retrying in 1 second.
[2024-01-17 00:31:58 +0000] [1] [ERROR] Can't connect to ('::', 24817)

To Reproduce
Steps to reproduce the behavior:

1. Setup a Kubernetes cluster with hosts that have ipv6.disable=1 in the Kernel command line via GRUB_CMDLINE or grubby --update-kernel=ALL --args="ipv6.disable=1"
2. Deploy the pulp operator and deploy a pulp CR
3. Observe the pulp-api and pulp-content pods terminating

Expected behavior
Pulp should be deployable on systems that disable IPv6.

Additional context
As a workaround, I was able to mark the Pulp CR as unmanaged and edit the deployments to use /bin/sh as an entrypoint and command that includes an alternate bind address similar to how the telemetry system in the operator updates the pods at https://github.com/pulp/pulp-operator/blob/main/controllers/telemetry.go#L35-L48.

Another workaround I found was to create a Dockerfile from the quay.io/pulp/pulp-minimal:3.44.0 image and update the entrypoints at /usr/bin/pulp-api and /usr/bin/pulp-content to use an alternate bind address in both statements of the PULP_API_ENTRYPOINT and PULP_CONTENT_ENTRYPOINT environment variables.

[markleary]

I also ran into this issue and used a similar workaround to rebuild the pulp-minimal image. However, pulp-web also binds to ipv6 using a nginx configmap which is hardcoded here:

https://github.com/pulp/pulp-operator/blob/main/controllers/repo_manager/web.go#L379

[mikedep333]

FYI: We moved the issue here because it lies in the pulp-oci-images scripts rather than in pulp-operator.

[plnordquist]

This is probably the best place to make the change although there may be follow on changes required in the operator since the OpenTelemetry feature overrides the default entrypoint to instrument the execution of pulp-api.

I use the ingress-nginx ingress controller at https://github.com/kubernetes/ingress-nginx so I set the is_nginx_ingress option on my Pulp CR and that avoids deploying pulp-web pods. Maybe there needs to be a separate issue for the pulp-web pods in the operator repository.