Check DNS in postinst scripts in package and image mode

[mh21]

When provisioning current c10s in image mode in Beaker, the postinst script is unable to resolve DNS names.

This works fine for FC41, or for c10s package mode.

Please consider adding tests that check for DNS resolution in postinst.

Reproducer: beaker.zip

[cgwalters]

When I last looked at this what I suspect is going on is that Anaconda is setting up /sysroot/etc/resolv.conf, but for ostree/bootc we actually use the deployment root.

https://github.com/rhinstaller/anaconda/blob/91fe91ad6f03903114e6637d792b8d7caa679103/pyanaconda/installation.py#L426

[cgwalters]

Filed https://issues.redhat.com/browse/RHEL-73499 to track this

[jkonecny12]

I'm afraid this will not be the only case with such a bug.

Before the image mode we more or less knew how the system will look like after the installation because it was a fresh installation. However, with bootc it is no longer true because of any customization done by the user. Thanks for filing the issue.

[cgwalters]

Before the image mode we more or less knew how the system will look like after the installation because it was a fresh installation. However, with bootc it is no longer true because of any customization done by the user.

Can you elaborate a bit more? In theory admins can make (nearly) equally arbitrary changes in the set of RPMs input. Although now it is definitely possible to do things like not having rpm in the target system at all, etc.

[jkonecny12]

Yes, they can do that now with:

• kickstart file
• Image Builder (tar payload)
• bootc

I guess, the change is that bootc is much more suited for these changes than the other methods? So, it's not something new, it's more like blind spot on our side because these changes were not that easy to achieve before IMHO.

[cgwalters]

Yes, actually on this topic - certainly someone could put /etc/resolv.conf in an RPM, but it'd likely feel a bit weird to do. But not only is it way easier and more obvious to do that with bootc, it actually always happens today with podman build because of containers/buildah#4242 (comment) even if you don't want it...

[rvykydal]

When I last looked at this what I suspect is going on is that Anaconda is setting up /sysroot/etc/resolv.conf, but for ostree/bootc we actually use the deployment root.

https://github.com/rhinstaller/anaconda/blob/91fe91ad6f03903114e6637d792b8d7caa679103/pyanaconda/installation.py#L426

Seems it can be the cause of the problem: https://issues.redhat.com/browse/RHEL-73499?focusedId=26380845&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-26380845