dledford
commented
2 years ago There are two separate issues here. I'm going to open a new issue for one and we can handle the other in this thread. The first issue is that the rdma_setup.sh scripts rely on the system using the old SysV init network files and Fedora rawhide (and maybe Fedora 36 too) have switched to NetworkManager connections without the old SysV init network files (also known as rhconfig network files). That is now being handled in issue #104 .
The second problem, and it's not related to the first but instead appears to be specific to the debug options used in Rawhide kernels, is the kernel panic. To fix that, I think we need to find out what exactly is causing the kernel panic and then use a command line option to disable it. It appears to be coming from the audit subsystem, which would mean SELinux. However, it could also be any number of other things especially since there appears to be a number of debugging options turned on in the Rawhide kernel and those can sometimes cause more problems than they solve. But, without getting a system up and running to see what the kernel's debug options are, it can be a guessing game to try and figure out which options to disable (and whether they even can be disabled from the command line or if they require a different kernel to be built).
JSpewock
In recent CKI testing it has been found that on some occasions the
rdma_setup.shscript fails producing many errors that resemble/etc/sysconfig/network-scripts/*: No such file or directory. After this setup script fails, the system restarts and ends in a kernel panic. An example of a job where this setup script fails can be found here and the console logs containing the errors can be found here.