RPM and yum decided to install a 32 bit kernel on a 64 bit system. I don’t know why, because they didn’t do so on an identically configured server that was its clustermate. This rendered the system unbootable.
Once I’d gotten someone to power cycle it, I went in and removed the 32 bit kernels. This made RPM remove all the initrd images that had been set up for the 32 bit kernel. Unfortunately, it didn’t bother to put back the initrd images for the 64 bit kernels that were still there. This made the system even more unbootable.
Maybe I’m crazy, but it seems to me that the default for a 64 bit install should be not to install 32 bit kernels unless you go out of your way to ask for them. And it seems to me that any changes to the set of kernels installed should automatically result in mkinitrd being run to generate any necessary initrd images for the installed kernels.
Oh, and while we’re at it, maybe yum reinstall should actually work on the kernel. And maybe RPM’s log files should actually record when each package was installed, rather than just a list of the package names. Or it could even, you know, use syslog.
(Note: You can get the info from rpm -qa --last | tac — if you’re chrooted on your broken system and RPM is working and RPM hasn’t recently corrupted its database.)