Installing CentOS on D03s/D05s is simple and requires three steps using mr-provisioner.
Flashing a new firmware
To flash the new firmware, follow these steps:
- First, boot the machine and wait for the UEFI boot prompt and select the EBL option, as described here.
- Then enter the FTP command line as described here.
To know what firmware you need for which ERP/CentOS version, contact the SDI/Server groups at LEG.
Here's a list of the know firmware/ERP pairs available in the MLab:
| ERP CentOS Version | Firmware Build |
|---|---|
| 17.08 (build 95) | 153 |
To upload your own firmware to the FTP, contact the SDI group with FTP instructions. Remember to chmod new dir/files, or the update process won't be able to see it.
Choosing the right PXE options
Using mr-provisioner, edit the machine you want and add these options:
- Kernel/initrd: Make sure they're the same and compatible
- Kernel opts: You have to set two important flags:
- init.stage2: set this to the URL of where you downloaded the kernel/initrd
- noverifyssl: necessary on first install, as UEFI's date is not set on first boot, and CentOS forces HTTPS downloads (and certificate is invalid). Remaining installs don't need this.
- inst.cmdline: if you're using a kickstart script, this makes sure there are no interruptions
- acpi=on: this may not be necessary, but CentOS only works with ACPI, so doesn't hurt
- A good Anaconda KickStart script (the one attached here is a good start)
- Netboot enabled: Yes
Example:
Download kernel/initrd from http://builds.96boards.org/snapshots/reference-platform/components/centos-installer-staging/94/images/pxeboot/
Set init.stage2 to http://builds.96boards.org/snapshots/reference-platform/components/centos-installer-staging/94/
This will download the squashfs.img from: http://builds.96boards.org/snapshots/reference-platform/components/centos-installer-staging/94/LiveOS/
Press the [PXE Boot] button and follow the console without interruptions.
To learn about all kernel arguments for the PXE installer, check this page.
To learn about all Anaconda's Kick Start script options, read this.
Running the installer
After boot, it can take up to 15 minutes to get to the PXE boot Grub. Go make some tea.
As soon as Grub downloads the kernel/initrd, that'll setup the network and it'll download squashfs's stage 2 image. Once that boots, you'll see a screen with a message:
Starting installer, one moment...
That will hang in there for a while if you have a broken IPMI image.
Working around the IPMI bug
There is a bug in the IPMI tool for now, so you'll need to kill it before the installer can execute. It'll hang saying waiting for the installer. This is due to the IPMI tool being locked.
Until we find the cause of the hang, you can work around by killing the process.
The installer is running inside tmux, so you'll need to go to the shell pane (ctrl+b+n), then pgrep ipmi, then kill the PID that it prints.
This will free the installer to proceed. To go back to the installer pane, just use ctrl+b+p.
Installer options
If you haven't selected inst.cmdline, the installer will ask you what type you want.
If you want to use VNC, you'll have to attach it somehow. If you just want a minimal install, use the Kick Start script attached here and select "text mode".
If you have added the right KS file in the machine setup and have chosen inst.cmdline, the installation will proceed automatically.
After the installation process starts, it can take up to 15 minutes more to finish. Go walk the dog.
Working around the IPMI bug (again)
Once it gets to "Running post-install scripts", the ipmi tool will hang again.
Same thing as last time, (ctrl+b+n), pgrep ipmi, kill the PID.
Restarting
Finally, the installation will say it finished. If all went well, you should be able to reboot (no PXE) and wait long enough for a CentOS prompt.
Change Passwords
If you followed the kick start file above, the root password is "centos" and the user "centos" has password "centos" and has sudo.
Change the passwords immediately!