During a recent PoC I was faced with an unusual scenario where, despite meeting all the documented prerequisites, the installation kept looping through the PXE boot process and the installation eventually timed out. Monitoring the console of the first node, the undercloud, during the boot process revealed very little – other than the verification that PXE was responding and an IP address was successfully being allocated through DHCP.

The one thing that was different this time to other installations was that this time I was using 3PAR storage to boot the physical nodes from SAN.
My first thought was that maybe the 2 TiB drives that were assigned may be too large. These drives were quickly modified and re-presented as 1.9TiB but this had no impact on the installation process, still timing out.

Next, I booted one of the servers using a CentOS image to ensure that the SAN drives were ok and what was being presented by 3PAR was the correct size, this also revealed no surprises – everything was as expected.

Finally, one of my colleagues suggested that I modify the baremetal.csv file and put a smaller value than what’s being presented for the server storage parameters. Within minutes after changing 1900 GiB to 1850 GiB all was well – the installation completed successfully!

Despite the documentation stating that everything must be in GiB, not exceed 2 TiB and the size of the local storage should match the storage parameter in the baremetal.csv – Don’t be afraid to understate the disk size in the baremetal.csv file.