Solved: Kickstart clearpart Not Working on RHEL

Kickstart clearpart should erase partition tables before Anaconda lays down a new layout. On reused lab VMs and rack servers with old LVM, RAID, or leftover signatures, Anaconda sometimes stops with:

• not enough free space on the install disk
• clearpart did not remove partitions even when clearpart --all is in the file

This article documents the failure we hit on RHEL 10 vm2 during PXE kickstart testing and the fix: scope disks with ignoredisk, tighten clearpart --drives=, and run a %pre script that deactivates volume groups and wipefs the install disk before partitioning.

Problem: clearpart not working

In a kickstart file you use clearpart to erase partitions before autopart or manual part lines. A minimal pattern looks like this:

zerombr
clearpart --all --initlabel
autopart --type=lvm

On vm2 the install failed during storage setup. Anaconda reported insufficient free space on sda because an old LVM stack from a prior install was still present. clearpart --all without --drives= also risked touching every disk. vm2 has a second disk (sdb) that the harness must keep for ignoredisk demos.

Solution: scope disks and wipe in %pre

Two kickstart changes plus a %pre script fixed the lab run.

Step 1 — ignoredisk and scoped clearpart

Keep sdb untouched and limit clearing to sda:

ignoredisk --drives=sdb
zerombr
clearpart --all --initlabel --drives=sda
autopart --type=lvm --nohome

ignoredisk tells Anaconda to ignore sdb for partitioning. clearpart --drives=sda avoids wiping every disk when --all is set.

Step 2 — %pre script to break old LVM and wipe signatures

When clearpart alone is not enough, add a %pre block that runs before partitioning. This is the script from the lab harness on vm1 (/var/www/html/ks/rhel10-lab.cfg):

%pre --log=/tmp/kickstart_pre.log
#!/bin/bash
set -x
echo "=== %pre wipe install disk $(date -Is) ==="
INSTALL_DISK=sda
udevadm settle
vgchange -an 2>/dev/null || true
dmsetup remove_all 2>/dev/null || true
if [[ -b /dev/${INSTALL_DISK} ]]; then
  wipefs -af "/dev/${INSTALL_DISK}" || true
  for part in /dev/${INSTALL_DISK}*; do
    [[ -b "$part" ]] && wipefs -af "$part" || true
  done
  lsblk "/dev/${INSTALL_DISK}"
fi
%end

The script deactivates volume groups, removes device-mapper nodes, and wipes partition and filesystem signatures on one disk only. That matches ignoredisk and clearpart --drives=sda.

For RAID-heavy hosts, extend the pattern with mdadm --stop and per-RAID wipefs before the disk loop. See the extended example in Kickstart %pre script examples.

Step 3 — validate before netboot

Run ksvalidator on the full kickstart file on the server:

ksvalidator /var/www/html/ks/rhel10-lab.cfg

Checking kickstart file /var/www/html/ks/rhel10-lab.cfg

A silent exit after that line means the parser accepted the storage and script sections.

If storage still fails, inspect installer logs on the target during the failed run:

/tmp/storage.log
/tmp/anaconda.log

After a successful harness install on vm2, confirm scoped clearing and ignoredisk on the running system:

lsblk -o NAME,SIZE,TYPE,FSTYPE,MOUNTPOINT

sda                   20G disk
├─sda2                 2G part xfs         /boot
└─sda3                18G part LVM2_member
  ├─rhel_vm200-root   16G lvm  xfs         /
  └─rhel_vm200-swap    2G lvm  swap        [SWAP]
sdb                   10G disk
└─sdb1                10G part LVM2_member

sda carries the new install. sdb was not partitioned by Anaconda, but a leftover LVM partition from earlier lab runs can remain on the ignored disk. That is expected with ignoredisk --drives=sdb; wipe sdb manually in %pre only when you need a completely blank second disk for other demos.

When to rely on clearpart alone

clearpart without %pre is enough on blank disks or after a single prior kickstart that did not leave active LVM. Add the wipe %pre when you:

• Reinstall the same VM repeatedly in a lab
• See "not enough free space" with existing dm-* or rl_* volume names in lsblk
• Run kickstart on servers that previously used software RAID or complex LVM

References

• Red Hat Enterprise Linux 10 — clearpart kickstart command

Summary

When kickstart clearpart does not remove old partitions, combine ignoredisk, scoped clearpart --drives=, and a %pre script that deactivates VGs and runs wipefs on the install disk only.

The lab harness on vm2 uses sda for install and ignores sdb for partitioning. After a successful run, lsblk should show LVM on sda (rhel_vm200-* when the hostname is vm2.lab.example). sdb may still show an old LVM2_member partition if prior installs used it; ignoredisk prevents Anaconda from clearing or reusing that disk. Adapt disk names for your hardware before production netboots, and never run a blanket wipe loop across every /dev/sd* on a multi-disk server.