starlingx/metal
Code Issues Proposed changes

Merge "Fix wipedisk to not break Ceph OSDs during B&R"

Browse Source
This commit is contained in:
Zuul 2020-09-24 17:05:54 +00:00 committed by Gerrit Code Review
commit f5725ad694
2 changed files with 19 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
bsp-files/kickstarts/pre_disk_setup_common.cfg
View File

@ -148,6 +148,7 @@ else
WIPE_HDD=""
# Partition type OSD has a unique globally identifier
CEPH_OSD_GUID="4FBD7E29-9D25-41B8-AFD0-062C0CEFF05D"
CEPH_JOURNAL_GUID="45B0969E-9B03-4F30-B4C6-B4B80CEFF106"
# Check if we wipe OSDs
if [ "$(curl -sf http://pxecontroller:6385/v1/ihosts/wipe_osds 2>/dev/null)" = "true" ]; then
@ -227,9 +228,9 @@ do
part=${dev}p${part_number}
;;
esac
sgdisk_part_info=$(sgdisk -i $part_number $dev)
part_type_guid=$(echo "$sgdisk_part_info" | grep "$part_type_guid_str" | awk '{print $4;}')
if [ "$dev" == "$rootfs_device" ]; then
sgdisk_part_info=$(sgdisk -i $part_number $dev)
part_type_guid=$(echo "$sgdisk_part_info" | grep "$part_type_guid_str" | awk '{print $4;}')
part_fstype=$(exec_retry 5 0.5 "blkid -s TYPE -o value $part")
if [ "$part_type_guid" == $BACKUP_PART_GUID -a "${part_fstype}" == "ext4" ]; then
wlog "Skipping wipe backup partition $part"
@ -238,8 +239,17 @@ do
fi
fi
wlog "Wiping partition $part"
dd if=/dev/zero of=$part bs=512 count=34
dd if=/dev/zero of=$part bs=512 count=34 seek=$((`blockdev --getsz $part` - 34))
if [ $WIPE_CEPH_OSDS == "true" -a "$part_type_guid" == $CEPH_JOURNAL_GUID ]; then
# Journal partitions require additional wiping. Based on the ceph-manage-journal.py
# script in the integ repo (at the ceph/ceph/files/ceph-manage-journal.py location)
# wiping 100MB of data at the beginning of the partition should be enough. We also
# wipe 100MB at the end, just to be safe.
dd if=/dev/zero of=$part bs=1M count=100
dd if=/dev/zero of=$part bs=1M count=100 seek=$(( `blockdev --getsz $part` / (1024 * 2) - 100 ))
else
dd if=/dev/zero of=$part bs=512 count=34
dd if=/dev/zero of=$part bs=512 count=34 seek=$((`blockdev --getsz $part` - 34))
fi
exec_retry 5 0.5 "parted -s $dev rm $part_number"
# LP 1876374: On some nvme systems udev doesn't correctly remove the
# links to the deleted partitions from /dev/nvme* causing them to be

36
mtce/src/scripts/wipedisk
View File

@ -70,22 +70,6 @@ done
WIPE_HDD="$pvs_to_delete $WIPE_HDD"
# During host reinstalls ceph journals also require wiping, so we also gather information on
# journal partitions. Even if this script is also called during upgrades, there was no issue
# observed during that operation, so we skip wiping the journals during upgrades.
JOURNAL_DISKS=""
HOST_IN_UPGRADE=$(curl -sf http://controller:6385/v1/upgrade/$(hostname)/in_upgrade 2>/dev/null)
# The "ceph-disk list" command works even if the ceph cluster is not operational (for example if
# too many monitors are down) so we can grab journal info from the node, even in such scenarios.
# As a safety measure, we also wrap the command in a timeout command; it should never take long
# for the command to return, but if it does it's safer to just time it out after 15 seconds.
CEPH_DISK_OUTPUT=$(timeout 15 ceph-disk list 2>/dev/null)
if [[ $? == 0 && "$HOST_IN_UPGRADE" != "true" ]]; then
JOURNAL_DISKS=$(echo "$CEPH_DISK_OUTPUT" | grep "ceph journal" | awk '{print $1}')
fi
WIPE_HDD="$JOURNAL_DISKS $WIPE_HDD"
if [ ! $FORCE ]
then
echo "This will result in the loss of all data on the hard drives and"
@ -158,21 +142,11 @@ do
echo "Wiping $dev..."
wipefs -f -a $dev
echo "$JOURNAL_DISKS" | grep -qw "$dev"
if [[ $? == 0 ]]; then
# Journal partitions require additional wiping. Based on the ceph-manage-journal.py
# script in the integ repo (at the ceph/ceph/files/ceph-manage-journal.py location)
# wiping 100MB of data at the beginning of the partition should be enough. We also
# wipe 100MB at the end, just to be safe.
dd if=/dev/zero of=$dev bs=1M count=100
dd if=/dev/zero of=$dev bs=1M count=100 seek=$((`blockdev --getsz $dev` - 204800))
else
# Clearing previous GPT tables or LVM data
# Delete the first few bytes at the start and end of the partition. This is required with
# GPT partitions, they save partition info at the start and the end of the block.
dd if=/dev/zero of=$dev bs=512 count=34
dd if=/dev/zero of=$dev bs=512 count=34 seek=$((`blockdev --getsz $dev` - 34))
fi
# Clearing previous GPT tables or LVM data
# Delete the first few bytes at the start and end of the partition. This is required with
# GPT partitions, they save partition info at the start and the end of the block.
dd if=/dev/zero of=$dev bs=512 count=34
dd if=/dev/zero of=$dev bs=512 count=34 seek=$((`blockdev --getsz $dev` - 34))
fi
fi
done