#!/bin/sh
# 17:19, 2018-02-09
# Nico Schottelius

# Based on ceph-disk -v prepare --bluestore /dev/sdc --osd-id ${ID} --osd-uuid $(uuidgen) --crush-device-class "ssd"

# Create:
# - block       -> link to partuuid
# - block_uuid -e> uuid if the block
# - ceph_fsid   -> get from ceph-conf
# crush_device_class -> ssd, hdd
# fsid          -> uuidgen!
# magic         -> string "ceph osd volume v026"
# type          -> bluestore

fsid=$(ceph-conf --cluster=ceph --name=osd. --lookup fsid)
fs_uuid=$(uuidgen)
magic="ceph osd volume v026"

set -x
set -e

if [ $# -lt 2 ]; then
    echo "$0 disk class [osdweight]"
    echo "class = hdd or ssd"
    exit 1
fi

export DEV=$1;shift
export CLASS=$1; shift


uuid_metadata=$(uuidgen)
uuid_block=$(uuidgen)

osd_id=$(ceph osd create)

dev_metadata="/dev/disk/by-partuuid/$uuid_metadata"
dev_block="/dev/disk/by-partuuid/$uuid_block"

/sbin/sgdisk --new=0:0:+100M --change-name="0:ceph data" \
             --partition-guid="0:$uuid_metadata" \
             --typecode=0:4fbd7e29-9d25-41b8-afd0-062c0ceff05d \
             --mbrtogpt -- $DEV
/sbin/udevadm settle --timeout=600

# Using gdisk --largest-new does not change the name or set guid;
# So use 2 steps instead
/sbin/sgdisk --largest-new=0 --mbrtogpt -- $DEV
/sbin/udevadm settle --timeout=600


lastpart=$(gdisk -l $DEV | tail -n1 | awk '{ print $1 }')
/sbin/sgdisk --change-name="${lastpart}:ceph block" \
             --partition-guid="${lastpart}:$uuid_block" \
             --typecode="${lastpart}:cafecafe-9b03-4f30-b4c6-b4b80ceff106" \
             --mbrtogpt -- $DEV
/sbin/udevadm settle --timeout=600

/sbin/mkfs -t xfs -f -i size=2048 -- "$dev_metadata"

mountpath=/var/lib/ceph/osd/ceph-${osd_id}

mkdir -p "$mountpath"
mount "$dev_metadata" "$mountpath"

ln -s $dev_block        "$mountpath/block"

echo "$uuid_block"    > "$mountpath/block_uuid"
echo "$fsid"          > "$mountpath/ceph_fsid"
echo "$magic"         > "$mountpath/magic"

# Not needed (anymore) - using ceph osd crush set-device-class below
#echo "$CLASS"         > "$mountpath/crush_device_class"

# Important, otherwise --mkfs later will try to create filestore
echo bluestore        > "$mountpath/type"

ceph auth get-or-create "osd.${osd_id}" osd \
     'allow *' mon 'allow profile osd' > $mountpath/keyring

echo ${osd_id}         > "$mountpath/whoami"
touch                   "$mountpath/sysvinit"

ceph-osd --cluster ceph -i "${osd_id}" --mkfs
chown -R ceph:ceph      "$mountpath"

# Also allow access to the blockdevices - via symlink
chown ceph:ceph "${mountpath}/block"

if [ $# -eq 1 ]; then
    WEIGHT=$1; shift
else
    devname=$(readlink -f $dev_block)
    nodev=$(echo $devname | sed 's,/dev/,,')
    WEIGHT=$(lsblk -l -b | awk "/^$nodev/ { print \$4/(1024^4) }")
fi

ceph osd crush add osd.${osd_id} ${WEIGHT} host=$(hostname)

# Ensure previous assigned class is gone
ceph osd crush rm-device-class osd.${osd_id}
ceph osd crush set-device-class $CLASS osd.${osd_id}

echo "$metadata_dev /var/lib/ceph/osd/ceph-${osd_id} xfs noatime 0 0" >> /etc/fstab

# Starting with monit, if available
if [ -e /etc/monit ]; then
    /opt/ungleich-tools/monit-ceph-create-start osd.${osd_id}
else
    /etc/init.d/ceph start osd.${osd_id}
fi

#rm "$mountpath/crush_device_class"