DRBD9 and LINSTOR the easy way

tested on focal

Install

on all nodes

the v9 module won’t always build, but it does with a kernel.org longterm, or ubuntu’s equivalent

uname -r
dpkg -l | grep headers
ls -lF /lib/modules/*/build
ls -lF /lib/modules/*/source

#apt install linux-generic

#cd /usr/src/linux/
#make olddefconfig && echo OK
#make prepare && echo OK

apt install software-properties-common apt-transport-https ca-certificates

#add-apt-repository ppa:npalix/coccinelle
#apt install coccinelle

add-apt-repository ppa:linbit/linbit-drbd9-stack
apt install drbd-dkms

--> postfix: local only meanwhile you setup DMA
--> postfix: FQDN as found from hosts // otherwise force it by adding the domain
    ...and fix your freaking /etc/hosts by adding the domain even to the cluster entries...

dkms status
ls -lF /lib/modules/*/updates/dkms/
dpkg -l | grep drbd #drbd-utils is there also

apt install linstor-controller linstor-satellite linstor-client python-linstor
dpkg -l | grep linstor

Cluster setup

on all nodes

systemctl enable --now linstor-controller
systemctl status linstor-controller
linstor node list

#LS_CONTROLLERS=lin1,lin2,lin3 linstor node list
vi /etc/linstor/linstor-client.conf

[global]
controllers=pro5s1,pro5s2

Now it’s time to check whether you are either LVM2/thin or ZFS ready

dpkg -l | grep lvm2
lsmod | grep zfs

on a single node

cat /etc/hosts
linstor node create pro5s1 x.x.x.x
linstor node create pro5s2 x.x.x.x

Storage pools

LVM2

all nodes

    pvcreate /dev/xvdb
    vgcreate vdisks /dev/xvdb

pvs
vgs

controller only

linstor storage-pool create lvm lin1 vpool vdisks
linstor storage-pool create lvm lin2 vpool vdisks
linstor storage-pool create lvm lin3 vpool vdisks
#linstor storage-pool delete lin3 vpool

linstor storage-pool list

and check on every node

lvs

or at once

#linstor physical-storage list
#linstor physical-storage create-device-pool \
#   --pool-name vdisks LVMTHIN lin1 /dev/xvdb \
#   --storage-pool vpool

ZFS

all nodes

Assuming three disks. Making use of what is left on first disk

cfdisk /dev/sda
NEW / ALL REMAINING SPACE / be Solaris boot

and creating the ZFS pool for every node

zpool create tank1 /dev/sda3 /dev/sdb /dev/sdc
zpool create tank2 /dev/sda3 /dev/sdb /dev/sdc
    zpool list
zfs list

controller only

create a pool WITH THE SAME NAME across the farm

    linstor storage-pool create zfs pro5s1 vdiskpool tank1
    linstor storage-pool create zfs pro5s2 vdiskpool tank2

    linstor storage-pool list
linstor storage-pool list-properties pro5s1 vdiskpool
linstor storage-pool list-properties pro5s2 vdiskpool

    #linstor storage-pool delete NODE-NAME POOL-NAME

Storage resources

linstor resource-group create vdiskrg --storage-pool vdiskpool --place-count 2
linstor resource-group list

#linstor resource-group delete RES-GROUP

now create the VG

linstor volume-group create vdiskrg
linstor volume-group list vdiskrg
linstor volume-definition list

#linstor volume-group delete RG-NAME 0

and some resources within, the quick way

linstor resource-group spawn-resources vdiskrg vdisk0 10G

or the right way

linstor resource-definition create vdisk0
linstor volume-definition create vdisk0 10G
#linstor volume-definition set-size backups 0 100G

linstor resource create pro5s1 vdisk0 --storage-pool vdiskpool

and check

linstor resource-definition list
linstor resource list

linstor resource list-properties pro5s1 GUEST-NAME

#linstor resource-definition delete vres0
#linstor resource delete pro5s1 vres0

TODO

to be continued…

make the PoC evolve to it actually uses the LUNs
see how serve the LUNs over the network (NVME-oF, diskless, iSCSI)
finally check how good it is for convergence

Resources