XOSTOR hyperconvergence preview

olivierlambert

peter_webbird

Is CBT meant to be supported on XOSTOR?

I've been experimenting with XOSTOR recently, but upon testing a delta-backup, noticed this warning...
couldn't enable CBT

The error message behind this is SR_OPERATION_NOT_SUPPORTED when calling Async.VDI.enable_cbt.

Running xe sr-param-list uuid={uuid} shows the following:

[~]# xe sr-param-list uuid={...}
    uuid ( RO) : {...}
    name-label ( RW): CD6
    name-description ( RW): Array of Kioxia CD6 U.2 drives, one in each Host.
    host ( RO): <shared>
    allowed-operations (SRO): unplug; plug; PBD.create; update; PBD.destroy; VDI.resize; VDI.clone; scan; VDI.snapshot; VDI.mirror; VDI.create; VDI.destroy
    {...etc}
    type ( RO): linstor
    content-type ( RO): user
    shared ( RW): true
    introduced-by ( RO): <not in database>
    is-tools-sr ( RO): false
    other-config (MRW): auto-scan: true
    sm-config (MRO):
    {...etc}

Compared to another SR, the following allowed-operations are missing:

VDI.enable_cbt;
VDI.list_changed_blocks;
VDI.disable_cbt;
VDI.data_destroy;
VDI.set_on_boot

Is this the expected behaviour? Note that this is using XCP-ng 8.2 (I've yet to test out 8.3).

olivierlambert

Ping @ronan-a or @dthenot

ronan-a

@peter_webbird We've already had feedback on CBT and LINSTOR/DRBD, we don't necessarily recommend enabling it. We have a blocking dev card regarding a bug with LVM lvchange command that may fail on CBT volumes used by a XOSTOR SR. We also have other issues related to migration with CBT.

gb.123

@ronan-a @dthenot @Team-Storage

Guys, Can you please clarify which method to use for installing XOSTOR in XCP-ng 8.3 ?

Simple :

yum install xcp-ng-linstor
yum install xcp-ng-release-linstor
./install --disks /dev/nvme0n1 --thin

Or the script in the first post ?
Or Some other script ?

dthenot

@gb.123 Hello,
The instruction in the first post are still the way to go

JeffBerntsen

@dthenot said in XOSTOR hyperconvergence preview:

@gb.123 Hello,
The instruction in the first post are still the way to go

I'm curious about that as well but the first post says that the installation script is only compatible with 8.2 and doesn't mention 8.3. Is that still the case or is the installation script now compatible with 8.3 as well? If not, is there an installation script that is compatible with 8.3?

I know that using XO is the recommended method for installation but I'm interested in an installation script as I would like to try to integrate XOSTOR installation into an XCP-ng installation script I already have which runs via PXE boot.

dthenot

@JeffBerntsen That's why I meant, the way to install written in the first post still work in 8.3, the script still work as expected also, it basically only create the VG/LV needed on hosts before you create the SR.

JeffBerntsen

@dthenot said in XOSTOR hyperconvergence preview:

@JeffBerntsen That's why I meant, the way to install written in the first post still work in 8.3, the script still work as expected also, it basically only create the VG/LV needed on hosts before you create the SR.

Got it. Thanks!

henri9813

Hello,

I plan to install my XOSTOR cluster on a pool of 7 nodes with 3 replicas, but not all nodes at once because disks are in use.
consider:

node1
node2
node ...
node 5
node 6
node 7.

with 2 disks on each

sda: 128GB for the OS
sdb: 1TB for local sr ( for now )

I emptied node 6 & 7.

so, here is what i plan to do:

On ALL NODES: setup linstor packages

Run the install script on node 6 & 7 to add their disks
so:

node6# install.sh --disks /dev/sdb
node7# install.sh --disks /dev/sdb

Then, configure the SR and the linstor plugin manager as the following

xe sr-create \ 
    type=linstor name-label=pool-01 \
    host-uuid=XXXX \
    device-config:group-name=linstor_group/thin_device device-config:redundancy=3 shared=true device-config:provisioning=thin

Normally, i should have a linstor cluster running of 2 nodes ( 2 satellite and one controller randomly placed ) with only 2 disks and then, only 2/3 working replicas.

The cluster SHOULD be usable ( i'm right on this point ? )

The next step, would be to move VM from node 5 on it to evacuate node 5. and then add it to the cluster by the following

node5# install.sh --disks /dev/sdb
node5# xe host-call-plugin \
  host-uuid=node5-uuid \
  plugin=linstor-manager \
  fn=addHost args:groupName=linstor_group/thin_device

That should deploy satelite on node 5 and add the disk.

I normally should have 3/3 working replicas and can start to deploy others nodes progressively.

I'm right on the process ?

aS mentionned in the discord, i will post my feedbacks and results from my setup once i finalized it. ( maybe thought a blog post somewhere ).

Thanks to provide xostor in opensource, it's clearly the missing piece for this virtualization stack in opensource ( vs proxmox )