Strahil,
Thanks for all the follow up, I will try to reproduce the same scenario today, deploy a 9 node cluster, Completely kill the initiating node (vmm10) and see If i can recover using the extra server approach (Different IP/FQDN). If I am able to recover I will also try to test with your suggested second approach (Using same IP/FQDN).
My objective here is to document the possible recovery scenarios without any downtime or impact.

I have documented a few setup and recovery scenarios with 6 and 9 nodes already with a hyperconverged setup and I will make them available to the community, hopefully this week, including the tests that you have been helping me with. Hopefully this will provide help to others that are in the same situation that I am, and it will also provide me with feedback from more knowledgeable admins out there so that I can get this into production in the near future.


Thanks again.



On Wed, Jun 12, 2019 at 11:58 PM Strahil <hunter86_bg@yahoo.com> wrote:

Hi Adrian,

Please keep in mind that when a server dies, the easiest way to recover is to get another freshly installed server with different IP/FQDN .
Then you will need to use 'replace-brick' and once gluster replaces that node - you should be able to remove the old entry in oVirt.
Once the old entry is gone, you can add the new installation in oVirt via the UI.

Another approach is to have the same IP/FQDN for the fresh install.In this situation, you need to have the same gluster ID (which should be a text file) and the peer IDs. Most probably you can create them on your own , based on data on the other gluster peers.
Once the fresh install is available in 'gluster peer' , you can initiate a reset-brick' (don't forget to set the SELINUX , firewall and repos) and a full heal.
From there you can reinstall the machine from the UI and it should be available for usage.

P.S.: I know that the whole procedure is not so easy :)

Best Regards,
Strahil Nikolov

On Jun 12, 2019 19:02, Adrian Quintero <adrianquintero@gmail.com> wrote:
Strahil, I dont use the GUI that much, in this case I need to understand how all is tied together if I want to move to production. As far as Gluster goes, I can get do the administration thru CLI, however when my test environment was set up it was setup using geodeploy for Hyperconverged setup under oVirt.
The initial setup was 3 servers with the same amount of physical disks: sdb, sdc, sdc, sdd, sde(this last one used for caching as it is an SSD)

vmm10.mydomain.com:/gluster_bricks/brick1(/dev/sdb) engine
vmm10.mydomain.com:/gluster_bricks/brick2(/dev/sdb) vmstore1
vmm10.mydomain.com:/gluster_bricks/brick3(/dev/sdc) data1
vmm10.mydomain.com:/gluster_bricks/brick4(/dev/sdd) data2

vmm11.mydomain.com:/gluster_bricks/brick1(/dev/sdb) engine
vmm11.mydomain.com:/gluster_bricks/brick2(/dev/sdb) vmstore1
vmm11.mydomain.com:/gluster_bricks/brick3(/dev/sdc) data1
vmm11.mydomain.com:/gluster_bricks/brick4(/dev/sdd) data2

vmm12.mydomain.com:/gluster_bricks/brick1(/dev/sdb) engine
vmm12.mydomain.com:/gluster_bricks/brick2(/dev/sdb) vmstore1
vmm12.mydomain.com:/gluster_bricks/brick3(/dev/sdc) data1
vmm12.mydomain.com:/gluster_bricks/brick4(/dev/sdd) data2

As you can see from the above the the engine volume is conformed of hosts vmm10 (Initiating cluster server but now dead sever), vmm11 and vmm12 and on block device /dev/sdb (100Gb LV), also the vmstore1 volume is also on /dev/sdb (2600Gb LV).
/dev/mapper/gluster_vg_sdb-gluster_lv_engine                   xfs             100G  2.0G   98G   2% /gluster_bricks/engine
/dev/mapper/gluster_vg_sdb-gluster_lv_vmstore1                 xfs             2.6T   35M  2.6T   1% /gluster_bricks/vmstore1
/dev/mapper/gluster_vg_sdc-gluster_lv_data1                    xfs             2.7T  4.6G  2.7T   1% /gluster_bricks/data1
/dev/mapper/gluster_vg_sdd-gluster_lv_data2                    xfs             2.7T  9.5G  2.7T   1% /gluster_bricks/data2
vmm10.mydomain.com:/engine                                       fuse.glusterfs  300G  9.2G  291G   4% /rhev/data-center/mnt/glusterSD/vmm10.virt.iad3p:_engine
vmm10.mydomain.com:/vmstore1                                     fuse.glusterfs  5.1T   53G  5.1T   2% /rhev/data-center/mnt/glusterSD/vmm10.virt.iad3p:_vmstore1
vmm10.mydomain.com:/data1                                        fuse.glusterfs  8.0T   95G  7.9T   2% /rhev/data-center/mnt/glusterSD/vmm10.virt.iad3p:_data1
vmm10.mydomain.com:/data2                                        fuse.glusterfs  8.0T  112G  7.8T   2% /rhev/data-center/mnt/glusterSD/vmm10.virt.iad3p:_data2



before any issues I increased the size of the cluster and the gluster cluster with the following, creating 4 distributed replicated volumes (engine, vmstore1, data1, data2)

vmm13.mydomain.com:/gluster_bricks/brick1(/dev/sdb) engine
vmm13.mydomain.com:/gluster_bricks/brick2(/dev/sdb) vmstore1
vmm13.mydomain.com:/gluster_bricks/brick3(/dev/sdc) data1
vmm13.mydomain.com:/gluster_bricks/brick4(/dev/sdd) data2

vmm14.mydomain.com:/gluster_bricks/brick1(/dev/sdb) engine
vmm14.mydomain.com:/gluster_bricks/brick2(/dev/sdb) vmstore1
vmm14.mydomain.com:/gluster_bricks/brick3(/dev/sdc) data1
vmm14.mydomain.com:/gluster_bricks/brick4(/dev/sdd) data2

vmm15.mydomain.com:/gluster_bricks/brick1(/dev/sdb) engine
vmm15.mydomain.com:/gluster_bricks/brick2(/dev/sdb) vmstore1
vmm15.mydomain.com:/gluster_bricks/brick3(/dev/sdc) data1
vmm15.mydomain.com:/gluster_bricks/brick4(/dev/sdd) data2

vmm16.mydomain.com:/gluster_bricks/brick1(/dev/sdb) engine
vmm16.mydomain.com:/gluster_bricks/brick2(/dev/sdb) vmstore1
vmm16.mydomain.com:/gluster_bricks/brick3(/dev/sdc) data1
vmm16.mydomain.com:/gluster_bricks/brick4(/dev/sdd) data2

vmm17.mydomain.com:/gluster_bricks/brick1(/dev/sdb) engine
vmm17.mydomain.com:/gluster_bricks/brick2(/dev/sdb) vmstore1
vmm17.mydomain.com:/gluster_bricks/brick3(/dev/sdc) data1
vmm17.mydomain.com:/gluster_bricks/brick4(/dev/sdd) data2

vmm18.mydomain.com:/gluster_bricks/brick1(/dev/sdb) engine
vmm18.mydomain.com:/gluster_bricks/brick2(/dev/sdb) vmstore1
vmm18.mydomain.com:/gluster_bricks/brick3(/dev/sdc) data1
vmm18.mydomain.com:/gluster_bricks/brick4(/dev/sdd) data

with your first suggestion I dont think it is possible to recover as I will lose the engine if I stop the "engine" volume, It might be doable for vmstore1, data1 and data2 but not the engine.
A) If you have space on another gluster volume (or volumes) or on NFS-based storage, you can migrate all VMs live . Once you do it,  the simple way will be to stop and remove the storage domain (from UI) and gluster volume that correspond to the problematic brick. Once gone, you can  remove the entry in oVirt for the old host and add the newly built one. Then you can recreate your volume and migrate the data back.

I tried removing the brick using CLI but get the following error:
volume remove-brick start: failed: Host node of the brick vmm10.mydomain.com:/gluster_bricks/engine/engine is down

So I used the force command:
gluster vol remove-brick engine vmm10.mydomain.com:/gluster_bricks/engine/engine  vmm11.mydomain.com:/gluster_bricks/engine/engine  vmm12.mydomain.com:/gluster_bricks/engine/engine force
Remove-brick force will not migrate files from the removed bricks, so they will no longer be available on the volume.
Do you want to continue? (y/n) y
volume remove-brick commit force: success

so I lost my engine:
Please enter your authentication name: vdsm@ovirt
Please enter your password:
 Id    Name                           State
----------------------------------------------------
 3     HostedEngine                   paused

 hosted-engine --vm-start
The hosted engine configuration has not been retrieved from shared storage. Please ensure that ovirt-ha-agent is running and the storage server is reachable.

I guess this fail scenario is more complex than I thought, hosted engine should of survived, as far as gluster I am able to get around command line, the issue is the engine, though it was running on vmm18 and not running on any bricks belonging to vmm10, 11, or 12 (original setup) it still failed...
virsh list --all
Please enter your authentication name: vdsm@ovirt
Please enter your password:
 Id    Name                           State
----------------------------------------------------
 -     HostedEngine                   shut off

Now I cant get it to start:
hosted-engine --vm-start
The hosted engine configuration has not been retrieved from shared storage. Please ensure that ovirt-ha-agent is running and the storage server is reachable.
df -hT still showing mounts from old hosts bricks, could the problem be that this was the initiating host of the hyperconverged setup?
vmm10.mydomain.com:/engine                                       fuse.glusterfs  200G  6.2G  194G   4% /rhev/data-center/mnt/glusterSD/vmm10.mydomain.com:_engine


I will re-create everything from scratch and simulate this again, and see why is it too complex to recover ovirt's engine with gluster when a server dies completely. Maybe it is my lack of understanding with regards how ovirt integrates with gluster though I have a decent understanding of Gluster to work with it...

I will let you know once I have the cluster recreated and will kill the same server and see if I missed anything from the recommendations your provided.

Thanks,

--
Adrian.







On Tue, Jun 11, 2019 at 4:13 PM Strahil Nikolov <hunter86_bg@yahoo.com> wrote:
Do you have empty space to store the VMs ? If yes, you can always script the migration of the disks via the API . Even a bash script and curl can do the trick.

About the /dev/sdb , I still don't get it . A pure "df -hT" from a node will make it way clear. I guess '/dev/sdb' is a PV and you got 2 LVs ontop of it.

Note: I should admit that as an admin - I don't use UI for gluster management.

For now do not try to remove the brick. The approach is either to migrate the qemu disks to another storage or to reset-brick/replace-brick in order to restore the replica count.
I will check the file and I will try to figure it out.

Redeployment never fixes the issue, it just speeds up the recovery. If you can afford the time to spent on fixing the issue - then do not redeploy.

I would be able to take a look next week , but keep in mind that I'm not so in deep with oVirt - I have started playing with it when I deployed my lab.

Best Regards,
Strahil Nikolov

Strahil,

Looking at your suggestions I think I need to provide a bit more info on my current setup.


  1. I have 9 hosts in total

  2. I have 5 storage domains:

    • hosted_storage (Data Master)

    • vmstore1 (Data)

    • data1 (Data)

    • data2 (Data)

    • ISO (NFS) //had to create this one because oVirt 4.3.3.1 would not let me upload disk images to a data domain without an ISO (I think this is due to a bug)

  3. Each volume is of the type “Distributed Replicate” and each one is composed of 9 bricks.
    I started with 3 bricks per volume due to the initial Hyperconverged setup, then I expanded the cluster and the gluster cluster by 3 hosts at a time until I got to a total of 9 hosts.

    • Disks, bricks and sizes used per volume
      / dev/sdb engine 100GB
      / dev/sdb vmstore1 2600GB
      / dev/sdc data1 2600GB
      / dev/sdd data2 2600GB
      / dev/sde -------- 400GB SSD Used for caching purposes

      From the above layout a few questions came up:

      1. Using the web UI, How can I create a 100GB brick and a 2600GB brick to replace the bad bricks for “engine” and “vmstore1” within the same block device (sdb) ?

        What about / dev/sde (caching disk), When I tried creating a new brick thru the UI I saw that I could use / dev/sde for caching but only for 1 brick (i.e. vmstore1) so if I try to create another brick how would I specify it is the same / dev/sde device to be used for caching?


  1. If I want to remove a brick and it being a replica 3, I go to storage > Volumes > select the volume > bricks once in there I can select the 3 servers that compose the replicated bricks and click remove, this gives a pop-up window with the following info:

    Are you sure you want to remove the following Brick(s)?
    - vmm11:/gluster_bricks/vmstore1/vmstore1
    - vmm12.virt.iad3p:/gluster_bricks/vmstore1/vmstore1
    - 192.168.0.100:/gluster-bricks/vmstore1/vmstore1
    -
    Migrate Data from the bricks?

    If I proceed with this that means I will have to do this for all the 4 volumes, that is just not very efficient, but if that is the only way, then I am hesitant to put this into a real production environment as there is no way I can take that kind of a hit for +500 vms :) and also I wont have that much storage or extra volumes to play with in a real sceneario.

  2. After modifying yesterday / etc/vdsm/vdsm.id by following (https://stijn.tintel.eu/blog/2013/03/02/ovirt-problem-duplicate-uuids) I was able to add the server back to the cluster using a new fqdn and a new IP, and tested replacing one of the bricks and this is my mistake as mentioned in #3 above I used / dev/sdb entirely for 1 brick because thru the UI I could not separate the block device and be used for 2 bricks (one for the engine and one for vmstore1). So in the “gluster vol info” you might see vmm102.mydomain.com but in reality it is myhost1.mydomain.com

  3. I am also attaching gluster_peer_status.txt and in the last 2 entries of that file you will see and entry vmm10.mydomain.com (old/bad entry) and vmm102.mydomain.com (new entry, same server vmm10, but renamed to vmm102). Also please find gluster_vol_info.txt file.

  4. I am ready to redeploy this environment if needed, but I am also ready to test any other suggestion. If I can get a good understanding on how to recover from this I will be ready to move to production.

  5. Wondering if you’d be willing to have a look at my setup through a shared screen?


Thanks


Adrian


On Mon, Jun 10, 2019 at 11:41 PM Strahil <hunter86_bg@yahoo.com> wrote:

Hi Adrian,

You have several options:
A) If you have space on another gluster volume (or volumes) or on NFS-based storage, you can migrate all VMs live . Once you do it,  the simple way will be to stop and remove the storage domain (from UI) and gluster volume that correspond to the problematic brick. Once gone, you can  remove the entry in oVirt for the old host and add the newly built one.Then you can recreate your volume and migrate the data back.

B)  If you don't have space you have to use a more riskier approach (usually it shouldn't be risky, but I had bad experience in gluster v3):
- New server has same IP and hostname:
Use command line and run the 'gluster volume reset-brick VOLNAME HOSTNAME:BRICKPATH HOSTNAME:BRICKPATH commit'
Replace VOLNAME with your volume name.
A more practical example would be:
'gluster volume reset-brick data ovirt3:/gluster_bricks/data/brick ovirt3:/gluster_ ricks/data/brick commit'

If it refuses, then you have to cleanup '/gluster_bricks/data' (which should be empty).
Also check if the new peer has been probed via 'gluster peer status'.Check the firewall is allowing gluster communication (you can compare it to the firewalls on another gluster host).

The automatic healing will kick in 10 minutes (if it succeeds) and will stress the other 2 replicas, so pick your time properly.
Note: I'm not recommending you to use the 'force' option in the previous command ... for now :)

- The new server has a different IP/hostname:
Instead of 'reset-brick' you can use  'replace-brick':
It should be like this:
gluster volume replace-brick data old-server:/path/to/brick new-server:/new/path/to/brick commit force

In both cases check the status via:
gluster volume info VOLNAME

If your cluster is in production , I really recommend you the first option as it is less risky and the chance for unplanned downtime will be minimal.

The 'reset-brick'  in your previous e-mail shows that one of the servers is not connected. Check peer status on all servers, if they are less than they should check for network and/or firewall issues.
On the new node check if glusterd is enabled and running.

In order to debug - you should provide more info like 'gluster volume info' and the peer status from each node.

Best Regards,
Strahil Nikolov

On Jun 10, 2019 20:10, Adrian Quintero <adrianquintero@gmail.com> wrote:

>
> Can you let me know how to fix the gluster and missing brick?,
> I tried removing it by going to "storage > Volumes > vmstore > bricks > selected the brick
> However it is showing as an unknown status (which is expected because the server was completely wiped) so if I try to "remove", "replace brick" or "reset brick" it wont work
> If i do remove brick: Incorrect bricks selected for removal in Distributed Replicate volume. Either all the selected bricks should be from the same sub volume or one brick each for every sub volume!
> If I try "replace brick" I cant because I dont have another server with extra bricks/disks
> And if I try "reset brick": Error while executing action Start Gluster Volume Reset Brick: Volume reset brick commit force failed: rc=-1 out=() err=['Host myhost1_mydomain_com  not connected']
>
> Are you suggesting to try and fix the gluster using command line? 
>
> Note that I cant "peer detach"   the sever , so if I force the removal of the bricks would I need to force downgrade to replica 2 instead of 3? what would happen to oVirt as it only supports replica 3?
>
> thanks again.
>
> On Mon, Jun 10, 2019 at 12:52 PM Strahil <hunter86_bg@yahoo.com> wrote:

>>
>> Hi Adrian,
>> Did you fix the issue with the gluster and the missing brick?
>> If yes, try to set the 'old' host in maintenance an



--
Adrian Quintero


--
Adrian Quintero


--
Adrian Quintero