I'm struggling with the same problem (I say struggling because I'm still having stability issues for what i consider a stable cluster) but you can: - create a replica 3 engine gluster volume - create replica 2 data, iso and export volumes

Deploy the hosted-engine on the first VM (with the engine volume) froom the CLI, then log in Ovirt admin, enable gluster support, install *and deploy* from the GUI host2 and host3 (where the engine bricks are) and then install host4 without deploying. This should get you the 4 hosts online, but the engine will run only on the first 3

2016-09-23 11:14 GMT+02:00 Goorkate, B.J. <b.j.goorkate(a)umcutrecht.nl&gt;: > Dear all, > > I've tried to find a way to add a 4th oVirt-node to my existing > 3-node setup with replica-3 gluster storage, but found no usable > solution yet. > > From what I read, it's not wise to create a replica-4 gluster > storage, because of bandwith overhead. > > Is there a safe way to do this and still have 4 equal oVirt nodes? > > Thanks in advance! > > Regards, > > Bertjan > > ------------------------------------------------------------ > ------------------ > > De informatie opgenomen in dit bericht kan vertrouwelijk zijn en is > uitsluitend bestemd voor de geadresseerde. Indien u dit bericht onterecht > ontvangt, wordt u verzocht de inhoud niet te gebruiken en de afzender > direct > te informeren door het bericht te retourneren. Het Universitair Medisch > Centrum Utrecht is een publiekrechtelijke rechtspersoon in de zin van de > W.H.W. > (Wet Hoger Onderwijs en Wetenschappelijk Onderzoek) en staat > geregistreerd bij > de Kamer van Koophandel voor Midden-Nederland onder nr. 30244197. > > Denk s.v.p aan het milieu voor u deze e-mail afdrukt. > > ------------------------------------------------------------ > ------------------ > > This message may contain confidential information and is intended > exclusively > for the addressee. If you receive this message unintentionally, please do > not > use the contents but notify the sender immediately by return e-mail. > University > Medical Center Utrecht is a legal person by public law and is registered > at > the Chamber of Commerce for Midden-Nederland under no. 30244197. > > Please consider the environment before printing this e-mail. > _______________________________________________ > Users mailing list > Users(a)ovirt.org > http://lists.ovirt.org/mailman/listinfo/users > -- Davide Ferrari Senior Systems Engineer _______________________________________________ Users mailing list Users(a)ovirt.org http://lists.ovirt.org/mailman/listinfo/users

2016-09-23 13:17 GMT+02:00 Sahina Bose <sabose(a)redhat.com&gt;: > > What are the stability issues you're facing? Data volume if used as a > data storage domain should be a replica 3 volume as well. > Basically that after the first host installation+deploy (from the CLI), after I enable the gluster management in the cluster, I have to manually restart vdsmd on host1 to be able to install the other hosts. But maybe I should just wait more time for vdmsd catch up with everything, I don't know.

Then I have some other problem like a ghost VM stuck on one host after moving the host to maintenance and the VM (the hosted-engine, the only one running in the whole cluster) being correctly migrated to another host, solved only by a manual reboot of the whole host (and consequent HE fencing of the host). I must say that that particular host is giving ECC correction errors in one DIMM, so maybe it could just be an HW related problem.

> >> Deploy the hosted-engine on the first VM (with the engine volume) froom >> the CLI, then log in Ovirt admin, enable gluster support, install *and >> deploy* from the GUI host2 and host3 (where the engine bricks are) and then >> install host4 without deploying. This should get you the 4 hosts online, >> but the engine will run only on the first 3 >> > > Right. You can add the 4th node to the cluster, but not have any bricks > on this volume in which case VMs will be run on this node but will access > data from the other 3 nodes. > Well, actually I *do* have data bricks on the 4th host, it's just the engine volume that's not present there (but that host is not HE eligible anyway). Am I doing something wrong?

2016-09-23 13:50 GMT+02:00 Sahina Bose <sabose(a)redhat.com&gt;: Ok, if I encounter again similiar problems I will post logs here > > If you have additional capacity on the other 3 hosts , then yes, you can > create a new gluster volume with a brick on the newly added 4th node and > bricks from other nodes - this volume can be used as another storage > domain. You are not doing anything wrong :) Keep in mind that all gluster > volumes used as data storage domains should be replica 3 or replica 3 > -arbiter to avoid split-brain and data loss issues. > > Mmmmh this is ringing an alarm bell then. So, it's basically impossible (or at least not supported) a 4 hosts configuration with all the 4 hosts having a data domain in a replica 2 fashion? Is it only replica 3 arbiter 1 the supported HA configuration? So if I want to expand storage (a part from adding disks to the same machines) I must add machines 3 by 3 ? Currently I have 4 machines with 4 disks each in a RAID-10 configuration, exposed as one brick. Which is the best HA solution in this scenario then?

Thanks -- Davide Ferrari Senior Systems Engineer

2016-09-23 15:57 GMT+02:00 Sahina Bose <sabose(a)redhat.com&gt;: > > You could do this - where Node3 & Node 2 also has arbiter bricks. Arbiter > bricks only store metadata and requires very low storage capacity compared > to the data bricks. > > Node1 Node2 Node3 Node4 > brick1 brick1 arb-brick > arb-brick brick1 brick1 > Ok, cool! And this won't pose any problem if Node2 or Node4 fail? The syntax shuld be this: gluster volume create data replica 3 arbiter 1 node1:/brick node2:/brick node2:/arb_brick node3:/brick node4:/brick node4:/arb_brick is not a problem having more than a brick on the same host for the volume create syntax? Thanks again -- Davide Ferrari Senior Systems Engineer

Reading the glusterfs docs https://gluster.readthedocs.io/en/latest/Administrator% 20Guide/arbiter-volumes-and-quorum/ "In a replica 3 volume, client-quorum is enabled by default and set to 'auto'. This means 2 bricks need to be up for the writes to succeed. Here is how this configuration prevents files from ending up in split-brain:" So this means that if one of the machines with the 2 bricks (arbiter & normal) fails, the otherbrick will be set RO, or am I missing something? I mean, this config will be better in case of a network loss, and thus a split brain, but it's far worse in case of a machine failing or being rebooted for maintenance.

2016-09-23 16:11 GMT+02:00 Davide Ferrari <davide(a)billymob.com&gt;: > > > 2016-09-23 15:57 GMT+02:00 Sahina Bose <sabose(a)redhat.com&gt;: > >> >> You could do this - where Node3 & Node 2 also has arbiter bricks. >> Arbiter bricks only store metadata and requires very low storage capacity >> compared to the data bricks. >> >> Node1 Node2 Node3 Node4 >> brick1 brick1 arb-brick >> arb-brick brick1 brick1 >> > > Ok, cool! And this won't pose any problem if Node2 or Node4 fail? > > The syntax shuld be this: > > gluster volume create data replica 3 arbiter 1 node1:/brick node2:/brick > node2:/arb_brick node3:/brick node4:/brick node4:/arb_brick > > is not a problem having more than a brick on the same host for the volume > create syntax? > > Thanks again > > -- > Davide Ferrari > Senior Systems Engineer > -- Davide Ferrari Senior Systems Engineer

On Fri, Sep 23, 2016 at 4:14 PM, Davide Ferrari <davide(a)billymob.com&gt; wrote: I'm struggling with the same problem (I say struggling because I'm still having stability issues for what i consider a stable cluster) but you can: - create a replica 3 engine gluster volume - create replica 2 data, iso and export volumes What are the stability issues you're facing? Data volume if used as a data storage domain should be a replica 3 volume as well. Deploy the hosted-engine on the first VM (with the engine volume) froom the CLI, then log in Ovirt admin, enable gluster support, install *and deploy* from the GUI host2 and host3 (where the engine bricks are) and then install host4 without deploying. This should get you the 4 hosts online, but the engine will run only on the first 3 Right. You can add the 4th node to the cluster, but not have any bricks on this volume in which case VMs will be run on this node but will access data from the other 3 nodes. 2016-09-23 11:14 GMT+02:00 Goorkate, B.J. <b.j.goorkate(a)umcutrecht.nl&gt;: Dear all, I've tried to find a way to add a 4th oVirt-node to my existing 3-node setup with replica-3 gluster storage, but found no usable solution yet. From what I read, it's not wise to create a replica-4 gluster storage, because of bandwith overhead. Is there a safe way to do this and still have 4 equal oVirt nodes? Thanks in advance! Regards, Bertjan ------------------------------------------------------------ ------------------ De informatie opgenomen in dit bericht kan vertrouwelijk zijn en is uitsluitend bestemd voor de geadresseerde. Indien u dit bericht onterecht ontvangt, wordt u verzocht de inhoud niet te gebruiken en de afzender direct te informeren door het bericht te retourneren. Het Universitair Medisch Centrum Utrecht is een publiekrechtelijke rechtspersoon in de zin van de W.H.W. (Wet Hoger Onderwijs en Wetenschappelijk Onderzoek) en staat geregistreerd bij de Kamer van Koophandel voor Midden-Nederland onder nr. 30244197. Denk s.v.p aan het milieu voor u deze e-mail afdrukt. ------------------------------------------------------------ ------------------ This message may contain confidential information and is intended exclusively for the addressee. If you receive this message unintentionally, please do not use the contents but notify the sender immediately by return e-mail. University Medical Center Utrecht is a legal person by public law and is registered at the Chamber of Commerce for Midden-Nederland under no. 30244197. Please consider the environment before printing this e-mail. _______________________________________________ Users mailing list Users(a)ovirt.org http://lists.ovirt.org/mailman/listinfo/users -- Davide Ferrari Senior Systems Engineer _______________________________________________ Users mailing list Users(a)ovirt.org http://lists.ovirt.org/mailman/listinfo/users

On Tue, Sep 27, 2016 at 8:59 PM, Goorkate, B.J. <b.j.goorkate(a)umcutrecht.nl&gt; wrote: Hi Sahina, First: sorry for my delayed response. I wasn't able to respond earlier. I already planned on adding the 4th node as a gluster client, so thank you for confirming that this works. Why I was in doubt is that certain VMs with a lot of storage I/O on the 4th node have to replicate to 3 other hosts (the replica-3 gluster nodes) over the storage network, while a VM on 1 of the replica-3 gluster nodes only has to replicate to two other nodes over the network, thus creating less network traffic. Does this make sense? And if it does: can that be an issue? IIUC, the 4th node that you add to the cluster is serving only compute and there is no storage (bricks) capacity added. In this case, yes, all reads and writes are over the network - this is like a standard oVirt deployment where storage is over the network (non hyper converged). While thoeretically this looks like an issue, it may not be, as there are multiple factors affecting performance. You will need to measure the impact on guest performance when VMs run on this node and see if it is acceptable to you. One thing you could do is schedule VMs that do not have stringent perf requirements on the 4th node? There are also improvements planned in upcoming releases of gluster which improve the I/O performance further (compound FOPS, libgfapi access), so whatever you see now should improve further. Regards, Bertjan On Fri, Sep 23, 2016 at 04:47:25PM +0530, Sahina Bose wrote: > > > On Fri, Sep 23, 2016 at 4:14 PM, Davide Ferrari <davide(a)billymob.com&gt; wrote: > > I'm struggling with the same problem (I say struggling because I'm still > having stability issues for what i consider a stable cluster) but you can: > - create a replica 3 engine gluster volume > - create replica 2 data, iso and export volumes > > > What are the stability issues you're facing? Data volume if used as a data > storage domain should be a replica 3 volume as well. > > > > Deploy the hosted-engine on the first VM (with the engine volume) froom the > CLI, then log in Ovirt admin, enable gluster support, install *and deploy* > from the GUI host2 and host3 (where the engine bricks are) and then install > host4 without deploying. This should get you the 4 hosts online, but the > engine will run only on the first 3 > > > Right. You can add the 4th node to the cluster, but not have any bricks on this > volume in which case VMs will be run on this node but will access data from the > other 3 nodes. > > > > 2016-09-23 11:14 GMT+02:00 Goorkate, B.J. <b.j.goorkate(a)umcutrecht.nl >: > > Dear all, > > I've tried to find a way to add a 4th oVirt-node to my existing > 3-node setup with replica-3 gluster storage, but found no usable > solution yet. > > From what I read, it's not wise to create a replica-4 gluster > storage, because of bandwith overhead. > > Is there a safe way to do this and still have 4 equal oVirt nodes? > > Thanks in advance! > > Regards, > > Bertjan > > ------------------------------------------------------------ > ------------------ > > De informatie opgenomen in dit bericht kan vertrouwelijk zijn en is > uitsluitend bestemd voor de geadresseerde. Indien u dit bericht > onterecht > ontvangt, wordt u verzocht de inhoud niet te gebruiken en de afzender > direct > te informeren door het bericht te retourneren. Het Universitair Medisch > Centrum Utrecht is een publiekrechtelijke rechtspersoon in de zin van > de W.H.W. > (Wet Hoger Onderwijs en Wetenschappelijk Onderzoek) en staat > geregistreerd bij > de Kamer van Koophandel voor Midden-Nederland onder nr. 30244197. > > Denk s.v.p aan het milieu voor u deze e-mail afdrukt. > > ------------------------------------------------------------ > ------------------ > > This message may contain confidential information and is intended > exclusively > for the addressee. If you receive this message unintentionally, please > do not > use the contents but notify the sender immediately by return e-mail. > University > Medical Center Utrecht is a legal person by public law and is > registered at > the Chamber of Commerce for Midden-Nederland under no. 30244197. > > Please consider the environment before printing this e-mail. > _______________________________________________ > Users mailing list > Users(a)ovirt.org > http://lists.ovirt.org/mailman/listinfo/users > > > > > -- > Davide Ferrari > Senior Systems Engineer > > _______________________________________________ > Users mailing list > Users(a)ovirt.org > http://lists.ovirt.org/mailman/listinfo/users > > >