On Feb 5, 2018 5:00 PM, "Luca 'remix_tj' Lorenzetto" < lorenzetto.luca(a)gmail.com&gt; wrote: Hello, i'm starting the implementation of our disaster recovery site with RHV 4.1.latest for our production environment. Our production setup is very easy, with self hosted engine on dc KVMPDCA, and virtual machines both in KVMPDCA and KVMPD dcs. All our setup has an FC storage backend, which is EMC VPLEX/VMAX in KVMPDCA and EMC VNX8000. Both storage arrays supports replication via their own replication protocols (SRDF, MirrorView), so we'd like to delegate to them the replication of data to the remote site, which is located on another remote datacenter. In KVMPD DC we have some storage domains that contains non critical VMs, which we don't want to replicate to remote site (in case of failure they have a low priority and will be restored from a backup). In our setup we won't replicate them, so will be not available for attachment on remote site. Can be this be an issue? Do we require to replicate everything?

What about master domain? Do i require that the master storage domain stays on a replicated volume or can be any of the available ones?

I've seen that since 4.1 there's an API for updating OVF_STORE disks. Do we require to invoke it with a frequency that is the compatible with the replication frequency on storage side.

We set at the moment RPO to 1hr (even if planned RPO requires 2hrs). Does OVF_STORE gets updated with the required frequency?

I've seen a recent presentation by Maor Lipchuk that is showing the "automagic" ansible role for disaster recovery: https://www.slideshare.net/maorlipchuk/ovirt-dr-site-tosite-using-ansible It's also related with some youtube presentations demonstrating a real DR plan execution. But what i've seen is that Maor is explicitly talking about 4.2 release. Does that role works only with >4.2 releases or can be used also on earlier (4.1) versions? Releases before 4.2 do not store complete information on the OVF store to perform such comprehensive failover. I warmly suggest 4.2! Y.

I've tested a manual flow of replication + recovery through Import SD followed by Import VM and worked like a charm. Using a prebuilt ansible role will reduce my effort on creating a new automation for doing this. Anyone has experiences like mine? Thank you for the help you may provide, i'd like to contribute back to you with all my findings and with an usable tool (also integrated with storage arrays if possible).

Luca (Sorry for duplicate email, ctrl-enter happened before mail completion) -- "E' assurdo impiegare gli uomini di intelligenza eccellente per fare calcoli che potrebbero essere affidati a chiunque se si usassero delle macchine" Gottfried Wilhelm von Leibnitz, Filosofo e Matematico (1646-1716) "Internet è la più grande biblioteca del mondo. Ma il problema è che i libri sono tutti sparsi sul pavimento" John Allen Paulos, Matematico (1945-vivente) Luca 'remix_tj' Lorenzetto, http://www.remixtj.net , < lorenzetto.luca(a)gmail.com&gt; _______________________________________________ Users mailing list Users(a)ovirt.org http://lists.ovirt.org/mailman/listinfo/users

On Mon, Feb 5, 2018 at 7:20 PM, Maor Lipchuk <mlipchuk(a)redhat.com&gt; wrote: > Hi Luca, > > Thank you for your interst in the Disaster Recovery ansible solution, it is > great to see users get familiar with it. > Please see my comments inline > > Regards, > Maor > > On Mon, Feb 5, 2018 at 7:54 PM, Yaniv Kaul <ykaul(a)redhat.com&gt; wrote: >> >> >> >> On Feb 5, 2018 5:00 PM, "Luca 'remix_tj' Lorenzetto" >> <lorenzetto.luca(a)gmail.com&gt; wrote: >> >> Hello, >> >> i'm starting the implementation of our disaster recovery site with RHV >> 4.1.latest for our production environment. >> >> Our production setup is very easy, with self hosted engine on dc >> KVMPDCA, and virtual machines both in KVMPDCA and KVMPD dcs. All our >> setup has an FC storage backend, which is EMC VPLEX/VMAX in KVMPDCA >> and EMC VNX8000. Both storage arrays supports replication via their >> own replication protocols (SRDF, MirrorView), so we'd like to delegate >> to them the replication of data to the remote site, which is located >> on another remote datacenter. >> >> In KVMPD DC we have some storage domains that contains non critical >> VMs, which we don't want to replicate to remote site (in case of >> failure they have a low priority and will be restored from a backup). >> In our setup we won't replicate them, so will be not available for >> attachment on remote site. Can be this be an issue? Do we require to >> replicate everything? > > > No, it is not required to replicate everything. > If there are no disks on those storage domains that attached to your > critical VMs/Templates you don't have to use them as part of yout mapping > var file > Excellent. >> >> What about master domain? Do i require that the master storage domain >> stays on a replicated volume or can be any of the available ones? > > > > You can choose which storage domains you want to recover. > Basically, if a storage domain is indicated as "master" in the mapping var > file then it should be attached first to the Data Center. > If your secondary setup already contains a master storage domain which you > dont care to replicate and recover, then you can configure your mapping var > file to only attach regular storage domains, simply indicate > "dr_master_domain: False" in the dr_import_storages for all the storage > domains. (You can contact me on IRC if you need some guidance with it) > Good, that's my case. I don't need a new master domain on remote side, because is an already up and running setup where i want to attach replicated storage and run the critical VMs. >> >> >> I've seen that since 4.1 there's an API for updating OVF_STORE disks. >> Do we require to invoke it with a frequency that is the compatible >> with the replication frequency on storage side. > > > > No, you don't have to use the update OVF_STORE disk for replication. > The OVF_STORE disk is being updated every 60 minutes (The default > configuration value), > What i need is that informations about vms is replicated to the remote site with disk. In an older test i had the issue that disks were replicated to remote site, but vm configuration not! I've found disks in the "Disk" tab of storage domain, but nothing on VM Import.

>> >> We set at the moment >> RPO to 1hr (even if planned RPO requires 2hrs). Does OVF_STORE gets >> updated with the required frequency? > > > > OVF_STORE disk is being updated every 60 minutes but keep in mind that the > OVF_STORE is being updated internally in the engine so it might not be > synced with the RPO which you configured. > If I understood correctly, then you are right by indicating that the data of > the storage domain will be synced at approximatly 2 hours = RPO of 1hr + > OVF_STORE update of 1hr > We require that we can recover vms with a status that is up to 2 hours ago. In worst case, from what you say, i think we'll be able to. [cut] > > Indeed, > We also introduced several functionalities like detach of master storage > domain , and attach of "dirty" master storage domain which are depndant on > the failover process, so unfortunatly to support a full recovery process you > will need oVirt 4.2 env. > Ok, but if i keep master storage domain on a non replicate volume, do i require this function?

I have to admit that i require, for subscription and support requirements, to use RHV over oVirt. I've seen 4.2 is coming also from that side, and we'll upgrade for sure when available. [cut] > > > Please feel free to share your comments and questions, I would very > appreciate to know your user expirience. Sure, i'll do! And i'll bother you on irc if i need some guidance :-) Thank you so much, Luca -- "E' assurdo impiegare gli uomini di intelligenza eccellente per fare calcoli che potrebbero essere affidati a chiunque se si usassero delle macchine" Gottfried Wilhelm von Leibnitz, Filosofo e Matematico (1646-1716) "Internet è la più grande biblioteca del mondo. Ma il problema è che i libri sono tutti sparsi sul pavimento" John Allen Paulos, Matematico (1945-vivente) Luca 'remix_tj' Lorenzetto, http://www.remixtj.net , < lorenzetto.luca(a)gmail.com&gt;

On Tue, Feb 6, 2018 at 9:33 PM, Maor Lipchuk <mlipchuk(a)redhat.com&gt; wrote: [cut] >> What i need is that informations about vms is replicated to the remote >> site with disk. >> In an older test i had the issue that disks were replicated to remote >> site, but vm configuration not! >> I've found disks in the "Disk" tab of storage domain, but nothing on VM >> Import. > > > > Can you reproduce it and attach the logs of the setup before the disaster > and after the recovery? > That could happen in case of new created VMs and Templates which were not > yet updated in the OVF_STORE disk, since the OVF_STORE update process was > not running yet before the disaster. > Since the time of a disaster can't be anticipated, gaps like this might > happen. > I haven't tried the recovery yet using ansible. It was an experiment of possible procedure to be performed manually and was on 4.0. I asked about this unexpected behavior and Yaniv returned me that was due to OVF_STORE not updated and that in 4.1 there is an api call that updates OVF_STORE on demand. I'm creating a new setup today and i'll test again and check if i still hit the issue. Anyway if the problem persist i think that engine, for DR purposes, should upgrade the OVF_STORE as soon as possible when a new vm is created or has disks added.

[cut] >> >> Ok, but if i keep master storage domain on a non replicate volume, do >> i require this function? > > > Basically it should also fail on VM/Template registration in oVirt 4.1 since > there are also other functionalities like mapping of OVF attributes which > was added on VM/Templates registeration. > What do you mean? That i could fail to import any VM/Template? In what case?

Another question: we have 2 DCs in main site, do we require to have also 2 DCs in recovery site o we can import all the storage domains in a single DC on recovery site? There could be uuid collisions or similar?

Thank you so much for your replies, Luca -- "E' assurdo impiegare gli uomini di intelligenza eccellente per fare calcoli che potrebbero essere affidati a chiunque se si usassero delle macchine" Gottfried Wilhelm von Leibnitz, Filosofo e Matematico (1646-1716) "Internet è la più grande biblioteca del mondo. Ma il problema è che i libri sono tutti sparsi sul pavimento" John Allen Paulos, Matematico (1945-vivente) Luca 'remix_tj' Lorenzetto, http://www.remixtj.net , < lorenzetto.luca(a)gmail.com&gt;