TOPIC:

Hi there,
after a little lab we are going to think to implement a 2 node cluster with XCP-NG 7.5 and halizard + nosan option. The phisycal nodes are made by 2 Dell R470 Servers with 7x8TB 7200RPM SATA drives, 2 dual port 10GB nic and 2 1 gbit nics.
Our idea was to use 2 ports of the 2 10gb nic in lacp configuration for storage replication, the other 2 10 GB ports for vm traffic and the 2 1gb nic for the management.
The 2 machines will be placed in two different building and we are struggling on if do a direct connection for the storage replica or use the existing 10gb infrastructure.
Would you mind to give me an advice on this configuration?
Kind regards

Giorgio

Hi Giorgio
Direct link is always better especially if you plan on running live VMs on both the master and the slave. Directly attaching the replication interfaces eliminates the possibility of HW failure in between.

Since your dom0s are in separate buildings, there is still much than can go wrong even when using direct links.

I would recommend running all live VMs on the master only. This way, if your replication link is interrupted, the VMs will continue to have access to the storage. Otherwise, a replication link outage would prevent any VMs running on the slave from accessing the storage.

So, in summary, limiting VMs to run on the master will make this a safe environment. You are, of course, able to migrate machines freely for performing maintenance on your hosts.

Thabks that was also my thought. As an alternative there could be the possibility to jeep the less critical machine (also in termos of disk iops) on the slave node. In case of intwrruption (also not brief) the recovery mechanism is automatic or do i have to scratch the replicated storage?! I speak not about a split brain situation that with 2 node would be difficult. What do you suggest to use as a euristic IP address?! Would be using the core switch a good idea?! Last question: does the automatic script of ha and nosan wotks on xcp-ng?! Kind regards

HA will work fine for VMs running on the slave, however, there are some special considerations in your case that must be considered. We need to look at 3 types of failure scenarios to plan accordingly with hosts operating across a network between buildings:

1) Host failure
In this case, HA will work normally, and if either of the 2 hosts fails, any affected VMs will be started on the peer host. Additionally, if the master fails, the slave will be promoted to the new pool master. This is normal operation assuming both hosts can see the heuristic device AND there is no network connectivity problem on the management network.

2) Management Network outage.
Your situation is a bit unique given that the hosts are in separate buildings and more prone to network issues and loss of heuristic IP connectivity. If you lose just the management network, then, whichever of the 2 hosts can see the heuristic device(s) will take over the workloads and the host that cannot see the heuristic device will immediately shutdown all of its VMs and enter a restricted mode which blocks it from starting VMs until the network issue is restored. In this scenario, you have 2 choices. (1) use a single heuristic IP on one side of the network which allows 1 host to take over in the case of a network interruption (2) configure 2 heuristic IPs, one on each side of the network. In this case, a host MUST be able to see both IPs in order to recover services. In a network outage, BOTH hosts would immediately stop ALL VMs, leaving no VMs running because they cannot access all of the configured heuristic IPs.

3) Replication network outage (and management network still OK)
In a replication network outage, no HA event is triggered. Any VMs running on the slave will lose access to their disks and continue to run diskless. This would require manual intervention to clean up.

In answer to your last question. Yes, the installer is working fine on XCP 7.5.

Thanks a lot Salvatore.
I will insist either to have a clustered environment between the storage connection (lacp trunk possibly) or direct connection if possible, we'll see.
I've to ask you a clarification on 2 points regarding what you wrote about.
- Situation 1) is clear what happens in case of failure of the host my question is: what should I do once the host is repaired-restored? Do I have simply to reboot the host? Let's assume that the master node fails and that it stays down for 4 days, which are the actions we should put in place? In terms of Xen serve, if i do not force the promotion of the other node as a master it will be available again, my worries are related to the storage replica.
- Situation 2: Same doubt as before, what should I do once the connection is restored and, furthermore, It's not really clear your idea of 2 heuristic ip? Is this a parameters I can set up manually? If i remember properly the wizard ask for a single heuristic IP but the logic you explained would be really useful.

Kind regards

Giorgio

One more question, i forgot to mention it before.
The servers we have to work with are built with a set of two little ssd raid 1 (used for the installation of the OS) and a huge RAID5.
I don't know if the Xen installation wizard will distinguish the two storages permitting me to assign the huge one to "Local Storage" or if I have to introduce the huge one later on.
I do think that the script will recognize also a storage introduced after the installation and whose name is not Local storage, am I right?

Let me know, kind regards