Can Multipathing on Fibre Channel be changed while VM's are running on host? Any Impact?

[Johan_Marais1]

Good Day Guys

I got a new client which was previously managed by another company, to say the least this is environment is so messed up. One of the things I noticed is that a Fibre Channel Connector MutliPathing on Host01 is configured as VMware (Fixed) ..... the same fibre Connector on Host02 is configured to VMware (Round Robin) I want to go through all the Hosts and change all the adapters multipathing to VMware (Round Robin). I already checked with the Storage Vendor and Round Robin is recommended by them.

If I change the Multipathing from VMware (Fixed) on the adapters to all be on VMWare (Round Robin) will this have any disconnects of any Running VM's or impact of downtime? This is crucial for me to log the Change Control.

Regards

Johan

[daphnissov]

There *could* be, so my recommendation when fiddling with storage settings on an ESXi host with running VMs is to put it into maintenance mode first to be safe.

[Johan_Marais1]

Thanks Daphnissov - Much Appreciated a bit difficult getting all the VM's of the host to another Host as their resources wont allow for all the VM's to my migrated to the other 2 host without 100% Utilization on the remainder  of the 2 Hosts

[Johan_Marais1]

Even if the Disconnect is temp and quick its okay ..... its difficult to get complete downtime on these systems. If they lose connection breifly that would be okay - Whats your thoughts? They would reconnect to storage fairly quickly after multipathing change right?

[a_p_]

I've done this many times, and never experienced any issues. The recommendation is to do this off hours, i.e. when usage isn't very high.

André

[Johan_Marais1]

Thanks Andre

I have never done this while VM's are running and just afraid it might disconnect the storage to the VM's and they orphan/ofline or something ........ this is now changing the fibre channel multipathing from VMware (Most Recently used) to VMware (Round Robin) ...... I noticed that a lot of my storage disconnects/re-connects are from one specific host. So I browse to my SAN Storage datastores and select manage, then connectivity & multipathing - there if I click on the data stores i can see how they are configured for each host. All the other hosts have the datastores configured as VMware (Round Robin) while only this host has it configured for all Data Stores as VMware(Most Recently Used). You recon I can change it to Round Robin without having any downtime on the VM's?

Regards

Johan

[a_p_]

As mentioned before, I've done this many times in production environments, and never had any issues.

The key is to ensure that the Round-Robin path policy is supported by the storage system, and that there's no reason not to use it (e.g. for MSCS RDM LUNs on older ESXi versions). You may also consider to pick a time when I/O is low on the VMs running on that host.

André

[Johan_Marais1]

Thanks Andre, yes i did check, the Vendor actually recommends Rounds Robin for their Storage.

Here is my Setup & Issues explained from bullet 4:

• 4 x ESXi 5.1 Hosts
• Using HP EVA Storage Aray & Using Netapp SAN storage
• Hosts 2, 3 & 4 The fibre channel adapters are configured to VMware Round Robin (Random times they have disconnects to the storage and almost immediately reconnects)
• Host 5 This one has the most disconnects/ re-connects to the storage which cause snaps taking more than a hour to delete after successful backup. This is also the Host where all the Fibre Channel  adapters are configure as Fixed VMware, this I want to change to Round Robin and see if it minimizes the storage disconnects/reconnects
• Also I saw that the HP Fibre RAID Controller is on Fixed (Vmware) ...... should this be changed to Round Robin too? Impact?
• Last screenshot is of the Host 5 losing connections to the Volumes although no VM's go Down.

[dhanarajramesh]

Surely you can do it on fly and it wont affect anything at the back as you have one active link connected. once enabled, the host will just start distribute the load to other available links.  if still concerned, just migrate all the vm from one host and leave  2 or 3 low priority vm and change it on fly and observe the flow.

[daphnissov]

Just clarification from my side, you may be able to do this change online just fine. Often times it does not cause any impact as mentioned by others. But I have seen issues from some customers where making this change can cause issues with I/O. In a couple of cases, this caused ESXi to hang and all VMs to go down. This is not a typical response, but it illustrates that the safest way to make changes to a hypervisor's I/O policy is to do so with no VMs running on it.

[Johan_Marais1]

daphnissov​

Thanks for the reply, I have decided to shutdown all the VM's on the Host then make the changes with the VM's shutdown remianing on the Host. Once the changes have been done I will boot the VM's up again. Is that a better option?

[daphnissov]

That would be a *safe* option, but may not be *better*. Again, though, making this change online is usually fine, but that's by no means guaranteed. Most times I've made this changes, things are fine, but the fact is when you alter how I/O is issued to backend storage while I/Os are in-flight, you always run the risk interrupting that path. If you want to be absolutely certain and safe that no harm will come, do so with VMs that are not running--however you'd like to ensure that.

[Johan_Marais1]

daphnissov

I meant I will shutdown all the VM's on that Host before making the changes. But the VM's will remain on the Host in the Shutdown state and then be booted after the changes was done...... Your answer confuses me a bit LOL ..... will this be the safest option?

[daphnissov]

That would be fine.

[a_p_]

CAUTION: AFAIK Round-Robin is only supported for disks/LUNs. Don't change the path policy for the "array" type, unless you find a documentation regarding this, or if someone can confirm that this is supported.

André