NVMe/TCP and ESXi issue

[glass crane]

Hi,
for the first time I'm trying to configure a VMware datastore with NVMe/TCP.
I followed this guide from @elder nebula
https://community.netapp.com/t5/Tech-ONTAP-Blogs/How-to-Configure-NVMe-TCP-with-vSphere-8-0-Update-1-and-ONTAP-9-13-1-for-VMFS/ba-p/445429

I have an unknown issue with this. Let me details the 2 option (one works the other not)

1. If I configure the NetApp NVMe Subsystem with 1 ESXi host NQN and I map the NVMe Namespace with this subsystem, the ESXi can see the device.

2. If I configure the NetApp NVMe Subsystem with 3 ESXi host NQN and I map the NVMe Namespace with this, no ESXi can see the device.

Any help?

[reef stream]

Can you show a screenshot for the 2nd scenario? Do you get any error message?

[glass crane]

honestly i don't see any error..
the only thing I can see is that from ESXi no device appears...

[reef stream]

I meant the configuration in ONTAP

[glass crane]

here I am

the odd thng is that from UI, in the NVMe subsystem, I see something different

[glass crane]

inoltre controllando gli Host NQN sembra tutto ok

[glass crane]

I solved this

[sullen owl]

why is your SVM called "NFS" when it serves NVMe namespaces? 🤯 this would confuse me endlessly

[glass crane]

well, just the original name..
It's also possible to enable on the same SVM the NFS and NVMeTCP service..

[dark axle]

How crazy would @sullen owl go when a cluster named clus_2720 is actually an upgraded fas2820?

I totally love (no hate, really) when a customer decides it is a great idea to name the cluster the model of the controllers...to turn around a few years later to add nodes of a different mode or upgrade to a newer model

[sullen owl]

oh boy... 🙂
I have seen a MetroCluster yesterday where the cluster one site was called "corp-MC1" and the other side's cluster was called "cl-foo" or something. Totally unrelated names

another customer thought it would be a good idea to store the SVM name, protocol, node name and port in the name of each data LIF. like "ESX-nfs-node1-e0a-112"
When we had to move them around for some reason, he objected, "because then I will have to update and change all my scripts!"

[coarse ember]

Maybe we should start a thread about naming standard best practices 😀. Unless this already exists, I think it might be useful. For example, I never advertise the vserver name or LIF names to clients. I have DNS aliases created to the LIFs just in case we need to rename or re-IP. We simply only need to move the DNS alias to another "A" record when ready to shift clients.

[dry mauve]

Well... I seem to remember that "some" NetApp applications like SnapCenter do a lot of DNS lookups (also reverse) based on the SVM name... and sometimes things breaks in strange ways... so I alwas have a DNS record with the SVM name pointing to the management lif-IP of the SVM, just in case 🙂 And as SSL security becomes more and more standard, the DNS records have to match the certificate for it to work...