HA with ESXi Embedded

[WillGreen]

I have just got an evaluation of VMware Infrastructure working with some HP BL495 blades using ESXi embedded (on internal USB flash drives).

I have configured a datastore on an NFS partition on a NetApp datastores and the VMs run fine and VMotion works correctly.

However, when I try to create a cluster I get the following error:

HA agent has an error: Host in HA Cluster must have userworld swap enabled

I have tried following the KB article: but am not what path I should use with an NFS data store. All the obvious options fail.

I have seen a few comments around the web that suggest that it's not possible to use HA with userworld swap on NFS or iSCSI storage. Is this really the case? Is there any workaround if it is?

[vmroyale]

Hello. I actually edited this post to remove the questions, after I re-read it again. My apologies for any confusion this causes.

You should be able to create a nfs volume on the NetApp and then direct ScratchConfig.ConfiguredScratchLocation towards it. This could be similar to the process outlined on p.63 of TR3428.

Also check this discussion for more info.

Good Luck!

[WillGreen]

Thank you for your reply.

The hosts are HP BL495s with 32GB of RAM and two AMD Opteron 2384. I am using the HP supplied embedded ESXi 3.5.0 build-123629 2008-10-15.

The host network and NFS settings are as advised in the NetApp best practices guide you mention. The hosts are using a dedicated NFS swap volume.

I create a new cluster. Add a host (with no VMs) to it. It adds successfully, but fails with the userworld swap message. I have tried both swap settings for the cluster: Store the swap file in the same directory.... and store the swapfile in the datastore specified by the host. It makes no difference.

VMs run fine on the hosts and if the hosts are outside the cluster I can successfully VMotion between them.

[WillGreen]

Do you know what value I should use for ScratchConfig.ConfiguredScratchLocation? I'm not clear how I indicate my swap NFS mount: the example in the KB refers to VMFS. My datastore is called vmswap, is on server 172.16.0.01 and is mounted from /vol/vmswap.

[WillGreen]

Interestingly vmware let me specify a path of /vmfs/volumes/vmware, but once I rebooted this had changed to:

/vmfs/volumes/9f2a59a2-60ea5b66

However, I was able to add the host to HA.

I'm not convinced this is right though. Any thoughts?

[AntonVZhbankov · Accepted solution]

This scary string is an actual name of datastore. Name you see in VI client is just a symbolic link, so it's ok.

My recommendation: do not point uwswap to datastore. Create some folder like ./uwswap/---

VMware vExpert '2009

http://blog.vadmin.ru

[WillGreen]

Thank you.

By logging onto an ESXi host and looking at /vmfs/volumes I can see that my NFS mounts are under there. I will now do as you suggest and create dedicated swap locations.

I think VMware should clarify with KB article.

[AntonVZhbankov]

To be more clear: do not point uwswap location to datastore root, of course. Create dedicated folders on datastore for each host.

---

VMware vExpert '2009

http://blog.vadmin.ru

[sakacc]

If you create a unique subdir off the NFS datastore for each ESXi host, you can point to it (make sure that your path in the scratch.config lcation is correct - a typo results in an error of course).

I agree that this should be in a VMware KB article - as it's not NetApp behavior - it's true of any NFS server (I did it with an EMC Celerra).