Yellow Bricks

by Duncan Epping

6.7

Which vSAN Maintenance Mode option was used on a host?

· ·

Last week there was a question on VMTN around maintenance mode, this customer wanted to find out which vSAN Maintenance Mode option was used while the host was placed in maintenance mode. The person who placed the host into maintenance wasn’t around, and I guess the customer wanted to know if data was moved from the host or not. They looked at the events and tasks, but unfortunately it wasn’t obvious from that vCenter section, next up were the logs. I also looked at the logs, and you would expect this info to be stored in hostd.log, but it wasn’t, so where is it? Well it actually is a configuration item, and you can dig it up here:

Pre ESXi 7.0 the info is stored in the “esx.conf” file, just grep for “hostDecommission”. Let me show you:

$ grep "/vsan/hostDecommission" /etc/vmware/esx.conf
/vsan/hostDecommissionVersion = "10"
/vsan/hostDecommissionState = "decom-state-decommissioned"
/vsan/hostDecommissionMode = "decom-mode-ensure-object-accessibility"

If the ESX is at 7.0 or later, just run the below config store command:

$ configstorecli config current get -c vsan -g system -k host_state
{
   "decom_mode": "ENSUREOBJECT_ACCESSIBILITY",
   "decom_state": "DECOMMISSIONED",
   "decom_version": 0
}

I hope that helps others who have a similar question!

VMs which are not stretched in a stretched cluster, which policy to use?

· ·

I’ve seen this question popping up regularly. Which policy setting (“site disaster tolerance” and “failures to tolerate”) should I use when I do not want to stretch my VMs? Well, that is actually pretty straight forward, in my opinion, you really only have two options you should ever use:

  • None – Keep data on preferred (stretched cluster)
  • None – Keep data on non-preferred (stretched cluster)

Yes, there is another option. This option is called “None – Stretched Cluster” and then there’s also “None – Standard Cluster”. Why should you not use these? Well, let’s start with “None – Stretched Cluster”. In the case of “None – Stretched Cluster”, vSAN will per object decide where to place it. As you hopefully know, a VM consists of multiple objects. As you can imagine, this is not optimal from a performance point of view, as you could end up having a VMDK being placed in Site A and a VMDK being placed in Site B. Which means it would read and write from both locations from a storage point of view, while the VM would be sitting in a single location from a compute point of view. It is also not very optimal from an availability stance, as it would mean that when the intersite link is unavailable, some objects of the VM would also become inaccessible. Not a great situation. What would it look like? Well, potentially something like the below diagram!

Then there’s “None – Standard Cluster”, what happens in this case? When you use “None – Standard Cluster” with “RAID-1”, what is going to happen is that the VM is configured with FTT=1 and RAID-1, but in a stretched cluster “FTT” does not exist, and FTT automatically will become PFTT. This means that the VM is going to be mirrored across locations, and you will have SFTT=0, which means no resiliency locally. It is the same as “Dual Site Mirroring”+”No Data Redundancy”!

In summary, if you ask me, “none – standard cluster” and “none – stretched cluster” should not be used in a stretched cluster.

vSphere HA reporting not enough failover resources fault with stretched cluster failure scenario

· ·

Last few months I had a couple of customers asking why vSphere HA was reporting “not enough failover resources” fault in a stretched cluster failure scenario for virtual machines that are still up and running. Now before I explain why, let’s paint a picture first to make it clear what is happening here. When you run a stretched cluster you can have a scenario where a particular VM (or multiple VMs) are not mirrored/replicated across locations. Now note, with vSAN you can specify for any given VM on a VM level how and if the VM should be available across locations. Typically you would see a VM with RAID-1 across locations, and then RAID-1/5/6 within a location. However, you can also have a scenario where a VM is not replicated across locations, but from a storage point of view only available within a location, this is depicted in the diagram below.

Now in this scenario, when Site A is somehow partitioned from Site B, you will see alarms/errors which indicate that vSphere HA has tried to restart the VM that is located in Site B in Site A and that is has failed as a result of not having enough failover resources.

This, of course, is not the result of not having sufficient failover resources, but it is the result of the fact that Site A does not have access to the required storage components to restart the VM. Basically what HA is reporting is that it doesn’t have the resources which have the ability to restart the impacted VM(s).

Now, if you have paid attention, you will probably wonder why HA tries to restart the VM in the first place, as the VM will still be running in this scenario. Why is it still running? Well the VM isn’t stretched, and this is a partition and not an isolation, which means the isolation response doesn’t kill the VM. So why restart it? Well, as Site A is partitioned from Site B, Site A does not know what the status is of Site B. Site A only knows that Site B is not responding at all, and the only thing it can do is assume the full site has failed. As a result it will attempt a failover for all VMs that were/are running in Site B and were protected by vSphere HA.

Hope that explains why this happens. If you are not sure you understand the full scenario, I recorded a quick five minute video actually walking through the scenario and explaining what happens. You can watch that below, or simply go to youtube.

Running vSphere 6.7 or 6.5 and configured HA APD/PDL responses? Read this…

· ·

If you are running vSphere 6.7 or 6.5 and have not installed 6.7 P02 yet (6.5 P05 is available soon) and you have APD/PDL responses configured within vSphere HA it could be that an issue causes VMs not to be failed over when an APD or PDL occurs. This is a known issue in the release, and P02 or P05 solves this problem. What is the problem? Well, a bug causes VMs which are listed in “VM overrides” to have settings that are not configured to be set to “disabled” instead of “unset”, in specific the APD/PDL setting.

This means that even though you have APD/PDL responses configured on a cluster level, the VM level configuration overrides it as it would be set to “disabled”. It doesn’t matter really why you added them to VM Overrides, could be to configure VM Restart Priority for instance. The frustrating part is that the UI doesn’t show you it is disabled as it looks like it is not configured.

If you can’t install the patch just yet, for whatever reason, but you do have VMs in VM Overrides, make sure to go to VM Overrides and explicitly configure the VMs to have the APD/PDL responses enabled similar to what it is configured to on a cluster level as shown in the screenshots below.

vGPUs and vMotion, why the long stun times?

· ·

Last week one of our engineers shared something which I found very interesting. I have been playing with Virtual Reality technology and NVIDIA vGPUs for 2 months now. One thing I noticed is that we (VMware) introduced support for vMotion in vSphere 6.7 and support for vMotion of multi vGPU VMs in vSphere 6.7 U3. In order to enable this, you need to set an advanced setting first. William Lam described this in his blog how to set this via Powershell or the UI. Now when you read the documentation there’s one thing that stands out, and that is the relatively high stun times for vGPU enabled VMs. Just as an example, here are a few potential stun times with various sized vGPU frame buffers:

  • 2GB – 16.5 seconds
  • 8GB – 61.3 seconds
  • 16GB – 100+ seconds (time out!)

This is all documented here for the various frame buffer sizes. Now there are a couple of things to know about this. First of all, the time mentioned was tested with 10GbE and the NVIDIA P40. This could be different for an RTX6000 or RTX8000 for instance. Secondly, they used a 10GbE NIC. If you use multi-NIC vMotion or for instance a 25GbE NIC than results may be different (times should be lower). But more importantly, the times mentioned assume the full frame buffer memory is consumed. If you have a 16GB frame buffer and only 2GB is consumed then, of course, the stun time would be lower than the above mentioned 100+ seconds.

Now, this doesn’t answer the question yet, why? Why on earth are these stun times this long? The vMotion process is described in this blog post by Niels in-depth, so I am not going to repeat it. It is also described in our Clustering Deep Dive book which you can download here for free. The key reason why with vMotion the “down time” (stun times) can be kept low is that vMotion uses a pre-copy process and tracks which memory pages are changed. In other words, when vMotion is initiated we copy memory pages to the destination host, and if a page has changed during that copy process we mark it as changed and copy it again. vMotion does this until the amount of memory that needs to be copied is extremely low and this would result in a seamless migration. Now here is the problem, it does this for VM memory. This isn’t possible for vGPUs unfortunately today.

Okay, so what does that mean? Well if you have a 16GB frame buffer and it is 100% consumed, the vMotion process will need to copy 16GB of frame buffer memory from the source to the destination host when the VM is stunned. Why when the VM is stunned? Well simply because that is the point in time where the frame buffer memory will not change! Hence the reason this could take a significant number of seconds unfortunately today. Definitely something to consider when planning on using vMotion on (multi) vGPU enabled VMs!