I recently decided to switch back to using USB Media for booting my VMware ESXi hosts. My main thought was that I want to use the local HHD for either some kind of VSA testing (such as StorMagic’s SvSAN) or else when I have the budget buy in SSDs to make a physical VSAN cluster, rather being dependent and reliant on a virtual nested VSAN setup. I went out and got some 32GB USB media to make sure they would be big enough to create the scratch partition.

I’m often wiping my VMware ESXi hosts to try out new builds – all to roll-backwards to previous builds for testing purposes. For that I used the UDA together with some kickstart scripts to do the bulk of the customisation. So times I just lay down a clean build with no customisation and use PowersHell scripting to build up the environment to the level I want it. I have that in a modular way that allows me to lay down, some but not all, of the configuration depending on my development needs.

Anyway, to script an install to USB media you can use this in the kickstart:

install --firstdisk=usb --novmfsondisk

The –firstdisk flag actually can take a series of arguments separated with a comma. It is possible to specify an order of particular disks type (not just the first USB, HHD/SSD or SAN based LUN) the installer searches for. And you can use to this set your own orders for how disks are discovered. For instance –firstdisk=usb,remote,local would first try to install to USB, then a LUN on a SAN, before lastly trying the local disk.

Also what’s supported is specifying the model of the disk like so –firstdisk=ST3120814A. I used this in the past with hyper-converged appliances that I was doing ground-zero resets on – and just so happened the bootdisk I was using had a unique model number. Of course this isn’t very helpful when a server contains disks that all have the same model number…

The other option is to use instead of install –firstdisk, but use install –disk=mpx.vmhba1:C0:T0:L0. This allows you to indicate the disk by which adapter its configured for, and the controller/target used (note these values are often 0, just the L number value that changes. For many years we have referred to this as the vmhba syntax or “Runtime Name”. In the past its been hard to trust 100% these values, but I think they are more reliable nowadays, as the way the VMware ESXi host boots the vmkernel is VERY different now.

Probably the easiest way to find the MODEL name/number and vmhba syntax is using the esxcli commands like so:

esxcli storage core path list

Here’s a cut and paste of the USB output – 

usb.vmhba32-usb.0:0-t10.SanDisk00Ultra00000000000000000000004C530001270222101242 UID: usb.vmhba32-usb.0:0-t10.SanDisk00Ultra00000000000000000000004C530001270222101242 

Runtime Name: vmhba32:C0:T0:L0 

Device: t10.SanDisk00Ultra00000000000000000000004C530001270222101242 

Device Display Name: Local USB Direct-Access (t10.SanDisk00Ultra00000000000000000000004C530001270222101242) 

Adapter: vmhba32 Channel: 0 Target: 0 LUN: 0 Plugin: 

NMP State: active Transport: usb 

Adapter Identifier: usb.vmhba32 

Target Identifier: usb.0:0 

Adapter Transport Details: Unavailable or path is unclaimed 

Target Transport Details: Unavailable or path is unclaimed Maximum IO Size: 32768

I imagined a system with multiple UBS media (!??!) would report C0:T0:L1 or else C0:T1:L1.

Oddly enough I found a second USB device added to my ESXi host doesn’t appear – not even when doing a manual install with ISO attached the HP ILO. I’m not sure why this is – but it may (or may not) be significant which USB slot the device gets inserted too, or that there are limits around the way the VMkernel enumerates USB device – perhaps only enumerating the first USB memory stick found?  This isn’t terribly important – but I will ask around my contacts and see what I can dig up. It’s a bit obscure, but I’m curious like that.

I’m running VSAN in a nested configuration, and I generally shutdown my homelab in the evening each day. I grew tired of doing the manual process of maintenance mode for these nested nodes before shutting them down, and then shutting down the host that they run on. I did a bit of googling for the PowersHell that does that – and was see quite complicated scripts. I’m pleased I went back to the documentation for PowerCLI which is the primary source for the cmdlets:

https://code.vmware.com/doc/preview?id=6702#/doc/Set-VMHost.html

I discovered that the Set-VMhost cmdlet supports a -VsanDataMigrationMode switch with three different options for:

• Full
• EnsureAccessibility
• NoDataMigration

1..4 | Foreach {
 $Num = "{0:00}" -f $_
 Set-VMHost -VMHost esx"$Num"nj.corp.local -State "Maintenance" -VsanDataMigrationMode NoDataMigration 
 }

1..4 | Foreach {
 $Num = "{0:00}" -f $_
 Stop-VMHost esx"$Num"nj.corp.local -Confirm:$False
 }

Nesting VMware ESXi has become easier and easier as the years roll by. In case you don’t know “nesting” is the term used for running VMware ESXi inside a VM for development purposes. It’s the basis of VMware’s popular “Hands-on-Lab”, as well as homelabs – and in recent years “nesting” has gone to the Cloud with service like Ravello on the Oracle Infrastructure Cloud (OIC).

In years gone by there were many settings needed at the Physical and Virtual layer – that required hand-edits to configuration files to make this work. With the advent of vSphere 6.5 U1 a lot of that disappeared. HOWEVER, there’s still plenty of work that needs to be done at the physical and virtual layer to allow for networking to work properly. As well as successfully passomh some of the “health checks” around technologies like VMware VSAN that has specific networking requirements for your web-client to light up with little green ticks of happiness.

Here’s a brief check list:

1. Physical Switch needs:
   • MTU of 9000,
   • with VLAN Tagging enabled as necessary
2. Physical ESXi host vSwitch needs
   • MTU of 9000
   • vSwitch security policy of Accept, Accept, Accept
   • Portgroups used by the virtual “nested ESX” enabled for 4096 to allow for VLAN tagging in the nested layer to pass-thru to the physical switch
3. Virtual Nested ESXi vSwitch needs
   • MTU of 9000
   • vSwitch security policy of Accept, Accept, Accept
   • Portgroups for VMs and VMkernel enabled for VLAN Tagging as necessary

Standard Switches work well with nesting, and have the added benefit of not being tied to a vCenter to the host. This makes blowing away the nested layer after your lab period ends a breeze. It’s slower and more awkward “clean-up” process if your using DvSwitches. To use the special MacLearn VIBs that improve network performance DvSwitches are needed at the physical layer – this isn’t an option if you main physical ESXi host is stand-alone and not managed vCenter.

1. Physical Switch Needs:

In my case I have a HP ProCurve 1810G – 24 G switch – its not a bad unit, whisper quiet and simple to configure:

I have simple VLAN configuration (mainly used to demonstrate/explain the VLAN Tagging concept) but I also VLAN off my VMotion traffic.

Physical ESXi host vSwitch needs

I set my MTU and Security policy on the properties of vSwitch0, which means all the portgroups inherit those settings. Its simple quick and easy.

To do that in the context of the VMware ESXi VMKernel you could use the ESXCLI command – and these commands could be part of a kickstart install script:

## - Enabling Jumbo Frames on vSwitch0 to pass VSAN "Configuration Assistant" tests.
esxcli network vswitch standard set -m 9000 -v vSwitch0

## - Lower security on vSwitch0 to allow traffic to flow in a nested environment. 
esxcli network vswitch standard policy security set -v vSwitch0 -f=true -m=true -p=true

The VLAN configuration can be set when the portgroup is being created as is the case with VMotion being on VLAN10 like so:

esxcfg-vswitch -A "VMotion" vSwitch0
esxcfg-vswitch -v 10 -p "VMotion" vSwitch0
esxcfg-vmknic -a "VMotion" -i [VMOT_IP] -n 255.255.0.0 -p "VMotion" vSwitch0
esxcli network ip interface tag add -i vmk1 -t VMotion

Note: In this case [VMOT_IP] is variable used as part of the UDA appliance.

For an existing portgroup for example the “VM Network” the VLAN value can be set to be one that passes thru the VLAN Tagging from the nested layer to the physical layer:

esxcli network vswitch standard portgroup set -p "VM Network" --vlan-id 4095

Alternatively, if your Physical ESXi hosts are managed by vCenter – you could do this with a PowersHell for each loop like so:

1..3 | Foreach {
 $Num = "{0:00}" -f $_
 $vswitch0 = Get-VirtualSwitch -VMHost esx"$Num"nyc.corp.local -Name vSwitch0
 Set-VirtualSwitch -VirtualSwitch $vswitch0 -MTU 9000 -Confirm:$false
 }

1..3 | Foreach {
 $Num = "{0:00}" -f $_
 $vswitch0 = Get-VirtualSwitch-VMHost esx"$Num"nyc.corp.local -Name vSwitch0
 Get-VirtualSwitch -Name $vSwitch0 | Get-SecurityPolicy | Set-SecurityPolicy -MacChanges $true
 Get-VirtualSwitch -Name $vSwitch0 | Get-SecurityPolicy | Set-SecurityPolicy -ForgedTransmitsInherited $true
 Get-VirtualSwitch -Name $vSwitch0 | Get-SecurityPolicy | Set-SecurityPolicy -AllowPromiscuous $true
 }

The VLAN configuration for an existing portgroup can be updated using the following:

1..3 | Foreach {
 $Num = "{0:00}" -f $_
 $vswitch0 = Get-VirtualSwitch -VMHost esx"$Num"nyc.corp.local -Name vSwitch0
 $VMNetworkPG = Get-VirtualPortGroup -VirtualSwitch $vSwitch0 -Name "VM Network"
 Set-VirtualPortGroup -VirtualPortGroup $VMNetworkPG -VLanId 4095
 }

Virtual Nested ESXi vSwitch needs

The setting on vSwitch0 on the Nested ESXi host mirror that of the physical… so everything is aligned from vESX>pESX>pSwitch everything is MTU 9000 with weakened security. The VLANs I have from 101-104 could be added at the physical and virtual level with the esxcli command like so:

esxcli network vswitch standard portgroup add --portgroup-name=VLAN101 --vswitch-name=vSwitch0

esxcli network vswitch standard portgroup set -p VLAN101 --vlan-id 101

or if you were wanting to do this via PowersHell, you could use the following method. This approach differs from my previous examples – it merely “gets” every ESXi host in a vCenter, and creates VLAN101, 102, 103, and VLAN 104.

101..104 | Foreach { $Num = $_ 

(Get-VMHost | sort-object name) | foreach 

{New-VirtualPortGroup -VirtualSwitch (Get-VirtualSwitch -Name vSwitch0 -VMHost $_) -Name VLAN$num -VLanId $num } 

}

A very common use of nested vSphere is to create a virtual/nested vSphere VSAN cluster – this is because not everyone can afford the servers and storage to build out VSAN on a physical level (3 hosts with 1xSDD, 1xHDD). For sometime its been possible to virtualise ESXi, as well as also marking specific VMDKs as either being HHD or SSD. Inside the nested ESXi environment you will need a VMKernel portgroup enabled for VSAN. A simple thing to do would be to enable either the Management Network or the VMotion network for dual usage. Alternatively, you could setup a net-new VMKernel portgroup who’s sole and only purpose is VSAN communications. That’s what I do (even if the traffic goes over the EXACT same nics – as it makes it clear what the usage of each portgroup/vmkernel port is.

You can do this with CLI using the following commands like so:

esxcfg-vswitch -A "VSAN" vSwitch0
esxcfg-vswitch -v 0 -p "VSAN" vSwitch0
esxcfg-vmknic -a "VSAN" -i [VSAN_IP] -n 255.255.255.0 -m 9000 -p "VSAN" vSwitch0
esxcli vsan network ipv4 add -i vmk2

Note: In my case [VSAN_IP] is a variable used in my kickstart sub-template as part of the UDA. The main thing is to know the number of vmk ports – vmk0 is always the default management port – in my script the next vmk port created is for VMotion, and that therefore make the VSAN port be vmk2.

or with PowersHell:

1..3 | Foreach {
 $Num = "{0:00}" -f $_
 $vswitch0 = Get-VirtualSwitch -VMHost esx"$Num"nj.corp.local -Name vSwitch0
 New-VMHostNetworkAdapter -VMHost esx"$Num"nyc.corp.local -VirtualSwitch $vswitch0 -PortGroup VSAN -IP 10.20.33.1"$Num" -SubnetMask 255.255.255.0 -VsanTrafficEnabled $true
 }

https://blogs.vmware.com/vsphere/2018/05/vsphere-6-5-update-2-now-available.html

DISCLAIMER: Firstly, and most importantly. This issue likely to effect a tiny proportion of customers. Despite what many think as soon as new version of vSphere hits the streets for most large organisations it can take anything between 12-24 months to actually complete an upgrade from one flavour of vSphere to another. Therefore by the time a customer who has rolled out vSphere 6.5 U2, the chances are that there WILL be an upgrade path to vSphere 6.7 probably in the shape of U1 or U2 release. The important thing to remember is if your intention is in the short-term to upgrade to vSphere 6.7 for whatever reason – is to be aware of the title of this blogpost. You may as well skip U2 and head straight to vSphere 6.7 if that’s the case…

UPDATE (10th May, 2018): It looks like the VAMI interface on the VCSA is set to deploy vSphere 6.5 U2. It’s set as just a “bugfix” and that its severity is set to “critical”. I’m research what the situation is with update manager.

WHY DOES THIS HAPPEN: This has happened on more than one occasion in recent memory. I don’t personally think that precedent can be used to justify this. A much simpler reason exists. The reality is VMware now has a very complicated, and richly interdependent and tightly-coupled series of software and services. Attempts have been made in the past to build a “Train Release” by which the core platform like vCenter/ESX are the “head” or “engine” of the train, and the related products that sit on top of it – are couple behind it on a release schedule that if correctly managed should arrive in the customer station at roughly the same time. Occasionally, a carriage gets de-coupled and is left in the sidings somewhere outside Colchester. For customers for whom that software package is critical it can and will delay and upgrade processes until the last part of the train gets into the station. This can be difficult where one SR calls for upgrade to fix one problem, only to “break” a piece of software elsewhere.

This kind of sequencing of software releases is very, very difficult to do in a large multi-national software company where software is shrinked-wrapped and installed on-prem. It’s actually a compelling argument for buying software like vSphere-as-a-service, and heading off into the Public Cloud. You cease to have worry about this kind of poop.

HAS THIS BEEN COMMUNICATED EFFECTIVELY. I’m sorry to say this but no. Sadly, an almost like Apple like cloak of secrecy still envelopes VMware. There are good reasons for this in some cases, but other situations like this – it actively and effectively works against good communication. As ordinary mortal I am supposed to read all the blogs, read all the release notes and listen to all the podcasts. I would say I’m pretty well connected to the VMWorld – but I didn’t know this was happening. Heck, I didn’t know that vSphere 6.5 U2 was on its way – or know there wasn’t upgrade path.

Apparently, to some that’s my fault, and my problem – and it isn’t the responsibility of VMware to shout from the rooftops as if Four Horses of the Apocalypse were on their way. Before I rant on and loose the plot a little – I will point you back to this disclaimer. Look. I totally get it, and totally understand why and how this happens. And in many, many, many ways the vSphere 6.5 U2 release is a good, good,good thing because it quickly deals with the issue associated with “WHY DOES THIS HAPPEN”. The train becoming de-railed from the railway lines.

For those who are outside of the vendorland much of this just seems bug-eyed weird. Although “shit-sandwich” has become such a defacto standard in our industry many seem quiet happy to suck it up. Personally, I like to set my standards higher than that. Case in point. The download page for vSphere 6.5 U2 has one of those little yellow “read the KB” stickers – which incidentally disappears when you click “Go To Downloads”.

The downloads page of vSphere 6.7 (at the time of writing) has a tiny notification at all.

People will say “buyer beware” and you should read the Release Notes. Honestly, who has time for that – Release Notes are like EULAs. Practically no-one reads them (I’m unable to quantify that statement I know!).

Of course, if you fail to read all the blogs, the release notes and these stickers – then its “your own fault chum” according to some. Personally, I think vendors have to realise that in todays busy world with lots of competing channels searching for our eyeball time – something a little more noticeable is required.

With that. Read my disclaimer again…

Executive Summary: There are lot of methods. If at first you don’t succeed give it another go. 2nd Time lucky and all that.

I recently lost my Windows10 Jumpbox due to a power failure. It disappeared from my SAN array. I didn’t have a backup (which was stupid of me) and Windows wasn’t activated. In the meantime I realise I had an activated version of Windows10 running in VMware Fusion. So I thought heck might as well import to vSphere, and avoid re-installing and that annoying watermark in the right-hand corner of my RDP desktop.

Ironically, after completing the move – because Windows10 was sitting on a different virtualization hardware platform – it de-activated itself – and wouldn’t reactivate. So it was all a waste of time really – I could have just deployed a fresh copy of Windows 10.

Anyway, I’ve learn my lesson the hardware – always backup your Jumpbox and make sure any work in progress is held elsewhere such as Dropbox, Google Drive or OneDrive. That way if your jumpbox is toast you don’t loose that PowersHell script you have been perfecting all week.

METHOD1: Connect to Server from VMware Fusion

This initially failed when the target was vSphere6.5 U1, but worked with vSphere 6.7. Hard to tell if the vSphere uplift was the decisive difference, or whether it was a case of just try, try and try again.

VMware Fusion has a “Connect to Server” option

Once connected and authenticated to the vCenter or ESXi host – you can drag and drop a VMware Fusion VM like so:

METHOD2: Export/Import OVA

Another method is to export the VM from Fusion into a the OVF/OVA format. I prefer OVA as its pre-compressed and gives you a single file to deal with – as OVF gives you a text-based “descriptor” file and whole bunch of VMDK files.

Select the VM in Fusion, and in the File menu, choose Export to OVF

Once exported you can login to either VMware vCenter or VMware ESX to import there. Personally I’ve found the import/export process in vCenter/ESX a bit 50:50…. Again, this failed in vSphere 6.5 U1, but was successful in vSphere 6.7.

Select a host or cluster in vCenter, and choose Deploy OVF Template…

Watch out for the format of the disk – as it does not default to “Thin Provisioning”.

METHOD3: VMware Convertor

I tried this – a number of times, and it did not work. VM would not boot. This is kind of odd because I’ve had more success with this method for importing physicals into VMware Fusion, than I have had with other methods of “importing” OSes into VMware Fusion….

One of the most common questions I was asked when I was an instructor was “what works best, clean install or an upgrade”. Without fail it was the kind of question everyone already had their own answer to. So I couldn’t add much to the debate, although I knew where I stood.

The upgrade process for vSphere with its myriad of service dependences is getting longer, and longer – and thornier and thornier. Even so despite an in-place upgrade taking a long time – its is still the best route considering the impact of clean installation. About the ONLY thing I’d would be tempted to do a clean install of is VMware ESXi, but that comes now with its own set of complications not least if the customer uses VMware’s Distributed Switches.

So this blog is NOT an agonising blow-by-blow of upgrading that would rival your teeth being extracted. It’s high-level view of the upgrade (just of vCenter and ESX and nowt else) and what that feels like.

VMware vCenter/PSC Upgrade on vCSA

This went preachy-smooth, and despite having the complicated “external” PSC and vCenter model to allow for “Enhanced Linked Mode” the upgrade process was lengthy, but easy and worked perfectly first time. It’s not so much an “upgrade” process, more of duplication/mirror process – as the the upgrade wizard spawns a duplicate vCenter and PSC, and then sets about copying the data from one to the other. At the end their IPs are inverted and the old vCenter/PSC is shutdown. The only danger here is if you do reboots and bring up the old vCenter/PSC by mistake. I did that once which was confusing. The nice thing about this process is that its non-intrusive to the Old vCenter/PSC. So if something goes pear-shaped (as opposed to peachy-smooth) you have a failback position

Verdict: Upgrade, fill your boots!

VMware ESXi Upgrade on HP ML350e Gen8

For me this was a less than good experience. I could only get just one of my three hosts to upgrade. Some issue with Update Manager in the end. By the end of it I realise that kickstart script install from my UDA was the only way to go – so the dirty upgrade fails, and a clean install won the day. Perhaps my upgrade issues were caused by much more intractable problem – my servers aren’t even on the HCL anymore!

I have 3x HP ML350e Towers – and they fell off the HCL around the time of vSphere 6.0 U3

Despite this – there was release from HPE that was dubbed pre-Gen10 of vSphere 6.6 U1 – this was the build called “VMware-ESXi-6.5.0-Update1-6765664-HPE-650.U1.9.6.5.1-Nov2017.iso”, if you accidentally installed the build that came after it called “VMware-ESXi-6.5.0-Update1-7388607-HPE-650.U1.10.2.0.23-Feb2018.iso” in Feb of this year – you would be installing a code based designed for a Gen9/10 server – and that would make the fans go bizzy and cause alarms and alerts in vCenter…

Things don’t tend to go back on the HCL once they have left it – and I think I was lucky that there was a pre-Gen9 custom image to work with my Gen8 systems. I don’t expect to see this at all with vSphere 6.7.

Both the vSphere 6.7 generic and 6.7 from HPE successful install to my HP ML350e Gen8s. But its not without complication. The fans go faster and so the lab makes more noise – and fan alarm is trigger is lit in vSphere 6.7.

Put simply before the vSphere 6.7 install. The fans were going much slower, and vCenter had no alarms.

Merely acknowledging or “Reset to Green” only causes this alarm to temporarily to be dismissed before it comes back again like a bad smell or penny. The only way to turn off this “false positive” is by disabling the alarm itself. Navigate to:

>> Global Inventory Lists  >> vCenters >>Select vCenter FQDN >>Configure  >>More >>Alarm Definitions

Search for the alarm called “Host hardware fan status“. Select and click the Disable button

One the more pleasing aspects of vSphere 6.7 is the extension of the HTML5 client to cover more functionality. 95% feature parity with the Ye Old Flash Web-Client of teeth-extraction fame. Make no bones about it the HTML5 client above is the bees-knees – mainly because in comparison to its slow-coach predecessor it is blisteringly quick. Indeed, switching back to the previous version of the client that replaced the previous version of the client, that worked prefect fine – is an unpleasant reminder of how dreadful it was… However, there is something jivey about the way this simultaneous release of 3 different clients at one stage happened. This isn’t “Agile” devlopment its “Fragile” development at its worst – and sends a confused jagged and jarring message to customers. I would have preferred to have waited until 7.0 and full-release of a brand-new shinny clients. Rather than the HTML5 client being dribbled out in dribs and drabs.

I doubt very much if any customer purchases a product on the strength of the client front-end. I mean if they did with VMware they’d get a lot of bang for the buck when 3 of them were available. 😉 After all most customers are either automating much of the tasks it offers with PowersHell, or with some other overlay that abstracts away the complexity of full fat Web-Client.

With that said there is a tier of customers for whom vSphere is just 10-20% of their daily admin tasks (if that) and for whom an intuitive and easy to use interface is a must. Because unlike hard-core VMware-Fanatics like myself, they only log into it if the have need too. There’s some aspects of the new HTML5 client that irritate and some aspects that are broken.

I’ve yet to see either the Flash Web-Client or the HTML5 Web-Client successfully handle OVF/OVA exports and imports – with me having to resort to the use of OVFTOOL to get a process that is reliable and dependable.