Upgrading vSphere without Runtime Downtime
Page last updated:
This topic describes how to upgrade the vSphere components that host your Pivotal Application Service (PAS) or VMware Tanzu Kubernetes Grid Integrated Edition (TKGI) deployment without service disruption.
At minimum, vSphere contains a vCenter Server and one or more ESXi hosts.
You cannot perform an in-place upgrade of vSphere without at least two ESXi hosts in your cluster.
If you do not meet this requirement, you have insufficient resources to evacuate an entire host. Because of this, you may experience runtime downtime during the upgrade.
To upgrade vSphere with only one ESXi host or without sufficient headroom capacity, you must reduce your runtime deployment size. In other words, you can either reduce the number of Diego Cells in your deployment or pause VMs to make more capacity available. These actions can result in downtime for PAS or TKGI.
If you are running a PAS or TKGI deployment following the base reference architecture described in vSphere Reference Architecture, your vSphere deployment should have these components:
- One vCenter Server
- Three ESXi hosts per cluster
- Three or more clusters
- One (or HA pair) NSX Edge appliances
Note: VMware recommends having at least three ESXi hosts in your cluster to maintain high availability (HA) during your upgrade.
These sections describe the procedure to upgrade the vSphere management layer underneath Ops Manager.
For more information about how to upgrade vCenter, see Overview of the vCenter Server Upgrade Process in the vSphere documentation.
After a successful vCenter upgrade, you must upgrade your ESXi hosts one at a time, starting with the first ESXi host.
To upgrade your ESXi hosts:
Verify that your ESXi hosts have sufficient resources and headroom to evacuate the VM workload of a single ESXi host to the two remaining hosts.
Note: If you have enabled vSphere HA on your ESXi host, each ESXi host should have sufficient headroom capacity, since vSphere HA reserves 66% of available memory.
Use vMotion to move all the VMs on the host you want to upgrade to the other ESXi hosts. vMotion places the VMs on the other hosts based on available capacity. For more information, see Migration with vMotion in the vSphere documentation.
Upgrade the evacuated ESXi host. For example, you may be upgrading from ESXi v6.0 to ESX v6.5. For more information, see Upgrading ESXi Hosts in the vSphere documentation.
Repeat the previous steps for each remaining host, one at a time. vSphere automatically rebalances all VMs back onto the upgraded hosts through DRS after you have upgraded all hosts.
If your PAS deployment is on a network behind an Edge Services Gateway (ESG) as recommended in vSphere Reference Architecture, upgrade each ESG only after completing the upgrade of vCenter and your ESXi hosts.
When you upgrade an ESG on VMware NSX, you upgrade the NSX Manager software. This upgrade can cause some slight downtime, the amount of which depends on the number of ESGs you are using:
If your deployment only has one ESG, you can expect a downtime of five minutes for network reconvergence.
If your ESGs are deployed in HA, upgrade the first ESG. Then upgrade the second ESG. This upgrade causes only 15-20 seconds of downtime.
For more information, see NSX Upgrade Guide in the VMware documentation.