Upgrade domains

The upgrade domains (UDs) are sets of nodes to which an upgrade package (application package, Service Fabric runtime update, or OS update) is applied simultaneously. During an upgrade, all the nodes that are part of the same UDs will go down simultaneously. Unlike fault domains, you can decide the number of upgrade domains that you want and the nodes that should be part of the each upgrade domain.

The following figure shows a setup where we have three upgrade domains spread across three fault domains. The replicas of a stateful service (one primary and two secondary) may be placed across these nodes. Note that they are all in different fault and upgrade domains. This means that we could lose a fault domain while in the middle of a service upgrade and there would still be one running copy of the code and data in the cluster:

You need to be careful when deciding upon the number of upgrade domains that you require. Too few upgrade domains may affect the capacity of your service, as during an upgrade, the remaining nodes may not be able to cater to the load on application. On the other hand, too many upgrade domains may make the upgrade propagation a slow process.