Azure Stack puts cloud services in a location you chose, usually to satisfy a latency, sovereignty, or connectivity requirement. It also hands your team responsibilities the public cloud had been quietly absorbing: physical infrastructure, local capacity, and a network link that can fail independently of everything else.

That link is the failure mode most likely to be untested. Applications running on Stack are typically written as though the connection to Azure is always available, because during development it was. What happens when it degrades, whether workloads keep serving locally or block on a call that will never complete, is usually unknown.

Install the Gremlin Agent on your Stack VMs and test it deliberately. Blackhole experiments sever the connection back to Azure so you can establish whether local autonomy is real. Latency experiments simulate the degraded link, which is more common than a clean disconnection and harder for applications to handle. Shutdown experiments cover local VM failure, where there's no regional redundancy to fall back on.

If the deployment exists to keep operating when connectivity is unreliable, that capability is worth verifying rather than assuming, particularly at a site with nobody on hand to intervene.