Hey everyone. We've been running ZPA for about a year now, and while the day-to-day access is solid, I’ve always been curious about what *actually* happens during a real outage. The documentation talks about high availability and failover, but I wanted to see it in action before we had to rely on it.
So last weekend, with full management approval and during our maintenance window, I ran a controlled test. The goal was simple: simulate a primary ZPA App Connector going offline and observe how quickly and seamlessly users were redirected. I picked a few key internal apps for this – our dev wiki, the HR portal, and a critical database frontend.
Here’s a rough outline of what we did and what we saw:
First, we had a small group of users from different locations actively using the apps. Then, from the ZPA admin portal, I manually disabled the primary App Connector servicing those apps. The failover wasn't instantaneous – we observed a brief hiccup (about 15-20 seconds) where connections hung before being re-established through the secondary Connector. The users didn't get booted out, just a short pause before things resumed normally.
The key takeaway for us was that the failover works, but it’s not completely transparent. That 15-second window is something we now need to account for in our internal SLAs and user training. It also highlighted the importance of our App Connector placement and health checks – if the secondary had been under-provisioned, that failover would have been much rougher.
Has anyone else done similar resilience testing? I’d be especially curious if you’ve found ways to minimize that transition delay or if you have a specific monitoring setup to watch failover events in real-time.
-ian
ian