An end user just asked me: “don’t you wish we still had our own Exchange server so we could fix everything instead of waiting for MS”?

This is what's kinda annoying me about the discussion from an SRE point of view for a SaaS offering. Downtime and Downtime being the same thing. Like, yeah, we do have outages. We've had a regional provider fuck up their networking backbones between their datacenters. Allegedly because two changes ran in parallel. In one place an electrician fucked up and put kv-sized power through networking equipment and at the redundancy someone fucked up configs. This took out the entire networking plane of that provider, and we were offline for 1-2 hours until the system came back online, self-healed and continued to function. Or in another case (before our database reworks), we ended up with some fucked up interaction between the HV, the underlying network interface and our VMs behavior, except the VM was a database master. This caused us to be effectively down for 2-3 hours because something in the network stack started dropping traffic on the floor after 20 - 30 minutes of runtime if the right (or wrong query) hit the database, as it seemed. This also rendered the VM unrebootable and it was pretty messy. The interaction with that vendor was also somewhat.. unenjoyable, so they are not hosting us anymore. But on the other hand, there is a lot of things customers don't see. One hoster had the AC in a redundant datacenter fail entirely. According to the temperature data, ambient temperature rose to "human adverse conditions" of 90 degrees celsius or more and they had firefighters on standby in case the entire thing flares through. At a software level, the system automatically removed thermally throttled nodes from the loadbalancing and failed some systems over out of that thermal hellscape. We mostly observed a response time increase at higher percentiles, because the system was figuring out what nodes to kill. Or we've had crash loops at an application level or even at a database level as well. Like, in one case, some misbehavior on production data caused tomcats to allocate about infinite memory, killing it rather quickly. Happened every 10 minutes or so to an instance. Was somewhat spicy, increased error rates slightly, but automated redeployments took care of that. In another case, some application misbehavior caused database servers to be overwhelmed by some insane abnormality for some reporting queries, eventually not reporting to queries again. Again, auto-failovers and healthchecks and loadbalancing took care of that. Increases in response time and error rates for sure, but even our most demanding customers were at most wondering why some sporadic requests had to be retried. This is why especially both the smallest, and the largest customers are very happy with the SaaS hosting. Our smallest customers need a system for 10 users and get one sized for hundreds of thousands of concurrent full time users, and our largest customers don't have to invest oodles and oodles of time to learn to run our software.