The banner, the pop-up, and the tag management (firing any third-party tag that requires consent, the IAB CMP API, etc.) all happen without any server request (consent is cached locally in a cookie) so that your user experience is not impacted.
💻 Back-end servers
We do send HTTP requests when consent is collected or for counting page views (we sample the number of page views by counting between 10% and 20% of them). These requests are async and always happen after all the front-end operations so that, even if they were to fail, that would not impact the user. That also means that if our servers were to be unavailable, the experience of your users would not be impacted and consent is stored locally and sent to our servers later on.
Our back-end servers are monitored 24/7 with alerting through New Relic and AWS CloudWatch to make sure that we get notified if there is any issue. We auto-scale them to ensure that we add (or remove) servers to our stack as needed throughout the day or when we sign up new clients.
We usually do not monitor clients individually (we might do exceptionally from time to time) as we focus on making sure that all our infrastructure is elastic, can scale without any human intervention, and that our servers being down do not impact the critical path of collecting/managing consent for end users. We know we can handle 10 times more traffic than we do today without any issue (and that growth has happened a few times over the last few months).