When we are hit by a network attack (which happens daily), our automation launches new instances to meet demand. On some occasions (a few times this year) there may be some degraded performance for between 1 and 5 minutes while scaling launches.
Earlier in the year we had an outage with our upstream provider which was incredibly frustrating.
No one likes degraded performance or downtime, let alone our team who are woken by automated alarms at 2am to come and deal with it, but it's the nature of the internet. When you compile a list it may look significant, but it's still just tiny blips in an otherwise robust service. We really do our best to minimise disruption and we'll continue to revise our alarms, scaling and database instances.