Bugs are unavoidable, and software resilience is crucial, especially in modern systems composed of hundreds of services. Check out the first paper on Cross-System Interaction (CSI) failures from @tianyin_xu 's team. From yesterday's outage: one click to update, but it took forever to recover. The recommended recovery path required either "try to reboot affected machines over and over" or 20 different manual steps. CSI failures are hard to deal with because a minor error can have cascading effects on hundreds of other components. It's challenging to revert everything to a consistent previous state due to obscure cross-system dependencies. Bugs will happen -- more research on CSI testing, verification, and fault tolerance is needed!