manually deallocation = predictable performance = lower memory overhead = more bugs

automatically deallocation = free the programmer from the burden = performance variations/overhead

push CDN: requires engineering effort to propagate updates to the CDN caches; the first request is fast bc CDN server doesn't need to travel to the original server. more control over cache timing

solutions: use apache flink for distributed state management; use monitoring and observability tools like prometheus, grafana and confluent to track down and resolve issues; use a kafka "checkpoints" + exactly-once semantics to prevent data loss and duplication of events

I also discovered that I’m a passive learner (I think we all are since our brains tend to be lazy). To counter this, I need to consciously stay active by explaining what I’ve just learned. It’s an effective way to keep my brain engaged.