Hey all! I was going through the "Real-Time data flow" documented here. I had two questions: 1. If there are n replicas of a segment, is it the correct understanding that all n servers consume the events from Kafka and only 1 of them "wins" to commit to deep storage? 2. In a rare event that "all" n servers for a segment go down before the segment could get completed (and committed to deep storage from memory), does this in-memory data get lost forever or do the new servers manage to consume these "lost" events again?

1. Yes 2. No data loss, anytime a server restarts it will start consuming from the previous committed offset.

Thanks!