1. Why does raft use randomized timout values?

  2. “GFS and HFDS use a seprate replicated stat machine to manage leader election and store configuration”, what does the author mean? Give concrete examples.

  3. What is the max # of servers that can fail while raft is still avaiable? what is the min # of servers that can fail while raft becomes unavailable?

  4. Descript the seqence of actions when a CANDIDATE handle an incoming AppendEntries call?

  5. How can a candidate discover current leader or a new term? Why it becomes a follower upon such discovery?

  6. During election, it is possible that no one is elected, how can that happen? How does raft deal with that?

  7. consider the cases of leader/follower crashes, where can these scenarios happen? 
         a. follower may have missing entries

     b. follower may have extra uncommited entries

     c. follower may have both missing entries and extra entries

  8. When and how will raft fix the inconsitences in Figure 7)?

  9. Why we need to ensure that any elected leader for an given has ALL of commited entries from the previous term? Why this property is satisfied in raft?

  10. Describe the scenario in Figure 8 
        a.replicated on majority of servers but not commited entry

    b.such entry in a) got overwritten

  11. What if broadCastTime « electionTimeout is violated? what if electionTimeout !« mean time between failures (MTBF)?

  12. Describe the disjoint majorities problem, how they can be triggered during membership changes, when each server just switch directly

  13. What happened when cluster leader itself is not included in the new membership?

  14. Give a scenario where removed cluster members can trick the new leaders into a follower, if without the remedy done by the raft.

  15. What kind of metadata should snapshot include?