Leaders with no log before snapshot get stuck in a loop when replicating the snapshot

[oleschoenburg]

Describe the bug

When a leader's first log entry is the next entry after the snapshot, the leader will get stuck in a snapshot replication loop, sending the same snapshot over and over again. The follower won't receive regular appends and will not catch up to the leader.

To Reproduce

Given this scenario:
Leader has a snapshot with index X, and a log that starts at index X+1.

After finishing snapshot replication to a member, the members position is reset to the snapshot position X + 1. That causes a seek for the previous position, X, followed by a read of the next entry, X+1. The position X does not exist (log starts with X+1) which means that the member context has no current entry and thus currentIndex = 0. This means that shouldReplicateSnapshot will decide to replicate the snapshot again, closing the loop.

Expected behavior

After successful snapshot replication, regular event replication should continue and the same snapshot shouldn't be sent over and over again.

This is a valid scenario that has worked previously but we removed one condition from shouldReplicateSnapshot that prevented repeated snapshot replication: 6ffc0cb

if (member.getSnapshotIndex() >= persistedSnapshot.getIndex()) {
  return false;
}

Support Case: SUPPORT-13931