Signed-off-by: Michał Maślanka <[email protected]>
(cherry picked from commit a9922a5)

Signed-off-by: Michał Maślanka <[email protected]>
(cherry picked from commit 00f27d9)

In Raft it is a common operation to access ids of all nodes from the
group configuration. Previously every time caller called `all_node_ids`
it built the vector of unique vnodes. Caching the value makes is much
more convenient to access/iterate over all replicas as it doesn't
require any copy or allocation.

Signed-off-by: Michał Maślanka <[email protected]>
(cherry picked from commit 46688af)

Refactored voter priority tracking by moving it from the raft protocol
implementation. The voter priority may be handled outside of
`consensus`.

Signed-off-by: Michał Maślanka <[email protected]>
(cherry picked from commit 3189c03)

Change the readiness for election to be handled inside of raft priority
tracker rather than leaking the priority internals outside of raft
protocol.

Signed-off-by: Michał Maślanka <[email protected]>
(cherry picked from commit 319843c)

Signed-off-by: Michał Maślanka <[email protected]>
(cherry picked from commit ca94cfa)

The maintenance mode should never completely prevent the leader of the
raft group to be elected. Even if the only possible leader candidate is
in maintenance mode. Change the leadership blocking logic to use the min
voter priority. This way the replica with blocked leadership can still
become a leader if it is the only eligible candidate.

Signed-off-by: Michał Maślanka <[email protected]>
(cherry picked from commit cb0e2f5)