Fix a bug where the leader is never elected even if the majority of the members are alive.

[sile]

Proposed Changes

This PR addresses the issue reported in #439.

To summarize #439, if there is a candidate member and a pre_vote member where the pre_vote member has a higher log index than the candidate member, neither of them can ever be elected as the leader.
(This holds true even if there are additional N / 2 - 1 or fewer followers without election timers, where N is the cluster size.)

This PR adds a branch to ra_server:handle_candidate(#pre_vote_rpc{}, ...) to handle cases where the pre_vote member has a higher log index. By the new branch, when such a message is received, the candidate transitions to the follower state.

I think this is somewhat ad-hoc. However, since I don't know much about the ra code base (especially regarding the role of the pre_vote state), I made a patch to minimize the impact range.
Feel free to suggest any better alternative approaches.

Closes #439.

FYI

By applying the patch for reproduction from issue #439 to this PR branch, the execution result became as follows:

(foo@localhost)1> repro:run().
# create cluster
* [repro_a] init
* [repro_c] init
* [repro_b] init
* [repro_a] state_enter: recover
* [repro_c] state_enter: recover
* [repro_b] state_enter: recover
* [repro_a] state_enter: recovered
* [repro_c] state_enter: recovered
* [repro_b] state_enter: recovered
* [repro_a] state_enter: follower
* [repro_c] state_enter: follower
* [repro_b] state_enter: follower
* [repro_c] state_enter: pre_vote
* [repro_c] state_enter: candidate
* [repro_c] state_enter: leader
# Please wait 5 seconds...
# trigger election
ok
* [repro_a] state_enter: pre_vote
* [repro_a] state_enter: candidate
* [repro_c] state_enter: follower
* [repro_c] state_enter: pre_vote
* [repro_a] state_enter: follower
* [repro_c] state_enter: candidate
* [repro_c] state_enter: leader  # new leader is elected
* [repro_a] state_enter: await_condition
* [repro_a] state_enter: follower

Types of Changes

What types of changes does your code introduce to this project?
Put an x in the boxes that apply

• Bug fix (non-breaking change which fixes issue There is a possibility that the leader is never elected even if the majority of the members are alive #439)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation (correction or otherwise)
• Cosmetics (whitespace, appearance)

Checklist

Put an x in the boxes that apply. You can also fill these out after creating
the PR. If you're unsure about any of them, don't hesitate to ask on the
mailing list. We're here to help! This is simply a reminder of what we are
going to look for before merging your code.

• I have read the CONTRIBUTING.md document
• I have signed the CA (see https://cla.pivotal.io/sign/rabbitmq)
• All tests pass locally with my changes
• I have added tests that prove my fix is effective or that my feature works
• I have added necessary documentation (if appropriate)
• Any dependent changes have been merged and published in related repositories

Further Comments

If this is a relatively large or complex change, kick off the discussion by
explaining why you chose the solution you did and what alternatives you
considered, etc.

[michaelklishin]

@sile thank you for your ongoing contributions! We will get to this PR in the next couple of weeks.

[kjnilsson]

Higher states are only really exited when encountering a higher term which this goes against. In this case I think it should be sufficient for the candidate to simply reply to the pre-vote request (with a success) as this would allow the pre-voter to potentially proceed to candidate state itself and eventually win the election.

[sile]

@kjnilsson That sounds reasonable. Thank you for your comment!

I have incorporated the change in commit e6bbd1a. Here are the current results of the repro function:

> repro:run().
# create cluster
* [repro_b] init
* [repro_c] init
* [repro_a] init
* [repro_b] state_enter: recover
* [repro_c] state_enter: recover
* [repro_a] state_enter: recover
* [repro_b] state_enter: recovered
* [repro_c] state_enter: recovered
* [repro_a] state_enter: recovered
* [repro_b] state_enter: follower
* [repro_c] state_enter: follower
* [repro_a] state_enter: follower
* [repro_c] state_enter: pre_vote
* [repro_c] state_enter: candidate
* [repro_c] state_enter: leader
# Please wait 5 seconds...
# trigger election
ok
* [repro_a] state_enter: pre_vote
* [repro_a] state_enter: candidate
* [repro_c] state_enter: follower
* [repro_c] state_enter: pre_vote
* [repro_c] state_enter: candidate
* [repro_a] state_enter: follower  # Unlike before the commit e6bbd1a, repro_a becomes follower after repro_c becomes candidate.
* [repro_c] state_enter: leader
* [repro_a] state_enter: await_condition
* [repro_a] state_enter: follower

[sile]

Thank you for reviewing and merging this PR!