[Tarantool-patches] [PATCH v2 4/5] raft: introduce split vote detection

[Serge Petrenko]

20.01.2022 03:43, Vladislav Shpilevoy пишет:
> Split vote is a situation when during election nobody can win and
> will not win in this term for sure. Not a single node could get
> enough votes. For example, with 4 nodes one could get 2 votes and
> other also 2 votes. Nobody will get quorum 3 in this term.
>
> The patch makes raft able to notice that situation and speed up
> the term bump. It is not bumped immediately though, because nodes
> might do that simultaneously and will get the split vote again
> after voting for self. There is a random delay. But it is just max
> 10% of election timeout, so it should speed up split vote
> resolution on 90% at least.
>
> Part of #5285
> ---
>   src/lib/raft/raft.c         | 129 +++++++++++++++-
>   src/lib/raft/raft.h         |   6 +
>   test/unit/raft.c            | 301 +++++++++++++++++++++++++++++++++++-
>   test/unit/raft.result       |  64 +++++++-
>   test/unit/raft_test_utils.c |  12 ++
>   test/unit/raft_test_utils.h |   5 +
>   6 files changed, 512 insertions(+), 5 deletions(-)
>

Thanks for the fixes! Please find 2 comments below.

> +static void
> +raft_check_split_vote(struct raft *raft)
> +{
> +	/* When leader is known, there is no election. Thus no vote to split. */
> +	if (raft->leader != 0)
> +		return;
> +	/* Not a candidate = can't trigger term bump anyway. */
> +	if (!raft->is_candidate)
> +		return;
> +	/*
> +	 * WAL write in progress means the state is changing. All is rechecked
> +	 * when it is done.
> +	 */
> +	if (raft->is_write_in_progress)
> +		return;
> +	if (!raft_has_split_vote(raft))
> +		return;
> +	assert(raft_ev_is_active(&raft->timer));
> +	/*
> +	 * Could be already detected before. The timeout would be updated by now
> +	 * then.
> +	 */
> +	if (raft->timer.repeat < raft->election_timeout)
> +		return;

I don't think you should decrease timer.repeat.
This 'vote speedup' is for a single term only.

Besides the check below about delay >= remaining is enough
to test if split vote detection was already triggered.

> +
> +	assert(raft->state == RAFT_STATE_FOLLOWER ||
> +	       raft->state == RAFT_STATE_CANDIDATE);
> +	struct ev_loop *loop = raft_loop();
> +	struct ev_timer *timer = &raft->timer;
> +	double delay = raft_new_random_election_shift(raft);
> +	/*
> +	 * Could be too late to speed up anything - probably the term is almost
> +	 * over anyway.
> +	 */
> +	double remaining = raft_ev_timer_remaining(loop, timer);
> +	if (delay >= remaining)
> +		delay = remaining;
> +	say_info("RAFT: split vote is discovered - %s, new term in %lf sec",
> +		 raft_scores_str(raft), delay);
> +	raft_ev_timer_stop(loop, timer);
> +	raft_ev_timer_set(timer, delay, delay);


...
> diff --git a/test/unit/raft_test_utils.h b/test/unit/raft_test_utils.h
> index c68dc3b22..2138a829e 100644
> --- a/test/unit/raft_test_utils.h
> +++ b/test/unit/raft_test_utils.h
> @@ -32,6 +32,7 @@
>   #include "fakesys/fakeev.h"
>   #include "fiber.h"
>   #include "raft/raft.h"
> +#include "raft/raft_ev.h"

Why do you need it here?

>   #include "unit.h"
>   
>   /** WAL simulation. It stores a list of rows which raft wanted to persist. */
> @@ -105,6 +106,7 @@ struct raft_node {
>   	int cfg_election_quorum;
>   	double cfg_death_timeout;
>   	uint32_t cfg_instance_id;
> +	int cfg_cluster_size;
>   	struct vclock *cfg_vclock;
>   };
>   
> @@ -227,6 +229,9 @@ raft_node_cfg_is_enabled(struct raft_node *node, bool value);
>   void
>   raft_node_cfg_is_candidate(struct raft_node *node, bool value);
>   
> +void
> +raft_node_cfg_cluster_size(struct raft_node *node, int value);
> +
>   void
>   raft_node_cfg_election_timeout(struct raft_node *node, double value);
>   

-- 
Serge Petrenko