[Tarantool-patches] [PATCH v2 2/8] txn_limbo: fix promote term filtering

[Cyrill Gorcunov]

On Fri, Jun 18, 2021 at 12:07:36AM +0300, Serge Petrenko wrote:
> txn_limbo_process() used to filter out promote requests whose term was
> equal to the greatest term seen. This wasn't correct for PROMOTE entries
> with term 1.
> 
> Such entries appear after box.ctl.promote() is issued on an instance
> with disabled elections. Every PROMOTE entry from such an instance has
> term 1, but should still be applied. Fix this in the patch.
> 
> Also, when an outdated PROMOTE entry with term smaller than already
> applied from some replica arrived, it wasn't filtered at all. Such a
> situation shouldn't be possible, but fix it as well.
> 
> Part-of #6034
> ---
>  src/box/txn_limbo.c | 22 +++++++++++++---------
>  1 file changed, 13 insertions(+), 9 deletions(-)
> 
> diff --git a/src/box/txn_limbo.c b/src/box/txn_limbo.c
> index 51dc2a186..a5d1df00c 100644
> --- a/src/box/txn_limbo.c
> +++ b/src/box/txn_limbo.c
> @@ -665,17 +665,21 @@ txn_limbo_process(struct txn_limbo *limbo, const struct synchro_request *req)
>  {
>  	uint64_t term = req->term;
>  	uint32_t origin = req->origin_id;
> -	if (txn_limbo_replica_term(limbo, origin) < term) {
> +	if (txn_limbo_replica_term(limbo, origin) < term)
>  		vclock_follow(&limbo->promote_term_map, origin, term);
> -		if (term > limbo->promote_greatest_term) {
> -			limbo->promote_greatest_term = term;
> -		} else if (req->type == IPROTO_PROMOTE) {
> -			/*
> -			 * PROMOTE for outdated term. Ignore.
> -			 */
> -			return;
> -		}
> +
> +	if (term > limbo->promote_greatest_term) {
> +		limbo->promote_greatest_term = term;
> +	} else if (req->type == IPROTO_PROMOTE &&
> +		   limbo->promote_greatest_term > 1) {
> +		/* PROMOTE for outdated term. Ignore. */
> +		say_info("RAFT: ignoring PROMOTE request from instance "
> +			 "id %"PRIu32" for term %"PRIu64". Greatest term seen "
> +			 "before (%"PRIu64") is bigger.", origin, term,
> +			 limbo->promote_greatest_term);
> +		return;
>  	}

Serge, here is a moment I don't understand. The promote_term_map
and promote_greatest_term both are zero initially, then we start
accepting requests from the net and previously we've been updating
promote_term_map and promote_greatest_term in same context, iow
promote_term_map is filtering incoming traffic and update
promote_greatest_term with max from promote_term_map, thus
we can denote it as

	if (term > promote_term_map[i]) {
		promote_term_map[i] = term;
		promote_greatest_term = max(promote_greatest_term, term);
	}

thus promote_greatest_term always = max({promote_term_map[i]}), so I don't
understart why say

	uint64_t term = req->term;
	uint32_t origin = req->origin_id;

	if (txn_limbo_replica_term(limbo, origin) < term) {
		vclock_follow(&limbo->promote_term_map, origin, term);
		if (term > limbo->promote_greatest_term)
			limbo->promote_greatest_term = term;
	}

	if (iproto_type_is_promote_request(req->type) &&
	    limbo->promote_greatest_term > 1) {
		/* PROMOTE for outdated term. Ignore. */
		say_info("RAFT: ignoring %s request from instance "
			 "id %"PRIu32" for term %"PRIu64". Greatest term seen "
			 "before (%"PRIu64") is bigger.",
			 iproto_type_name(req->type), origin, term,
			 limbo->promote_greatest_term);
		return;
	}

won't do the trick? Do I miss something obvious here?