From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from smtp41.i.mail.ru (smtp41.i.mail.ru [94.100.177.101]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by dev.tarantool.org (Postfix) with ESMTPS id 03595469719 for ; Mon, 21 Sep 2020 11:20:53 +0300 (MSK) References: <3c916b9b7e70c4fbfa2b95c4aeb4146e83c82c58.1599693319.git.v.shpilevoy@tarantool.org> <3e56d915-2078-1181-ae6c-1ee1743e0e1c@tarantool.org> From: Serge Petrenko Message-ID: Date: Mon, 21 Sep 2020 11:20:52 +0300 MIME-Version: 1.0 In-Reply-To: <3e56d915-2078-1181-ae6c-1ee1743e0e1c@tarantool.org> Content-Type: text/plain; charset="utf-8"; format="flowed" Content-Transfer-Encoding: 8bit Content-Language: ru Subject: Re: [Tarantool-patches] [PATCH v2 09/11] raft: introduce state machine List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Vladislav Shpilevoy , tarantool-patches@dev.tarantool.org, gorcunov@gmail.com 19.09.2020 18:50, Vladislav Shpilevoy пишет: > And my new changes on top of it, in a separate commit. > > ==================== > > diff --git a/src/box/applier.cc b/src/box/applier.cc > index c352faf5e..10186ab91 100644 > --- a/src/box/applier.cc > +++ b/src/box/applier.cc > @@ -885,6 +885,11 @@ static int > applier_handle_raft(struct applier *applier, struct xrow_header *row) > { > assert(iproto_type_is_raft_request(row->type)); > + if (applier->instance_id == 0) { > + diag_set(ClientError, ER_PROTOCOL, "Can't apply a Raft request " > + "from an instance without an ID"); > + return -1; > + } > > struct raft_request req; > struct vclock candidate_clock; > diff --git a/src/box/raft.c b/src/box/raft.c > index 07a49351f..b3ab94bd7 100644 > --- a/src/box/raft.c > +++ b/src/box/raft.c > @@ -347,6 +347,7 @@ raft_process_msg(const struct raft_request *req, uint32_t source) > if (req->vote != 0) { > switch (raft.state) { > case RAFT_STATE_FOLLOWER: > + case RAFT_STATE_LEADER: > /* > * Can't respond on vote requests when Raft is disabled. > */ > @@ -386,6 +387,12 @@ raft_process_msg(const struct raft_request *req, uint32_t source) > "indirect votes are not allowed"); > break; > } > + if (raft.leader != 0) { > + say_info("RAFT: vote request is skipped - the " > + "leader is already known - %u", > + raft.leader); > + break; > + } > /* > * Either the term is new, or didn't vote in the current > * term yet. Anyway can vote now. > @@ -415,16 +422,6 @@ raft_process_msg(const struct raft_request *req, uint32_t source) > } > raft_sm_become_leader(); > break; > - case RAFT_STATE_LEADER: > - /* > - * If the node is still a leader, it ignores all votes. > - * Indeed, if the received vote would be from a new > - * term, the node would bump its own term and would > - * enter the follower state by now. If the vote is from > - * the current term, then the leader can freely ignore > - * it. > - */ > - break; > default: > unreachable(); > } > @@ -466,6 +463,12 @@ end: > void > raft_process_heartbeat(uint32_t source) > { > + /* > + * Raft handles heartbeats from all instances, including anon instances > + * which don't participate in Raft. > + */ > + if (source == 0) > + return; > /* > * When not a candidate - don't wait for anything. Therefore do not care > * about the leader being dead. > @@ -746,6 +749,7 @@ raft_sm_schedule_new_vote(uint32_t new_vote) > { > say_info("RAFT: vote for %u, follow", new_vote, raft.volatile_term); > assert(raft.volatile_vote == 0); > + assert(raft.leader == 0); > assert(raft.state == RAFT_STATE_FOLLOWER); > raft.volatile_vote = new_vote; > raft_sm_pause_and_dump(); > @@ -929,15 +933,8 @@ raft_cfg_election_quorum(void) > return; > if (raft.vote_count < raft_election_quorum()) > return; > - /* > - * The node is a candidate. It means its state if fully synced with > - * disk. Otherwise it would be a follower. > - */ > - assert(!raft.is_write_in_progress); > - raft.state = RAFT_STATE_LEADER; > - raft.leader = instance_id; > + raft_sm_become_leader(); > raft_broadcast_new_state(); > - box_update_ro_summary(); > } > > void Hi! Thanks for the fixes! LGTM. -- Serge Petrenko