int raft_recv_requestvote_response(raft_server_t* me_, int node,
msg_requestvote_response_t* r)
{
raft_server_private_t* me = (void*)me_;
__log(me_, NULL, "node responded to requestvote: %d status: %s",
node, r->vote_granted == 1 ? "granted" : "not granted");
if (raft_is_leader(me_))
return 0;
assert(node < me->num_nodes);
// if (r->term != raft_get_current_term(me_))
// return 0;
if (1 == r->vote_granted)
{
int votes;
me->votes_for_me[node] = 1;
votes = raft_get_nvotes_for_me(me_);
if (raft_votes_is_majority(me->num_nodes, votes))
raft_become_leader(me_);
}
return 0;
}
static void drop_bads(void)
{
int i;
for (i = 0; i < MAX_CLIENTS; i++)
{
Client *c = server.clients + i;
if (c->sock < 0) continue;
if (!c->good || !raft_is_leader(raft)) remove_client(c);
}
}
int raft_recv_entry(raft_server_t* me_,
msg_entry_t* e,
msg_entry_response_t *r)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
int i;
/* Only one voting cfg change at a time */
if (raft_entry_is_voting_cfg_change(e))
if (-1 != me->voting_cfg_change_log_idx)
return -1;
if (!raft_is_leader(me_))
return -1;
__log(me_, NULL, "received entry t:%d id: %d idx: %d",
me->current_term, e->id, raft_get_current_idx(me_) + 1);
raft_entry_t ety;
ety.term = me->current_term;
ety.id = e->id;
ety.type = e->type;
memcpy(&ety.data, &e->data, sizeof(raft_entry_data_t));
raft_append_entry(me_, &ety);
for (i = 0; i < me->num_nodes; i++)
{
if (me->node == me->nodes[i] || !me->nodes[i] ||
!raft_node_is_voting(me->nodes[i]))
continue;
/* Only send new entries.
* Don't send the entry to peers who are behind, to prevent them from
* becoming congested. */
int next_idx = raft_node_get_next_idx(me->nodes[i]);
if (next_idx == raft_get_current_idx(me_))
raft_send_appendentries(me_, me->nodes[i]);
}
/* if we're the only node, we can consider the entry committed */
if (1 == me->num_nodes)
me->commit_idx = raft_get_current_idx(me_);
r->id = e->id;
r->idx = raft_get_current_idx(me_);
r->term = me->current_term;
if (raft_entry_is_voting_cfg_change(e))
me->voting_cfg_change_log_idx = raft_get_current_idx(me_);
return 0;
}
static bool accept_client(void)
{
int fd;
fprintf(stderr, "a new connection is queued\n");
fd = accept(server.listener, NULL, NULL);
if (fd == -1) {
fprintf(stderr, "failed to accept a connection: %s\n", strerror(errno));
return false;
}
fprintf(stderr, "a new connection fd=%d accepted\n", fd);
if (!raft_is_leader(raft)) {
fprintf(stderr, "not a leader, disconnecting the accepted connection fd=%d\n", fd);
close(fd);
return false;
}
return add_client(fd);
}
int raft_recv_requestvote(raft_server_t* me_,
raft_node_t* node,
msg_requestvote_t* vr,
msg_requestvote_response_t *r)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
if (raft_get_current_term(me_) < vr->term)
{
raft_set_current_term(me_, vr->term);
raft_become_follower(me_);
}
if (__should_grant_vote(me, vr))
{
/* It shouldn't be possible for a leader or candidate to grant a vote
* Both states would have voted for themselves */
assert(!(raft_is_leader(me_) || raft_is_candidate(me_)));
raft_vote_for_nodeid(me_, vr->candidate_id);
r->vote_granted = 1;
/* there must be in an election. */
me->current_leader = NULL;
me->timeout_elapsed = 0;
}
else
r->vote_granted = 0;
__log(me_, node, "node requested vote: %d replying: %s",
node, r->vote_granted == 1 ? "granted" : "not granted");
r->term = raft_get_current_term(me_);
return 0;
}
int raft_recv_appendentries(
raft_server_t* me_,
const int node,
msg_appendentries_t* ae)
{
int i;
raft_server_private_t* me = (void*)me_;
msg_appendentries_response_t r;
me->timeout_elapsed = 0;
__log(me_, NULL, "received appendentries from: %d", node);
r.term = me->current_term;
/* we've found a leader who is legitimate */
if (raft_is_leader(me_) && me->current_term <= ae->term)
raft_become_follower(me_);
/* 1. Reply false if term < currentTerm (�1) */
if (ae->term < me->current_term)
{
__log(me_, NULL, "AE term is less than current term");
r.success = 0;
goto done;
}
#if 0
if (-1 != ae->prev_log_idx &&
ae->prev_log_idx < raft_get_current_idx(me_))
{
__log(me_, NULL, "AE prev_idx is less than current idx");
r.success = 0;
goto done;
}
#endif
/* not the first appendentries we've received */
if (0 != ae->prev_log_idx)
{
raft_entry_t* e;
if ((e = raft_get_entry_from_idx(me_, ae->prev_log_idx)))
{
/* 2. Reply false if log doesn抰 contain an entry at prevLogIndex
whose term matches prevLogTerm (�3) */
if (e->term != ae->prev_log_term)
{
__log(me_, NULL, "AE term doesn't match prev_idx");
r.success = 0;
goto done;
}
/* 3. If an existing entry conflicts with a new one (same index
but different terms), delete the existing entry and all that
follow it (�3) */
raft_entry_t* e2;
if ((e2 = raft_get_entry_from_idx(me_, ae->prev_log_idx+1)))
{
log_delete(me->log, ae->prev_log_idx+1);
}
}
else
{
__log(me_, NULL, "AE no log at prev_idx");
r.success = 0;
goto done;
//assert(0);
}
}
/* 5. If leaderCommit > commitIndex, set commitIndex =
min(leaderCommit, last log index) */
if (raft_get_commit_idx(me_) < ae->leader_commit)
{
raft_entry_t* e;
if ((e = log_peektail(me->log)))
{
raft_set_commit_idx(me_, e->id < ae->leader_commit ?
e->id : ae->leader_commit);
while (1 == raft_apply_entry(me_));
}
}
if (raft_is_candidate(me_))
raft_become_follower(me_);
raft_set_current_term(me_, ae->term);
/* append all entries to log */
for (i=0; i<ae->n_entries; i++)
{
msg_entry_t* cmd;
raft_entry_t* c;
cmd = &ae->entries[i];
/* TODO: replace malloc with mempoll/arena */
c = malloc(sizeof(raft_entry_t));
c->term = me->current_term;
c->len = cmd->len;
c->id = cmd->id;
c->data = malloc(cmd->len);
memcpy(c->data, cmd->data, cmd->len);
if (0 == raft_append_entry(me_, c))
{
__log(me_, NULL, "AE failure; couldn't append entry");
r.success = 0;
goto done;
}
}
r.success = 1;
r.current_idx = raft_get_current_idx(me_);
r.first_idx = ae->prev_log_idx + 1;
done:
if (me->cb.send)
me->cb.send(me->cb_ctx, me, node, RAFT_MSG_APPENDENTRIES_RESPONSE,
(void*)&r, sizeof(msg_appendentries_response_t));
return 1;
}
int raft_recv_appendentries(
raft_server_t* me_,
const int node,
msg_appendentries_t* ae,
msg_appendentries_response_t *r
)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
me->timeout_elapsed = 0;
__log(me_, "received appendentries from: %d", node);
r->term = me->current_term;
/* we've found a leader who is legitimate */
if (raft_is_leader(me_) && me->current_term <= ae->term)
raft_become_follower(me_);
/* 1. Reply false if term < currentTerm (§5.1) */
if (ae->term < me->current_term)
{
__log(me_, "AE term is less than current term");
r->success = 0;
return 0;
}
#if 0
if (-1 != ae->prev_log_idx &&
ae->prev_log_idx < raft_get_current_idx(me_))
{
__log(me_, "AE prev_idx is less than current idx");
r->success = 0;
return 0;
}
#endif
/* not the first appendentries we've received */
if (0 != ae->prev_log_idx)
{
raft_entry_t* e = raft_get_entry_from_idx(me_, ae->prev_log_idx);
if (e)
{
/* 2. Reply false if log doesn't contain an entry at prevLogIndex
whose term matches prevLogTerm (§5.3) */
if (e->term != ae->prev_log_term)
{
__log(me_, "AE term doesn't match prev_idx");
r->success = 0;
return 0;
}
/* 3. If an existing entry conflicts with a new one (same index
but different terms), delete the existing entry and all that
follow it (§5.3) */
raft_entry_t* e2;
e2 = raft_get_entry_from_idx(me_, ae->prev_log_idx + 1);
if (e2)
log_delete(me->log, ae->prev_log_idx + 1);
}
else
{
__log(me_, "AE no log at prev_idx");
r->success = 0;
return 0;
}
}
/* 5. If leaderCommit > commitIndex, set commitIndex =
min(leaderCommit, last log index) */
if (raft_get_commit_idx(me_) < ae->leader_commit)
{
raft_entry_t* e = log_peektail(me->log);
if (e)
{
int id = e->id < ae->leader_commit ? e->id : ae->leader_commit;
raft_set_commit_idx(me_, id);
while (0 == raft_apply_entry(me_))
;
}
}
if (raft_is_candidate(me_))
raft_become_follower(me_);
raft_set_current_term(me_, ae->term);
int i;
/* append all entries to log */
for (i = 0; i < ae->n_entries; i++)
{
msg_entry_t* cmd = &ae->entries[i];
/* TODO: replace malloc with mempoll/arena */
raft_entry_t* c = (raft_entry_t*)malloc(sizeof(raft_entry_t));
c->term = me->current_term;
c->len = cmd->len;
c->id = cmd->id;
c->data = (unsigned char*)malloc(cmd->len);
memcpy(c->data, cmd->data, cmd->len);
if (-1 == raft_append_entry(me_, c))
{
__log(me_, "AE failure; couldn't append entry");
r->success = 0;
return -1;
}
}
r->success = 1;
r->current_idx = raft_get_current_idx(me_);
r->first_idx = ae->prev_log_idx + 1;
return 0;
}
int raft_recv_appendentries_response(raft_server_t* me_,
raft_node_t* node,
msg_appendentries_response_t* r)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
__log(me_, node,
"received appendentries response %s ci:%d rci:%d 1stidx:%d",
r->success == 1 ? "SUCCESS" : "fail",
raft_get_current_idx(me_),
r->current_idx,
r->first_idx);
/* Stale response -- ignore */
if (r->current_idx != 0 && r->current_idx <= raft_node_get_match_idx(node))
return 0;
if (!raft_is_leader(me_))
return -1;
/* If response contains term T > currentTerm: set currentTerm = T
and convert to follower (§5.3) */
if (me->current_term < r->term)
{
raft_set_current_term(me_, r->term);
raft_become_follower(me_);
return 0;
}
else if (me->current_term != r->term)
return 0;
/* stop processing, this is a node we don't have in our configuration */
if (!node)
return 0;
if (0 == r->success)
{
/* If AppendEntries fails because of log inconsistency:
decrement nextIndex and retry (§5.3) */
assert(0 <= raft_node_get_next_idx(node));
int next_idx = raft_node_get_next_idx(node);
assert(0 <= next_idx);
if (r->current_idx < next_idx - 1)
raft_node_set_next_idx(node, min(r->current_idx + 1, raft_get_current_idx(me_)));
else
raft_node_set_next_idx(node, next_idx - 1);
/* retry */
raft_send_appendentries(me_, node);
return 0;
}
assert(r->current_idx <= raft_get_current_idx(me_));
raft_node_set_next_idx(node, r->current_idx + 1);
raft_node_set_match_idx(node, r->current_idx);
if (!raft_node_is_voting(node) &&
-1 == me->voting_cfg_change_log_idx &&
raft_get_current_idx(me_) <= r->current_idx + 1 &&
me->cb.node_has_sufficient_logs &&
0 == raft_node_has_sufficient_logs(node)
)
{
raft_node_set_has_sufficient_logs(node);
me->cb.node_has_sufficient_logs(me_, me->udata, node);
}
/* Update commit idx */
int votes = 1; /* include me */
int point = r->current_idx;
int i;
for (i = 0; i < me->num_nodes; i++)
{
if (me->node == me->nodes[i] || !raft_node_is_voting(me->nodes[i]))
continue;
int match_idx = raft_node_get_match_idx(me->nodes[i]);
if (0 < match_idx)
{
raft_entry_t* ety = raft_get_entry_from_idx(me_, match_idx);
if (ety->term == me->current_term && point <= match_idx)
votes++;
}
}
if (me->num_nodes / 2 < votes && raft_get_commit_idx(me_) < point)
raft_set_commit_idx(me_, point);
/* Aggressively send remaining entries */
if (raft_get_entry_from_idx(me_, raft_node_get_next_idx(node)))
raft_send_appendentries(me_, node);
/* periodic applies committed entries lazily */
return 0;
}