void raft_send_appendentries_all(raft_server_t* me_)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
int i;
for (i = 0; i < me->num_nodes; i++)
if (me->nodeid != i)
raft_send_appendentries(me_, i);
}
void raft_send_appendentries_all(raft_server_t* me_)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
int i;
me->timeout_elapsed = 0;
for (i = 0; i < me->num_nodes; i++)
if (me->node != me->nodes[i])
raft_send_appendentries(me_, me->nodes[i]);
}
void raft_send_appendentries_all(raft_server_t* me_)
{
raft_server_private_t* me = (void*)me_;
int i;
for (i=0; i<me->num_nodes; i++)
{
if (me->nodeid == i) continue;
raft_send_appendentries(me_, i);
}
}
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;
}
int raft_recv_appendentries_response(raft_server_t* me_,
int node, msg_appendentries_response_t* r)
{
raft_server_private_t* me = (void*)me_;
raft_node_t* p;
__log(me_, NULL, "received appendentries response from: %d", node);
p = raft_get_node(me_, node);
if (1 == r->success)
{
int i;
for (i=r->first_idx; i<=r->current_idx; i++)
log_mark_node_has_committed(me->log, i);
while (1)
{
raft_entry_t* e;
e = log_get_from_idx(me->log, me->last_applied_idx + 1);
/* majority has this */
if (e && me->num_nodes / 2 <= e->num_nodes)
{
if (0 == raft_apply_entry(me_)) break;
}
else
{
break;
}
}
}
else
{
/* If AppendEntries fails because of log inconsistency:
decrement nextIndex and retry (�3) */
assert(0 <= raft_node_get_next_idx(p));
// TODO does this have test coverage?
// TODO can jump back to where node is different instead of iterating
raft_node_set_next_idx(p, raft_node_get_next_idx(p)-1);
raft_send_appendentries(me_, node);
}
return 1;
}
void raft_become_leader(raft_server_t* me_)
{
raft_server_private_t* me = (void*)me_;
int i;
__log(me_, NULL, "becoming leader");
raft_set_state(me_,RAFT_STATE_LEADER);
me->voted_for = -1;
for (i=0; i<me->num_nodes; i++)
{
if (me->nodeid == i) continue;
raft_node_t* p = raft_get_node(me_, i);
raft_node_set_next_idx(p, raft_get_current_idx(me_)+1);
raft_send_appendentries(me_, i);
}
}
void raft_become_leader(raft_server_t* me_)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
int i;
__log(me_, NULL, "becoming leader term:%d", raft_get_current_term(me_));
raft_set_state(me_, RAFT_STATE_LEADER);
for (i = 0; i < me->num_nodes; i++)
{
if (me->node == me->nodes[i] || !raft_node_is_voting(me->nodes[i]))
continue;
raft_node_t* node = me->nodes[i];
raft_node_set_next_idx(node, raft_get_current_idx(me_) + 1);
raft_node_set_match_idx(node, 0);
raft_send_appendentries(me_, node);
}
}
int raft_recv_entry(raft_server_t* me_, int node, msg_entry_t* e)
{
raft_server_private_t* me = (void*)me_;
raft_entry_t ety;
int res, i;
__log(me_, NULL, "received entry from: %d", node);
ety.term = me->current_term;
ety.id = e->id;
ety.data = e->data;
ety.len = e->len;
res = raft_append_entry(me_, &ety);
raft_send_entry_response(me_, node, e->id, res);
for (i=0; i<me->num_nodes; i++)
{
if (me->nodeid == i) continue;
raft_send_appendentries(me_,i);
}
return 0;
}
int raft_recv_entry(raft_server_t* me_, int node, msg_entry_t* e,
msg_entry_response_t *r)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
raft_entry_t ety;
int res, i;
__log(me_, "received entry from: %d", node);
ety.term = me->current_term;
ety.id = e->id;
ety.data = e->data;
ety.len = e->len;
res = raft_append_entry(me_, &ety);
for (i = 0; i < me->num_nodes; i++)
if (me->nodeid != i)
raft_send_appendentries(me_, i);
r->id = e->id;
r->was_committed = (0 == res);
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;
}