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;
}
int RaftNode::getNextIndex() const
{
if (!_node)
{
return -1;
}
return raft_node_get_next_idx(_node);
}
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;
}
void raft_send_appendentries(raft_server_t* me_, int node)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
__log(me_, "sending appendentries to: %d", node);
if (!(me->cb.send_appendentries))
return;
raft_node_t* p = raft_get_node(me_, node);
msg_appendentries_t ae;
ae.term = me->current_term;
ae.leader_id = me->nodeid;
ae.prev_log_term = raft_node_get_next_idx(p);
// TODO:
ae.prev_log_idx = 0;
ae.n_entries = 0;
me->cb.send_appendentries(me_, me->udata, node, &ae);
}
void raft_send_appendentries(raft_server_t* me_, int node)
{
msg_appendentries_t ae;
raft_server_private_t* me = (void*)me_;
__log(me_, NULL, "sending appendentries to: %d", node);
if (!(me->cb.send))
return;
raft_node_t* p = raft_get_node(me_, node);
ae.term = me->current_term;
ae.leader_id = me->nodeid;
ae.prev_log_term = raft_node_get_next_idx(p);
// TODO:
ae.prev_log_idx = 0;
ae.n_entries = 0;
me->cb.send(me->cb_ctx, me, node, RAFT_MSG_APPENDENTRIES,
(void*)&ae, sizeof(msg_appendentries_t));
}
int raft_send_appendentries(raft_server_t* me_, raft_node_t* node)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
assert(node);
assert(node != me->node);
if (!(me->cb.send_appendentries))
return -1;
msg_appendentries_t ae = {};
ae.term = me->current_term;
ae.leader_commit = raft_get_commit_idx(me_);
ae.prev_log_idx = 0;
ae.prev_log_term = 0;
int next_idx = raft_node_get_next_idx(node);
ae.entries = raft_get_entries_from_idx(me_, next_idx, &ae.n_entries);
/* previous log is the log just before the new logs */
if (1 < next_idx)
{
raft_entry_t* prev_ety = raft_get_entry_from_idx(me_, next_idx - 1);
ae.prev_log_idx = next_idx - 1;
if (prev_ety)
ae.prev_log_term = prev_ety->term;
}
__log(me_, node, "sending appendentries node: ci:%d t:%d lc:%d pli:%d plt:%d",
raft_get_current_idx(me_),
ae.term,
ae.leader_commit,
ae.prev_log_idx,
ae.prev_log_term);
me->cb.send_appendentries(me_, me->udata, node, &ae);
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;
}
void TestRaft_node_set_nextIdx(CuTest * tc)
{
raft_node_t *p = raft_node_new((void*)1, 1);
raft_node_set_next_idx(p, 3);
CuAssertTrue(tc, 3 == raft_node_get_next_idx(p));
}
