int raft_apply_all(raft_server_t* me_)
{
while (raft_get_last_applied_idx(me_) < raft_get_commit_idx(me_))
{
int e = raft_apply_entry(me_);
if (RAFT_ERR_SHUTDOWN == e)
return RAFT_ERR_SHUTDOWN;
}
return 0;
}
int raft_msg_entry_response_committed(raft_server_t* me_,
const msg_entry_response_t* r)
{
raft_entry_t* ety = raft_get_entry_from_idx(me_, r->idx);
if (!ety)
return 0;
/* entry from another leader has invalidated this entry message */
if (r->term != ety->term)
return -1;
return r->idx <= raft_get_commit_idx(me_);
}
void TestRaft_server_wont_apply_entry_if_we_dont_have_entry_to_apply(CuTest* tc)
{
raft_entry_t ety;
char *str = "aaa";
void *r = raft_new();
raft_set_commit_idx(r, 0);
raft_set_last_applied_idx(r, 0);
raft_apply_entry(r);
CuAssertTrue(tc, 0 == raft_get_last_applied_idx(r));
CuAssertTrue(tc, 0 == raft_get_commit_idx(r));
ety.term = 1;
ety.id = 1;
ety.data.buf = str;
ety.data.len = 3;
raft_append_entry(r, &ety);
raft_apply_entry(r);
CuAssertTrue(tc, 1 == raft_get_last_applied_idx(r));
CuAssertTrue(tc, 1 == raft_get_commit_idx(r));
}
void TestRaft_follower_recv_appendentries_set_commitidx_to_LeaderCommit(
CuTest * tc)
{
void *r = raft_new();
raft_add_node(r, (void*)1, 1);
raft_add_node(r, (void*)2, 0);
msg_appendentries_t ae;
msg_appendentries_response_t aer;
memset(&ae, 0, sizeof(msg_appendentries_t));
ae.term = 1;
ae.prev_log_idx = 0;
ae.prev_log_term = 1;
/* include entries */
msg_entry_t e[4];
memset(&e, 0, sizeof(msg_entry_t) * 4);
e[0].term = 1;
e[0].id = 1;
e[1].term = 1;
e[1].id = 2;
e[2].term = 1;
e[2].id = 3;
e[3].term = 1;
e[3].id = 4;
ae.entries = e;
ae.n_entries = 4;
raft_recv_appendentries(r, 1, &ae, &aer);
/* receive an appendentry with commit */
memset(&ae, 0, sizeof(msg_appendentries_t));
ae.term = 1;
ae.prev_log_term = 1;
ae.prev_log_idx = 3;
ae.leader_commit = 3;
/* receipt of appendentries changes commit idx */
raft_recv_appendentries(r, 1, &ae, &aer);
CuAssertTrue(tc, 1 == aer.success);
/* set to 3 because leaderCommit is lower */
CuAssertTrue(tc, 3 == raft_get_commit_idx(r));
}
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(
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;
}
void raft_apply_all(raft_server_t* me_)
{
while (raft_get_last_applied_idx(me_) < raft_get_commit_idx(me_))
raft_apply_entry(me_);
}
int raft_recv_appendentries(
raft_server_t* me_,
raft_node_t* node,
msg_appendentries_t* ae,
msg_appendentries_response_t *r
)
{
raft_server_private_t* me = (raft_server_private_t*)me_;
me->timeout_elapsed = 0;
if (0 < ae->n_entries)
__log(me_, node, "recvd appendentries from: %lx, t:%d ci:%d lc:%d pli:%d plt:%d #%d",
node,
ae->term,
raft_get_current_idx(me_),
ae->leader_commit,
ae->prev_log_idx,
ae->prev_log_term,
ae->n_entries);
r->term = me->current_term;
if (raft_is_candidate(me_) && me->current_term == ae->term)
{
me->voted_for = -1;
raft_become_follower(me_);
}
else if (me->current_term < ae->term)
{
raft_set_current_term(me_, ae->term);
r->term = ae->term;
raft_become_follower(me_);
}
else if (ae->term < me->current_term)
{
/* 1. Reply false if term < currentTerm (§5.1) */
__log(me_, node, "AE term %d is less than current term %d",
ae->term, me->current_term);
goto fail_with_current_idx;
}
/* Not the first appendentries we've received */
/* NOTE: the log starts at 1 */
if (0 < ae->prev_log_idx)
{
raft_entry_t* e = raft_get_entry_from_idx(me_, ae->prev_log_idx);
if (!e)
{
__log(me_, node, "AE no log at prev_idx %d", ae->prev_log_idx);
goto fail_with_current_idx;
}
/* 2. Reply false if log doesn't contain an entry at prevLogIndex
whose term matches prevLogTerm (§5.3) */
if (raft_get_current_idx(me_) < ae->prev_log_idx)
goto fail_with_current_idx;
if (e->term != ae->prev_log_term)
{
__log(me_, node, "AE term doesn't match prev_term (ie. %d vs %d) ci:%d pli:%d",
e->term, ae->prev_log_term, raft_get_current_idx(me_), ae->prev_log_idx);
assert(me->commit_idx < ae->prev_log_idx);
/* Delete all the following log entries because they don't match */
log_delete(me->log, ae->prev_log_idx);
r->current_idx = ae->prev_log_idx - 1;
goto fail;
}
}
/* 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) */
if (ae->n_entries == 0 && 0 < ae->prev_log_idx && ae->prev_log_idx + 1 < raft_get_current_idx(me_))
{
assert(me->commit_idx < ae->prev_log_idx + 1);
log_delete(me->log, ae->prev_log_idx + 1);
}
r->current_idx = ae->prev_log_idx;
int i;
for (i = 0; i < ae->n_entries; i++)
{
msg_entry_t* ety = &ae->entries[i];
int ety_index = ae->prev_log_idx + 1 + i;
raft_entry_t* existing_ety = raft_get_entry_from_idx(me_, ety_index);
r->current_idx = ety_index;
if (existing_ety && existing_ety->term != ety->term)
{
assert(me->commit_idx < ety_index);
log_delete(me->log, ety_index);
break;
}
else if (!existing_ety)
break;
}
/* Pick up remainder in case of mismatch or missing entry */
for (; i < ae->n_entries; i++)
{
int e = raft_append_entry(me_, &ae->entries[i]);
if (-1 == e)
goto fail_with_current_idx;
r->current_idx = ae->prev_log_idx + 1 + i;
}
/* 4. If leaderCommit > commitIndex, set commitIndex =
min(leaderCommit, index of most recent entry) */
if (raft_get_commit_idx(me_) < ae->leader_commit)
{
int last_log_idx = max(raft_get_current_idx(me_), 1);
raft_set_commit_idx(me_, min(last_log_idx, ae->leader_commit));
}
/* update current leader because we accepted appendentries from it */
me->current_leader = node;
r->success = 1;
r->first_idx = ae->prev_log_idx + 1;
return 0;
fail_with_current_idx:
r->current_idx = raft_get_current_idx(me_);
fail:
r->success = 0;
r->first_idx = 0;
return -1;
}
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;
}
