replication: Only update commit index if last stored is high enough

A leader might have its last_stored field lagging behind its commit_index,
after it committed an index that it did not yet persist, while enough followers
for forming a majority did.

We don't wont to update this leader's commit_index until last_stored has caught
up.

Signed-off-by: Free Ekanayaka <free@ekanayaka.io>

src/replication.c

@@ -902,7 +902,8 @@ static void appendFollowerCb(struct raft_io_append *req, int status)
      *   commitIndex, set commitIndex = min(leaderCommit, index of last new
      *   entry).
      */
-    if (args->leader_commit > r->commit_index) {
+    if (args->leader_commit > r->commit_index &&
+        r->last_stored >= r->commit_index) {
         r->commit_index = min(args->leader_commit, r->last_stored);
         rv = replicationApply(r);
         if (rv != 0) {
@@ -1101,6 +1102,7 @@ int replicationAppend(struct raft *r,
      */
     if (n == 0) {
         if ((args->leader_commit > r->commit_index) &&
+            r->last_stored >= r->commit_index &&
             !replicationInstallSnapshotBusy(r)) {
             r->commit_index = min(args->leader_commit, r->last_stored);
             rv = replicationApply(r);