Fix wrong replication in the sharded setting

dsg-uwaterloo · Feb 22, 2024 · 3f02c9f · 3f02c9f
1 parent 5dbe9ce
commit 3f02c9f
Show file tree

Hide file tree

Showing 3 changed files with 8 additions and 5 deletions.
diff --git a/ansible/templates/oblishard-oramnode.service.j2 b/ansible/templates/oblishard-oramnode.service.j2
@@ -13,7 +13,8 @@ ExecStart=/root/oblishard/oramnode/oramnode \
     -replicaid {{ item.replicaid }} \
     -raftport {{ item.raftport }} \
     {% if item.replicaid!=0 %}
-    {{'-joinaddr=' + oramnode_endpoints.endpoints[item.id].exposed_ip + ':' + oramnode_endpoints.endpoints[item.id].port | string }} \
+    # this assumes there are 3 replicas (Dangerous assmumption)
+    {{'-joinaddr=' + oramnode_endpoints.endpoints[item.id * 3].exposed_ip + ':' + oramnode_endpoints.endpoints[item.id].port | string }} \
     {% endif %}
     -conf /root/oblishard/ \
     -logpath {{ '/root/oblishard/oramnode/'  + item.id | string + '/' + item.replicaid | string + '/all.log' }}

diff --git a/ansible/templates/oblishard-shardnode.service.j2 b/ansible/templates/oblishard-shardnode.service.j2
@@ -13,7 +13,8 @@ ExecStart=/root/oblishard/shardnode/shardnode \
     -replicaid {{ item.replicaid }} \
     -raftport {{ item.raftport }} \
     {% if item.replicaid!=0 %}
-    {{'-joinaddr=' + shardnode_endpoints.endpoints[item.id].exposed_ip + ':' + shardnode_endpoints.endpoints[item.id].port | string }} \
+    # this assumes there are 3 replicas (Dangerous assmumption)
+    {{'-joinaddr=' + shardnode_endpoints.endpoints[item.id * 3].exposed_ip + ':' + shardnode_endpoints.endpoints[item.id].port | string }} \
     {% endif %}
     -conf /root/oblishard/ \
     -logpath {{ '/root/oblishard/shardnode/'  + item.id | string + '/' + item.replicaid | string + '/all.log' }}

diff --git a/pkg/shardnode/raft.go b/pkg/shardnode/raft.go
@@ -276,9 +276,10 @@ func (fsm *shardNodeFSM) Restore(rc io.ReadCloser) error {
 func startRaftServer(isFirst bool, bindIP string, advertiseIP string, replicaID int, raftPort int, shardshardNodeFSM *shardNodeFSM) (*raft.Raft, error) {
 
 	raftConfig := raft.DefaultConfig()
-	raftConfig.ElectionTimeout = 150 * time.Millisecond
-	raftConfig.HeartbeatTimeout = 150 * time.Millisecond
-	raftConfig.LeaderLeaseTimeout = 150 * time.Millisecond
+	// TODO: set these carefuly for the crash experiments
+	// raftConfig.ElectionTimeout = 150 * time.Millisecond
+	// raftConfig.HeartbeatTimeout = 150 * time.Millisecond
+	// raftConfig.LeaderLeaseTimeout = 150 * time.Millisecond
 	raftConfig.Logger = hclog.New(&hclog.LoggerOptions{Output: log.Logger})
 	raftConfig.LocalID = raft.ServerID(strconv.Itoa(replicaID))