lighthouse, manager: support multiple quorum rooms (#48)

Author: d4l3k

proto/torchft.proto

@@ -50,7 +50,11 @@ message Quorum {
 }
 
 message LighthouseQuorumRequest {
-    QuorumMember requester = 1;
+    // room_id is the specific quorum channel to use. All workers/replicas
+    // participating in the quorum must specify the same channel.
+    // Multiple channels can be active simultaneously.
+    string room_id = 1;
+    QuorumMember requester = 2;
 }
 
 message LighthouseQuorumResponse {
@@ -69,9 +73,13 @@ service LighthouseService {
 }
 
 message ManagerQuorumRequest {
-    int64 rank = 1;
-    int64 step = 2;
-    string checkpoint_server_addr = 3;
+    // room_id is the specific quorum channel to use. All workers/replicas
+    // participating in the quorum must specify the same channel.
+    // Multiple channels can be active simultaneously.
+    string room_id = 1;
+    int64 rank = 2;
+    int64 step = 3;
+    string checkpoint_server_addr = 4;
 }
 
 message ManagerQuorumResponse {

src/lib.rs

@@ -105,12 +105,14 @@ impl ManagerClient {
     fn quorum(
         &mut self,
         py: Python<'_>,
+        room_id: String,
         rank: i64,
         step: i64,
         checkpoint_server_addr: String,
     ) -> PyResult<(i64, i64, i64, String, String, i64, Option<i64>, i64, bool)> {
         py.allow_threads(move || {
             let mut request = tonic::Request::new(ManagerQuorumRequest {
+                room_id: room_id,
                 rank: rank,
                 step: step,
                 checkpoint_server_addr: checkpoint_server_addr,