lighthouse, manager: remove room support (#70)

pytorch · Jan 13, 2025 · 97ad397 · 97ad397
1 parent e0f76e1
commit 97ad397
Show file tree

Hide file tree

Showing 7 changed files with 101 additions and 186 deletions.
diff --git a/proto/torchft.proto b/proto/torchft.proto
@@ -50,11 +50,7 @@ message Quorum {
 }
 
 message LighthouseQuorumRequest {
-    // 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;
+    QuorumMember requester = 1;
 }
 
 message LighthouseQuorumResponse {
@@ -73,13 +69,9 @@ service LighthouseService {
 }
 
 message ManagerQuorumRequest {
-    // 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;
+    int64 rank = 1;
+    int64 step = 2;
+    string checkpoint_server_addr = 3;
 }
 
 message ManagerQuorumResponse {

diff --git a/src/lib.rs b/src/lib.rs
@@ -105,19 +105,17 @@ impl ManagerClient {
         })
     }
 
-    #[pyo3(signature = (room_id, rank, step, checkpoint_server_addr, timeout=None))]
+    #[pyo3(signature = (rank, step, checkpoint_server_addr, timeout=None))]
     fn quorum(
         &mut self,
         py: Python<'_>,
-        room_id: String,
         rank: i64,
         step: i64,
         checkpoint_server_addr: String,
         timeout: Option<Duration>,
     ) -> Result<(i64, i64, i64, String, String, i64, Option<i64>, i64, bool), StatusError> {
         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,