Fix panic in RemoteProcessAlloc select_all (#281)

Summary:
Pull Request resolved: #281

There was a bug in the remote process allocator, where if all hosts became closed, the loop
would hit a panic in `select_all` given an empty list of handles.
This change exits out of the loop instead, completing the task.

Added some asserts en route to the allocator to ensure it doesn't get a list of zero hosts to
query.

And some more `__repr__`'s which were helpful when debugging with logs.

Reviewed By: shayne-fletcher

Differential Revision: D76750206

fbshipit-source-id: 6c24d0a46a2205ba472aeefd1e806a7a3d94093f

Author: dulinriley

hyperactor_extension/src/alloc.rs

@@ -49,6 +49,17 @@ impl PyAlloc {
     }
 }
 
+#[pymethods]
+impl PyAlloc {
+    fn __repr__(&self) -> PyResult<String> {
+        let data = self.inner.lock().unwrap();
+        match &*data {
+            None => Ok("Alloc(None)".to_string()),
+            Some(wrapper) => Ok(format!("Alloc({})", wrapper.shape())),
+        }
+    }
+}
+
 /// Internal wrapper to translate from a dyn Alloc to an impl Alloc. Used
 /// to support polymorphism in the Python bindings.
 pub struct PyAllocWrapper {

hyperactor_mesh/src/alloc/remoteprocess.rs

@@ -529,6 +529,7 @@ impl RemoteProcessAlloc {
             let watcher = WatchStream::new(tx_status);
             tx_watchers.push((watcher, host.id.clone()));
         }
+        assert!(!tx_watchers.is_empty());
         let tx = self.comm_watcher_tx.clone();
         tokio::spawn(async move {
             loop {
@@ -558,6 +559,10 @@ impl RemoteProcessAlloc {
                             break;
                         }
                         tx_watchers.remove(index);
+                        if tx_watchers.is_empty() {
+                            // All of the statuses have been closed, exit the loop.
+                            break;
+                        }
                     }
                 }
             }
@@ -579,6 +584,9 @@ impl RemoteProcessAlloc {
             .initialize_alloc()
             .await
             .context("alloc initializer error")?;
+        if hosts.is_empty() {
+            anyhow::bail!("Initializer returned empty list of hosts");
+        }
         // prepare a list of host names in this allocation to be sent
         // to remote allocators.
         let hostnames: Vec<_> = hosts.iter().map(|e| e.hostname.clone()).collect();

python/monarch/actor_mesh.py

@@ -337,6 +337,9 @@ def __iter__(self):
     def __len__(self) -> int:
         return len(self._shape)
 
+    def __repr__(self) -> str:
+        return f"ValueMesh({self._shape})"
+
     @property
     def _ndslice(self) -> NDSlice:
         return self._shape.ndslice
@@ -702,6 +705,9 @@ def _new_with_shape(self, shape: Shape) -> "ActorMeshRef":
             self._mailbox,
         )
 
+    def __repr__(self) -> str:
+        return f"ActorMeshRef(class={self._class}, shape={self._actor_mesh_ref._shape})"
+
 
 class ActorError(Exception):
     """