feat: handle non-matching transport path removal

Signed-off-by: Diwakar Sharma <[email protected]>

control-plane/agents/src/bin/ha/node/server.rs

@@ -171,9 +171,31 @@ impl NodeAgentSvc {
 
         tracing::info!(new_path, "Connecting to NVMe target");
 
-        // Check to ensure Nqn is already connected
-        if Subsystem::try_from_nqn(&nqn).is_err() {
-            return Err(SvcError::ReplaceNqnNotFound { nqn });
+        // Check to ensure Nqn is already connected.
+        // Additionally, if a path is found for this nqn where the transport is
+        // not the same as current parsed_uri's transport, we disconnect that path.
+        // Such a path could be to the same host as parsed_uri host or possibly a
+        // different host perhaps, but it is fairly safe to disconnect it. This should
+        // also take care of the fact that - even when the nexus/subsystem is rehosted on
+        // same node/host, the traddr might be different because with rdma the interface
+        // name of RNIC is explicitly set and that could be different than default MY_POD_IP.
+        let paths_for_nqn = Subsystem::try_from_nqn(&nqn)
+            .map_err(|_| SvcError::ReplaceNqnNotFound { nqn: nqn.clone() })?;
+
+        for path in paths_for_nqn {
+            let path_transport = TrType::try_from(path.clone().transport)
+                .map_err(|_| SvcError::InvalidArguments {})?;
+            if parsed_uri.transport().ne(&path_transport) {
+                let p_disconn = path.clone();
+                tokio::task::block_in_place(move || {
+                    // Remove the subsystem.
+                    if let Err(error) = p_disconn.disconnect() {
+                        tracing::error!(?p_disconn, ?error, "Failed to disconnect stale NVMe path");
+                    } else {
+                        tracing::info!(?p_disconn, "Stale NVMe path successfully disconnected");
+                    }
+                });
+            }
         }
 
         // Check if the NVMe subsystem already exists to not