Merge c4c5c78 into 53404c7

ocaml/idl/datamodel_errors.ml

@@ -1143,7 +1143,9 @@ let _ =
   error Api_errors.cluster_host_not_joined [ "cluster_host" ]
     ~doc:"Cluster_host operation failed as the cluster_host has not joined the cluster." ();
   error Api_errors.cluster_host_is_last ["cluster_host"]
-    ~doc:"The last cluster host cannot be destroyed. Destroy the cluster instead" ()
+    ~doc:"The last cluster host cannot be destroyed. Destroy the cluster instead" ();
+  error Api_errors.no_cluster_hosts_reachable ["cluster"]
+    ~doc:"No other cluster host was reachable when joining" ()
 
 
 

ocaml/xapi-consts/api_errors.ml

@@ -616,3 +616,4 @@ let cluster_stack_in_use = "CLUSTER_STACK_IN_USE"
 let invalid_cluster_stack = "INVALID_CLUSTER_STACK"
 let pif_not_attached_to_host = "PIF_NOT_ATTACHED_TO_HOST"
 let cluster_host_not_joined = "CLUSTER_HOST_NOT_JOINED"
+let no_cluster_hosts_reachable = "NO_CLUSTER_HOSTS_REACHABLE"

ocaml/xapi/helpers.ml

@@ -63,10 +63,16 @@ let get_pool ~__context = List.hd (Db.Pool.get_all ~__context)
 let get_master ~__context =
   Db.Pool.get_master ~__context ~self:(get_pool ~__context)
 
-let get_management_iface_is_connected ~__context =
+let get_iface_is_connected ~__context iface =
   let dbg = Context.string_of_task __context in
-  Net.Bridge.get_physical_interfaces dbg (Xapi_inventory.lookup Xapi_inventory._management_interface)
-  |> List.exists (fun name -> Net.Interface.is_connected dbg name)
+  Net.Bridge.get_physical_interfaces dbg iface
+  |> List.exists (fun name ->
+    debug "Checking interface %s" name;
+    Net.Interface.is_connected dbg name)
+
+let get_management_iface_is_connected ~__context =
+  Xapi_inventory.lookup Xapi_inventory._management_interface
+  |> get_iface_is_connected ~__context
 
 let get_primary_ip_addr ~__context iface primary_address_type =
   if iface = "" then

ocaml/xapi/xapi_cluster_host.ml

@@ -72,11 +72,18 @@ let join_internal ~__context ~self =
       let cluster = Db.Cluster_host.get_cluster ~__context ~self in
       let cluster_token = Db.Cluster.get_cluster_token ~__context ~self:cluster in
       let ip = ip_of_pif (pIF, Db.PIF.get_record ~__context ~self:pIF) in
-      let ip_list = List.map (fun self ->
-        let p_ref = Db.Cluster_host.get_PIF ~__context ~self in
-        let p_rec = Db.PIF.get_record ~__context ~self:p_ref in
-          ip_of_pif (p_ref,p_rec)
+      let ip_list = Xapi_stdext_std.Listext.List.filter_map (fun self ->
+          let p_ref = Db.Cluster_host.get_PIF ~__context ~self in
+          let p_rec = Db.PIF.get_record ~__context ~self:p_ref in
+          (* parallel join: some hosts may not have an IP yet *)
+          try
+            let other_ip = ip_of_pif (p_ref,p_rec) in
+            if other_ip <> ip then Some other_ip
+            else None
+          with _ -> None
         ) (Db.Cluster.get_cluster_hosts ~__context ~self:cluster) in
+      if ip_list = [] then
+        raise Api_errors.(Server_error (no_cluster_hosts_reachable, [Ref.string_of cluster]));
 
       debug "Enabling clusterd and joining cluster_host %s" (Ref.string_of self);
       Xapi_clustering.Daemon.enable ~__context;

ocaml/xapi/xapi_mgmt_iface.ml

@@ -160,31 +160,33 @@ let rebind ~__context =
 let ip_mutex = Mutex.create ()
 let ip_cond = Condition.create ()
 
-let wait_for_management_ip ~__context =
-  let ip = ref (match Helpers.get_management_ip_addr ~__context with Some x -> x | None -> "") in
-  let is_connected = ref (Helpers.get_management_iface_is_connected ~__context) in
-  Mutex.execute ip_mutex
-    (fun () -> while !ip = "" && !is_connected = false do
-          Condition.wait ip_cond ip_mutex;
-          ip := (match Helpers.get_management_ip_addr ~__context with Some x -> x | None -> "");
-          is_connected := (Helpers.get_management_iface_is_connected ~__context)
-        done);
-  !ip
+let wait_for_ip get_ip is_connected =
+  let rec loop () =
+    let ip = get_ip () in
+    let connected = is_connected () in
+    if ip = "" || not connected then begin
+      debug "wait_for_ip (IP=%s, connected:%b), waiting" ip connected;
+      Condition.wait ip_cond ip_mutex;
+      loop ()
+    end
+    else ip
+  in
+  Mutex.execute ip_mutex loop
 
-let has_carrier ~__context ~self =
-  let metrics = Db.PIF.get_metrics ~__context ~self in
-  Db.PIF_metrics.get_carrier ~__context ~self:metrics
+let wait_for_management_ip ~__context =
+  let get_ip () = match Helpers.get_management_ip_addr ~__context with Some x -> x | None -> "" in
+  let is_connected () = Helpers.get_management_iface_is_connected ~__context in
+  wait_for_ip get_ip is_connected
 
 (* CA-280237: Called in startup sequence after creating cluster_hosts *)
 let wait_for_clustering_ip ~__context ~(self : API.ref_Cluster_host) =
   let pIF = Db.Cluster_host.get_PIF ~__context ~self in
-  let iP = ref (Db.PIF.get_IP ~__context ~self:pIF) in
-  Mutex.execute ip_mutex (* Don't return until PIF is plugged AND has a valid IP *)
-    (fun () -> while !iP="" || not (has_carrier ~__context ~self:pIF) do
-        Condition.wait ip_cond ip_mutex;
-        iP := Db.PIF.get_IP ~__context ~self:pIF
-      done);
-  !iP
+  let network = Db.PIF.get_network ~__context ~self:pIF in
+  let interface = Db.Network.get_bridge ~__context ~self:network in
+  debug "Waiting for clustering IP on interface %s (%s)" interface (Ref.string_of pIF);
+  let get_ip () = Db.PIF.get_IP ~__context ~self:pIF in
+  let is_connected () = Helpers.get_iface_is_connected ~__context interface in
+  wait_for_ip get_ip is_connected
 
 let on_dom0_networking_change ~__context =
   debug "Checking to see if hostname or management IP has changed";