Merge pull request xapi-project#232 from lindig/ely-bugfix

CA-229176 fix PVS Proxy status cache inconsistency: invalidate during attach

Author: robhoes

ocaml/xapi/monitor_dbcalls_cache.ml

@@ -46,6 +46,11 @@ let clear_cache_for_vm ~vm_uuid =
   Mutex.execute vm_memory_cached_m
     (fun _ -> Hashtbl.remove vm_memory_cached vm_uuid)
 
+(** [clear_pvs_status_cache] removes the cache entry for [vm_uuid] *)
+let clear_pvs_status_cache ~vm_uuid =
+  Mutex.execute pvs_proxy_cached_m
+    (fun _ -> Hashtbl.remove pvs_proxy_cached vm_uuid)
+
 (** Clear the whole cache. This forces fresh properties to be written into
  * xapi's database. *)
 let clear_cache () =

ocaml/xapi/monitor_pvs_proxy.ml

@@ -32,6 +32,18 @@ let find_rrd_files () =
   |> Array.to_list
   |> List.filter (String.startswith metrics_prefix)
 
+  (* The PVS Proxy status cache [pvs_proxy_cached] contains the status
+   * entries from PVS Proxies as reported via RRD. When the status
+   * changes, it is updated in the xapi database. However: The xapi
+   * databse is only updated for proxies that are currently attached.
+   * This can lead to divergence between the cache and the database,
+   * leading to error CA-229176. When the PVS Proxy is attached in
+   * xapi_xenops.ml, the cache entry for the PVS Proxy is invalidated
+   * such that it is picked up again and updated in the xapi database.
+   * Inconsistencies are thus limited to the time between when a PVS
+   * Proxy starts reporting its status and when it is attached.
+   *)
+
 let get_changes () =
   List.iter (fun filename ->
       try

ocaml/xapi/xapi_xenops.ml

@@ -1709,9 +1709,11 @@ let update_vif ~__context id =
                 | None -> ()
                 | Some proxy ->
                   debug "xenopsd event: Updating PVS_proxy for VIF %s.%s currently_attached <- %b" (fst id) (snd id) state.pvs_rules_active;
-                  if state.pvs_rules_active then
-                    Db.PVS_proxy.set_currently_attached ~__context ~self:proxy ~value:true
-                  else
+                  if state.pvs_rules_active then begin
+                    Db.PVS_proxy.set_currently_attached ~__context ~self:proxy ~value:true;
+                    (* force status to be read again by invalidating cache *)
+                    Monitor_dbcalls_cache.clear_pvs_status_cache (fst id)
+                  end else
                     Pvs_proxy_control.clear_proxy_state ~__context vif proxy
                );
                debug "xenopsd event: Updating VIF %s.%s currently_attached <- %b" (fst id) (snd id) (state.plugged || state.active);