CP-308863: Count vGPU migrations

This metric represents the rate of VM migrations with vGPUs per second.
The total count can be calculated as AVERAGE * seconds. For example,
for one-day data granularity, the total count for one day is
AVERAGE * 86400, in which 86400 is the seconds of one day.

Signed-off-by: Ming Lu <[email protected]>

Author: minglumlu

ocaml/xapi/message_forwarding.ml

@@ -2501,6 +2501,8 @@ functor
             let snapshot = Db.VM.get_record ~__context ~self:vm in
             reserve_memory_for_vm ~__context ~vm ~host ~snapshot
               ~host_op:`vm_migrate (fun () ->
+                if Db.VM.get_VGPUs ~__context ~self:vm <> [] then
+                  Xapi_stats.incr_pool_vgpu_migration_count () ;
                 forward_vm_op ~local_fn ~__context ~vm ~remote_fn
             )
         ) ;
@@ -2622,6 +2624,8 @@ functor
                   assert_can_migrate ~__context ~vm ~dest ~live ~vdi_map
                     ~vif_map ~vgpu_map ~options
               ) ;
+              if vgpu_map <> [] then
+                Xapi_stats.incr_pool_vgpu_migration_count () ;
               forward_migrate_send ()
           )
         in

ocaml/xapi/xapi_stats.ml

@@ -16,7 +16,16 @@ module D = Debug.Make (struct let name = "xapi_stats" end)
 
 let with_lock = Xapi_stdext_threads.Threadext.Mutex.execute
 
+let pool_vgpu_migration_count : int Atomic.t = Atomic.make 0
+
+let incr_pool_vgpu_migration_count () = Atomic.incr pool_vgpu_migration_count
+
+let stop_watch = ref (Mtime_clock.counter ())
+
 let generate_master_stats ~__context =
+  let elapsed = Mtime_clock.count !stop_watch |> Clock.Timer.span_to_s in
+  (* It's fine if counters change between here and collecting *)
+  stop_watch := Mtime_clock.counter () ;
   let session_count =
     Db.Session.get_all ~__context |> List.length |> Int64.of_int
   in
@@ -44,7 +53,24 @@ let generate_master_stats ~__context =
         ~min:0.0 ~units:"sessions/s" ()
     )
   in
-  [session_count_ds; task_count_ds; session_count_change_ds]
+  let vgpu_migration_rate =
+    let cnt = Atomic.exchange pool_vgpu_migration_count 0 in
+    float_of_int cnt /. elapsed
+  in
+  let vgpu_migration_count_ds =
+    ( Rrd.Host
+    , Ds.ds_make ~name:"pool_vgpu_migration_rate"
+        ~description:"Number of vGPU migrations occurred per second"
+        ~value:(Rrd.VT_Float vgpu_migration_rate) ~ty:Rrd.Absolute ~default:true
+        ~min:0. ~units:"migrations/s" ()
+    )
+  in
+  [
+    session_count_ds
+  ; task_count_ds
+  ; session_count_change_ds
+  ; vgpu_migration_count_ds
+  ]
 
 let gc_debug = ref true
 

ocaml/xapi/xapi_stats.mli

@@ -18,3 +18,6 @@ val start : unit -> unit
 
 val stop : unit -> unit
 (** Stop the stats reporting thread. *)
+
+val incr_pool_vgpu_migration_count : unit -> unit
+(** Increments the pool_vgpu_migration_count by 1 . *)