Mux mirror failure check for SXM (#6439)

Continuation of #6434, more mutiplexing for SXM, this time the mirror
status checking logic.

No functional change.

More to come...

Author: Vincent-lau

ocaml/xapi-idl/storage/storage_interface.ml

@@ -1179,6 +1179,27 @@ module StorageAPI (R : RPC) = struct
       let receive_cancel2 =
         declare "DATA.MIRROR.receive_cancel2" []
           (dbg_p @-> id_p @-> url_p @-> verify_dest_p @-> returning unit_p err)
+
+      let pre_deactivate_hook =
+        declare "DATA.MIRROR.pre_deactivate_hook" []
+          (dbg_p @-> dp_p @-> sr_p @-> vdi_p @-> returning unit_p err)
+
+      let has_mirror_failed =
+        let mirror_failed_p =
+          Param.mk ~name:"mirror_failed_p" ~description:[] Types.bool
+        in
+        declare "DATA.MIRROR.has_mirror_failed" []
+          (dbg_p @-> id_p @-> sr_p @-> returning mirror_failed_p err)
+
+      let list =
+        let result_p =
+          Param.mk ~name:"mirrors" TypeCombinators.(list (pair Mirror.(id, t)))
+        in
+        declare "DATA.MIRROR.list" [] (dbg_p @-> returning result_p err)
+
+      let stat =
+        let result_p = Param.mk ~name:"result" Mirror.t in
+        declare "DATA.MIRROR.stat" [] (dbg_p @-> id_p @-> returning result_p err)
     end
   end
 
@@ -1285,6 +1306,16 @@ module type MIRROR = sig
     -> url:string
     -> verify_dest:bool
     -> unit
+
+  val pre_deactivate_hook :
+    context -> dbg:debug_info -> dp:dp -> sr:sr -> vdi:vdi -> unit
+
+  val has_mirror_failed :
+    context -> dbg:debug_info -> mirror_id:Mirror.id -> sr:Sr.t -> bool
+
+  val list : context -> dbg:debug_info -> (Mirror.id * Mirror.t) list
+
+  val stat : context -> dbg:debug_info -> id:Mirror.id -> Mirror.t
 end
 
 module type Server_impl = sig
@@ -1759,6 +1790,14 @@ module Server (Impl : Server_impl) () = struct
         Impl.DATA.MIRROR.receive_finalize2 () ~dbg ~mirror_id ~sr ~url
           ~verify_dest
     ) ;
+    S.DATA.MIRROR.pre_deactivate_hook (fun dbg dp sr vdi ->
+        Impl.DATA.MIRROR.pre_deactivate_hook () ~dbg ~dp ~sr ~vdi
+    ) ;
+    S.DATA.MIRROR.has_mirror_failed (fun dbg mirror_id sr ->
+        Impl.DATA.MIRROR.has_mirror_failed () ~dbg ~mirror_id ~sr
+    ) ;
+    S.DATA.MIRROR.list (fun dbg -> Impl.DATA.MIRROR.list () ~dbg) ;
+    S.DATA.MIRROR.stat (fun dbg id -> Impl.DATA.MIRROR.stat () ~dbg ~id) ;
     S.DATA.import_activate (fun dbg dp sr vdi vm ->
         Impl.DATA.import_activate () ~dbg ~dp ~sr ~vdi ~vm
     ) ;

ocaml/xapi-idl/storage/storage_skeleton.ml

@@ -182,6 +182,16 @@ module DATA = struct
 
     let receive_cancel2 ctx ~dbg ~mirror_id ~url ~verify_dest =
       u "DATA.MIRROR.receive_cancel2"
+
+    let pre_deactivate_hook ctx ~dbg ~dp ~sr ~vdi =
+      u "DATA.MIRROR.pre_deactivate_hook"
+
+    let has_mirror_failed ctx ~dbg ~mirror_id ~sr =
+      u "DATA.MIRROR.has_mirror_failed"
+
+    let list ctx ~dbg = u "DATA.MIRROR.list"
+
+    let stat ctx ~dbg ~id = u "DATA.MIRROR.stat"
   end
 end
 

ocaml/xapi-storage-cli/main.ml

@@ -149,7 +149,7 @@ let string_of_file filename =
 
 let mirror_list common_opts =
   wrap common_opts (fun () ->
-      let list = Storage_migrate.list ~dbg in
+      let list = Client.DATA.MIRROR.list dbg in
       List.iter
         (fun (id, status) -> Printf.printf "%s" (string_of_mirror id status))
         list

ocaml/xapi-storage-script/main.ml

@@ -1927,6 +1927,10 @@ let bind ~volume_script_dir =
   S.DATA.MIRROR.receive_finalize2 (u "DATA.MIRROR.receive_finalize2") ;
   S.DATA.MIRROR.receive_cancel (u "DATA.MIRROR.receive_cancel") ;
   S.DATA.MIRROR.receive_cancel2 (u "DATA.MIRROR.receive_cancel2") ;
+  S.DATA.MIRROR.pre_deactivate_hook (u "DATA.MIRROR.pre_deactivate_hook") ;
+  S.DATA.MIRROR.has_mirror_failed (u "DATA.MIRROR.has_mirror_failed") ;
+  S.DATA.MIRROR.list (u "DATA.MIRROR.list") ;
+  S.DATA.MIRROR.stat (u "DATA.MIRROR.stat") ;
   S.DP.create (u "DP.create") ;
   S.TASK.cancel (u "TASK.cancel") ;
   S.TASK.list (u "TASK.list") ;

ocaml/xapi/storage_access.ml

@@ -446,7 +446,7 @@ let update_task ~__context id =
 let update_mirror ~__context id =
   try
     let dbg = Context.string_of_task __context in
-    let m = Storage_migrate.stat ~dbg ~id in
+    let m = Client.DATA.MIRROR.stat dbg id in
     if m.Mirror.failed then
       debug "Mirror %s has failed" id ;
     let task = get_mirror_task id in

ocaml/xapi/storage_migrate.ml

@@ -209,28 +209,20 @@ module MigrateLocal = struct
         stop ~dbg ~id:mirror_id ;
         raise e
 
-  let stat ~dbg:_ ~id =
+  let stat ~dbg ~id =
     let recv_opt = State.find_active_receive_mirror id in
     let send_opt = State.find_active_local_mirror id in
     let copy_opt = State.find_active_copy id in
+    let sr, _vdi = State.of_mirror_id id in
     let open State in
     let failed =
       match send_opt with
       | Some send_state ->
+          let (module Migrate_Backend) = choose_backend dbg sr in
           let failed =
-            match send_state.Send_state.tapdev with
-            | Some tapdev -> (
-              try
-                let stats = Tapctl.stats (Tapctl.create ()) tapdev in
-                stats.Tapctl.Stats.nbd_mirror_failed = 1
-              with _ ->
-                debug "Using cached copy of failure status" ;
-                send_state.Send_state.failed
-            )
-            | None ->
-                false
+            Migrate_Backend.has_mirror_failed () ~dbg ~mirror_id:id ~sr
           in
-          send_state.Send_state.failed <- failed ;
+          send_state.failed <- failed ;
           failed
       | None ->
           false
@@ -325,58 +317,9 @@ module MigrateLocal = struct
     State.clear ()
 end
 
-exception Timeout of Mtime.Span.t
-
-let reqs_outstanding_timeout = Mtime.Span.(150 * s)
-
-let pp_time () = Fmt.str "%a" Mtime.Span.pp
-
-(* Tapdisk should time out after 2 mins. We can wait a little longer *)
-
-let pre_deactivate_hook ~dbg:_ ~dp:_ ~sr ~vdi =
-  let open State.Send_state in
-  let id = State.mirror_id_of (sr, vdi) in
-  let start = Mtime_clock.counter () in
-  State.find_active_local_mirror id
-  |> Option.iter (fun s ->
-         (* We used to pause here and then check the nbd_mirror_failed key. Now, we poll
-            					   until the number of outstanding requests has gone to zero, then check the
-            					   status. This avoids confusing the backend (CA-128460) *)
-         try
-           match s.tapdev with
-           | None ->
-               ()
-           | Some tapdev ->
-               let open Tapctl in
-               let ctx = create () in
-               let rec wait () =
-                 let elapsed = Mtime_clock.count start in
-                 if Mtime.Span.compare elapsed reqs_outstanding_timeout > 0 then
-                   raise (Timeout elapsed) ;
-                 let st = stats ctx tapdev in
-                 if st.Stats.reqs_outstanding > 0 then (
-                   Thread.delay 1.0 ; wait ()
-                 ) else
-                   (st, elapsed)
-               in
-               let st, elapsed = wait () in
-               debug "Got final stats after waiting %a" pp_time elapsed ;
-               if st.Stats.nbd_mirror_failed = 1 then (
-                 error "tapdisk reports mirroring failed" ;
-                 s.failed <- true
-               )
-         with
-         | Timeout elapsed ->
-             error
-               "Timeout out after %a waiting for tapdisk to complete all \
-                outstanding requests"
-               pp_time elapsed ;
-             s.failed <- true
-         | e ->
-             error "Caught exception while finally checking mirror state: %s"
-               (Printexc.to_string e) ;
-             s.failed <- true
-     )
+let pre_deactivate_hook ~dbg ~dp ~sr ~vdi =
+  let (module Migrate_Backend) = choose_backend dbg sr in
+  Migrate_Backend.pre_deactivate_hook () ~dbg ~dp ~sr ~vdi
 
 let post_deactivate_hook ~sr ~vdi ~dp:_ =
   let open State.Send_state in

ocaml/xapi/storage_mux.ml

@@ -857,6 +857,22 @@ module Mux = struct
 
       let receive_cancel2 () ~dbg:_ ~mirror_id:_ ~url:_ ~verify_dest:_ =
         u __FUNCTION__
+
+      let pre_deactivate_hook _ctx ~dbg:_ ~dp:_ ~sr:_ ~vdi:_ =
+        u "DATA.MIRROR.pre_deactivate_hook"
+
+      let has_mirror_failed _ctx ~dbg:_ ~mirror_id:_ ~sr:_ =
+        u "DATA.MIRROR.has_mirror_failed"
+
+      let list () ~dbg =
+        with_dbg ~name:"DATA.MIRROR.list" ~dbg @@ fun di ->
+        info "%s dbg: %s" __FUNCTION__ dbg ;
+        Storage_migrate.list ~dbg:di.log
+
+      let stat () ~dbg ~id =
+        with_dbg ~name:"DATA.MIRROR.stat" ~dbg @@ fun di ->
+        info "%s dbg: %s mirror_id: %s" __FUNCTION__ di.log id ;
+        Storage_migrate.stat ~dbg:di.log ~id
     end
   end
 

ocaml/xapi/storage_smapiv1.ml

@@ -1157,6 +1157,14 @@ module SMAPIv1 : Server_impl = struct
 
       let receive_cancel2 _context ~dbg:_ ~mirror_id:_ ~url:_ ~verify_dest:_ =
         assert false
+
+      let pre_deactivate_hook _context ~dbg:_ ~dp:_ ~sr:_ ~vdi:_ = assert false
+
+      let has_mirror_failed _context ~dbg:_ ~mirror_id:_ ~sr:_ = assert false
+
+      let list _context ~dbg:_ = assert false
+
+      let stat _context ~dbg:_ ~id:_ = assert false
     end
   end