CA-380551: HA: assert that the HA SR is big enough for BOTH the state…

…file AND the redo log

Previously only the statefile size was checked, and available space for the database VDI was only checked
during DR in the metadata VDI code, but not during HA in the redolog code.

With this change XenCenter is able to grey out SRs that are too small.

After this change an HA heartbeat SR will have a physical utilization of `3997171712`, which is as close
as we can get to 4GB (the next 4MiB increment would go over 4GB).

Signed-off-by: Edwin Török <[email protected]>

ocaml/database/redo_log.ml

@@ -48,8 +48,17 @@ let mib megabytes =
   let ( ** ) = Int64.mul in
   Int64.of_int megabytes ** 1024L ** 1024L
 
-(* Make sure we have plenty of room for the database *)
-let minimum_vdi_size = mib 4096
+(* Make sure we have plenty of room for the database
+   There is also a 4MiB statefile, so make the sum be 4GB, which is easier to document.
+*)
+let minimum_vdi_size =
+  let ( // ) = Int64.div and ( ** ) = Int64.mul and ( -- ) = Int64.sub in
+  let align = mib 4 in
+  ((4_000_000_000L // align)
+  -- 2L
+     (* -2 because we also need room for a statefile, and an 'empty' SR seems to have a utilization of 4MiB *)
+  )
+  ** align
 
 let redo_log_sm_config = [("type", "raw")]
 

ocaml/xapi/xha_statefile.ml

@@ -29,34 +29,42 @@ open Client
 
 (** Return the minimum size of an HA statefile, as of
     XenServer HA state-file description vsn 1.3 *)
-let minimum_size =
+let minimum_statefile_size =
   let ( ** ) = Int64.mul and ( ++ ) = Int64.add in
   let global_section_size = 4096L
   and host_section_size = 4096L
   and maximum_number_of_hosts = 64L in
   global_section_size ++ (maximum_number_of_hosts ** host_section_size)
 
-let ha_fits_sr ~__context ~sr ~typ ~minimum_size =
+let round_to ~align n = Int64.(div (add n @@ sub align 1L) align |> mul align)
+
+(* SM doesn't actually allow us to create VDIs smaller than 4MiB, so we need to round up *)
+let minimum_sr_size =
+  [minimum_statefile_size; Redo_log.minimum_vdi_size]
+  |> List.map @@ round_to ~align:Int64.(shift_left 1L 22)
+  |> List.fold_left Int64.add Int64.zero
+
+let ha_fits_sr ~__context ~what ~sr ~typ ~minimum_size =
   let ha_fits self =
     Db.VDI.get_type ~__context ~self = typ
     && Db.VDI.get_virtual_size ~__context ~self >= minimum_size
   in
   match List.filter ha_fits (Db.SR.get_VDIs ~__context ~self:sr) with
   | x :: _ ->
-      debug "Would re-use existing statefile: %s"
+      debug "Would re-use existing %s: %s" what
         (Db.VDI.get_uuid ~__context ~self:x) ;
       Some x
   | [] ->
-      debug
-        "no suitable existing statefile found; would have to create a fresh one" ;
+      debug "no suitable existing %s found; would have to create a fresh one"
+        what ;
       let self = sr in
       let size = Db.SR.get_physical_size ~__context ~self in
       let utilisation = Db.SR.get_physical_utilisation ~__context ~self in
       let free_space = Int64.sub size utilisation in
-      if free_space < minimum_size then (
+      if free_space < minimum_sr_size then (
         let sr = Ref.string_of sr in
         info "%s: SR %s size=%Ld utilisation=%Ld free=%Ld needed=%Ld"
-          __FUNCTION__ sr size utilisation free_space minimum_size ;
+          __FUNCTION__ sr size utilisation free_space minimum_sr_size ;
         raise
           (Api_errors.Server_error
              (Api_errors.sr_source_space_insufficient, [sr])
@@ -126,12 +134,17 @@ let check_sr_can_host_statefile ~__context ~sr ~cluster_stack =
           (Api_errors.Server_error
              (Api_errors.sr_operation_not_supported, [Ref.string_of sr])
           ) ;
-      ha_fits_sr ~__context ~sr ~minimum_size ~typ:`ha_statefile
+      ha_fits_sr ~__context ~what:"statefile" ~sr
+        ~minimum_size:minimum_statefile_size ~typ:`ha_statefile
 
 let assert_sr_can_host_statefile ~__context ~sr ~cluster_stack =
   let (_ : 'a option) =
     check_sr_can_host_statefile ~__context ~sr ~cluster_stack
   in
+  let (_ : _ option) =
+    ha_fits_sr ~__context ~what:"redo-log" ~sr
+      ~minimum_size:Redo_log.minimum_vdi_size ~typ:`redo_log
+  in
   ()
 
 let list_srs_which_can_host_statefile ~__context ~cluster_stack =
@@ -146,7 +159,7 @@ let list_srs_which_can_host_statefile ~__context ~cluster_stack =
 
 let create ~__context ~sr ~cluster_stack =
   assert_sr_can_host_statefile ~__context ~sr ~cluster_stack ;
-  let size = minimum_size in
+  let size = minimum_statefile_size in
   Helpers.call_api_functions ~__context (fun rpc session_id ->
       Client.VDI.create ~rpc ~session_id ~name_label:"Statefile for HA"
         ~name_description:"Used for storage heartbeating" ~sR:sr