CP-54065, xenopsd: use domain_claim_pages on a single node, if possible

Xen currently supports to modes to claim memory for a domain: without any node
in particular, or claim memory in a single NUMA node.

When planning a domain, return the nodes that will host the domain, and how
much memory. In the case where the domain fits in a single NUMA node, claim
pages on that node, otherwise fall back to previous behaviour.

The memory claims need to happen while the memory measurements hold valid, that
is while no VMs are started, otherwise ENOMEM might be returned.

Because the current mode is a best-effort, log when the claim does not work.

Signed-off-by: Pau Ruiz Safont <pau.ruizsafont@cloud.com>

Author: psafont

ocaml/xenopsd/lib/softaffinity.ml

@@ -26,10 +26,9 @@ let plan host nodes ~vm =
       (Fmt.to_to_string NUMARequest.pp_dump requested)
       (Fmt.to_to_string NUMAResource.pp_dump allocated) ;
     let candidate = nodes.(nodeidx) in
-    ( NUMAResource.union allocated candidate
-    , node :: picked
-    , NUMARequest.shrink requested candidate
-    )
+    (* This is where the memory allocated to the node can be calculated *)
+    let remaining_request = NUMARequest.shrink requested candidate in
+    (NUMAResource.union allocated candidate, node :: picked, remaining_request)
   in
   let plan_valid (avg, nodes) =
     let allocated, picked, remaining =
@@ -72,8 +71,8 @@ let plan host nodes ~vm =
   | None ->
       debug "No allocations possible" ;
       None
-  | Some allocated ->
+  | Some (allocated, nodes) ->
       debug "Allocated resources: %s"
         (Fmt.to_to_string NUMAResource.pp_dump allocated) ;
       assert (NUMARequest.fits vm allocated) ;
-      Some allocated.NUMAResource.affinity
+      Some (allocated.NUMAResource.affinity, nodes)

ocaml/xenopsd/lib/softaffinity.mli

@@ -14,7 +14,11 @@
 
 open Topology
 
-val plan : NUMA.t -> NUMAResource.t array -> vm:NUMARequest.t -> CPUSet.t option
+val plan :
+     NUMA.t
+  -> NUMAResource.t array
+  -> vm:NUMARequest.t
+  -> (Topology.CPUSet.t * Topology.NUMA.node list) option
 (** [plan host nodes ~vm] returns the CPU soft affinity recommended for [vm],
     Such that the memory latency between the NUMA nodes of the vCPUs is small,
     and usage of NUMA nodes is balanced.

ocaml/xenopsd/lib/topology.ml

@@ -298,7 +298,7 @@ module NUMA = struct
       None
     else (
       List.iter (fun (Node n) -> t.node_usage.(n) <- t.node_usage.(n) + 1) nodes ;
-      Some result
+      Some (result, nodes)
     )
 
   let pp_dump_node = Fmt.(using (fun (Node x) -> x) int)

ocaml/xenopsd/lib/topology.mli

@@ -150,7 +150,10 @@ module NUMA : sig
       NUMA nodes > 16 it limits the length of the sequence to [n+65520], to
       avoid exponential blowup. *)
 
-  val choose : t -> (node list * NUMAResource.t) Seq.t -> NUMAResource.t option
+  val choose :
+       t
+    -> (node list * NUMAResource.t) Seq.t
+    -> (NUMAResource.t * node list) option
   (** [choose t resources] will choose one NUMA node deterministically, trying
       to keep the overall NUMA node usage balanced *)
 

ocaml/xenopsd/test/test_topology.ml

@@ -210,26 +210,28 @@ let test_allocate ?(mem = default_mem) (expected_cores, h) ~vms () =
          match Softaffinity.plan h nodes ~vm with
          | None ->
              Alcotest.fail "No NUMA plan"
-         | Some plan ->
-             D.debug "NUMA allocation succeeded for VM %d: %s" i
-               (Fmt.to_to_string CPUSet.pp_dump plan) ;
+         | Some (cpu_plan, mem_plan) ->
+             D.debug
+               "NUMA allocation succeeded for VM %d: [CPUS: %s]; [nodes: %s]" i
+               (Fmt.to_to_string CPUSet.pp_dump cpu_plan)
+               (Fmt.to_to_string Fmt.(Dump.list NUMA.pp_dump_node) mem_plan) ;
              let usednodes =
-               plan
+               cpu_plan
                |> CPUSet.elements
                |> List.map (NUMA.node_of_cpu h)
                |> List.sort_uniq compare
                |> List.to_seq
              in
              let costs_numa_aware =
-               vm_access_costs h plans (vm_cores, usednodes, plan)
+               vm_access_costs h plans (vm_cores, usednodes, cpu_plan)
              in
              let costs_default =
                vm_access_costs h plans (vm_cores, NUMA.nodes h, NUMA.all_cpus h)
              in
              cost_not_worse ~default:costs_default costs_numa_aware ;
              ( costs_default :: costs_old
              , costs_numa_aware :: costs_new
-             , ((vm_cores, List.of_seq usednodes), plan) :: plans
+             , ((vm_cores, List.of_seq usednodes), cpu_plan) :: plans
              )
        )
        ([], [], [])