Merge pull request #51 from psafont/nolog

.github/workflows/main.yml

@@ -7,39 +7,45 @@ on:
 jobs:
   ocaml-test:
     name: Ocaml tests
-    runs-on: ubuntu-20.04
-    env:
-      package: "gpumon"
+    runs-on: ubuntu-22.04
 
     steps:
       - name: Checkout code
-        uses: actions/checkout@v2
+        uses: actions/checkout@v4
 
       - name: Pull configuration from xs-opam
         run: |
           curl --fail --silent https://raw.githubusercontent.com/xapi-project/xs-opam/master/tools/xs-opam-ci.env | cut -f2 -d " " > .env
 
       - name: Load environment file
         id: dotenv
-        uses: falti/[email protected]
+        uses: falti/dotenv-action@v1
+
+      - name: Update Ubuntu repositories
+        shell: bash
+        run: sudo apt-get update
 
       - name: Use ocaml
-        uses: avsm/setup-ocaml@v1
+        uses: ocaml/setup-ocaml@v2
         with:
-          ocaml-version: ${{ steps.dotenv.outputs.ocaml_version_full }}
-          opam-repository: ${{ steps.dotenv.outputs.repository }}
+          ocaml-compiler: ${{ steps.dotenv.outputs.ocaml_version_full }}
+          opam-repositories: |
+            xs-opam: ${{ steps.dotenv.outputs.repository }}
+          dune-cache: true
+        env:
+          DUNE_CACHE_STORAGE_MODE: copy
 
       - name: Install dependencies
-        run: |
-          opam pin add . --no-action
-          opam depext -u ${{ env.package }}
-          opam install ${{ env.package }} --deps-only --with-test -v
+        shell: bash
+        run: opam install . --deps-only --with-test -v
 
       - name: Build
+        shell: bash
         run: |
           opam exec -- cp mocks/mock.ml lib/nvml.ml
           opam exec -- cp mocks/mock.c stubs/nvml_stubs.c
           opam exec -- make
 
       - name: Run tests
+        shell: bash
         run: opam exec -- make test

gpumon/gpumon.ml

@@ -11,7 +11,7 @@ let vgpu_config_dir = "/usr/share/nvidia/vgpu"
 (* acquire NVML interface but give up after timeout. Note that this does
    not attach the NVML libarary but it waits for someone else to attach it
    if necessary. *)
-let get_nvml_or_wait ?(timeout = 30.0) () =
+let get_nvml_or_wait ~log ?(timeout = 30.0) () =
   let rec loop delay waited =
     match Nvml.NVML.get () with
     | Some _ as interface ->
@@ -28,25 +28,28 @@ let get_nvml_or_wait ?(timeout = 30.0) () =
   | true ->
       loop 2.0 0.0
   | false ->
-      Process.D.info "%s not found - assuming NVML library unavailable"
-        vgpu_config_dir ;
+      if log then
+        Process.D.info "%s not found - assuming NVML library unavailable"
+          vgpu_config_dir ;
       None
 
 (* Like get_nvml_or_wait but keep looping. Only call this in a thread
    that does not block the main interaction like the RPC server. Again,
    this is not attaching the library but waits for it to be attached by
    some other thread. *)
 let get_nvml_or_wait_forever () =
-  let rec try_again_after delay =
-    match get_nvml_or_wait () with
+  let rec try_again_after ~log delay =
+    match get_nvml_or_wait ~log () with
     | Some interface ->
         interface
     | None ->
-        Process.D.info "Failed to open NVML interface - retrying in %4.0f" delay ;
+        if log then
+          Process.D.info "Failed to open NVML interface - retrying in %4.0f"
+            delay ;
         Thread.delay delay ;
-        try_again_after (min (delay *. 1.5) (20.0 *. 60.0))
+        try_again_after ~log:false (min (delay *. 1.5) (20.0 *. 60.0))
   in
-  try_again_after 60.0
+  try_again_after ~log:true 60.0
 
 let default_config : (int32 * Gpumon_config.config) list =
   let open Gpumon_config in
@@ -359,7 +362,7 @@ let () =
     exit 0
   in
   let module Gpumon_server = Gpumon_server.Make (struct
-    let interface () = get_nvml_or_wait ()
+    let interface () = get_nvml_or_wait ~log:true ()
   end) in
   (* create daemon module to bind server call declarations to implementations *)
   let module Daemon = Make (Gpumon_server) in

lib/gpumon_config.ml

@@ -142,8 +142,8 @@ let device_type_of_v2_format = function
       ( if List.mem_assoc "subsystem_device_id" dict then
           List.assoc "subsystem_device_id" dict |> unbox_string >>= id_of_string
           >>= fun id -> Ok (Match id)
-      else
-        Ok Any
+        else
+          Ok Any
       )
       (* Try to read the list of metrics. *)
       >>= fun subsystem_device_id ->