From 0c5883bbce990809800106eaad2512f9a6d1b7eb Mon Sep 17 00:00:00 2001
From: qbisi <qbisicwate@gmail.com>
Date: Mon, 21 Oct 2024 07:20:30 +0800
Subject: [PATCH] mpiCheckPhaseHook: add parameters to bypass errors in sandbox

---
 .../setup-hooks/mpi-check-hook/default.nix            |  4 ++++
 .../setup-hooks/mpi-check-hook/mpi-check-hook.sh      | 11 +++++++++++
 .../setup-hooks/mpi-check-hook/topology.xml           | 10 ++++++++++
 3 files changed, 25 insertions(+)
 create mode 100644 pkgs/build-support/setup-hooks/mpi-check-hook/topology.xml

diff --git a/pkgs/build-support/setup-hooks/mpi-check-hook/default.nix b/pkgs/build-support/setup-hooks/mpi-check-hook/default.nix
index 2834cfcc44ff..a49e8e338bea 100644
--- a/pkgs/build-support/setup-hooks/mpi-check-hook/default.nix
+++ b/pkgs/build-support/setup-hooks/mpi-check-hook/default.nix
@@ -2,4 +2,8 @@
 
 makeSetupHook {
   name = "mpi-checkPhase-hook";
+
+  substitutions = {
+    topology = ./topology.xml;
+  };
 } ./mpi-check-hook.sh
diff --git a/pkgs/build-support/setup-hooks/mpi-check-hook/mpi-check-hook.sh b/pkgs/build-support/setup-hooks/mpi-check-hook/mpi-check-hook.sh
index 41b6d864c8a5..d7275b62244d 100644
--- a/pkgs/build-support/setup-hooks/mpi-check-hook/mpi-check-hook.sh
+++ b/pkgs/build-support/setup-hooks/mpi-check-hook/mpi-check-hook.sh
@@ -44,6 +44,17 @@ setupMpiCheck() {
       # Disable CPU pinning
       export OMPI_MCA_hwloc_base_binding_policy=none
       export PRTE_MCA_hwloc_default_binding_policy=none
+
+      # OpenMPI get confused by the sandbox environment and spew errors like this (both to stdout and stderr):
+      #     [hwloc/linux] failed to find sysfs cpu topology directory, aborting linux discovery.
+      #     [1729458724.473282] [localhost:78   :0]       tcp_iface.c:893  UCX  ERROR scandir(/sys/class/net) failed: No such file or directory
+      # These messages contaminate test output, which makes the difftest to fail.
+      # The solution is to use a preset cpu topology file and disable ucx model.
+
+      # Disable sysfs cpu topology directory discovery.
+      export PRTE_MCA_hwloc_use_topo_file="@topology@"
+      # Use the network model ob1 instead of ucx.
+      export OMPI_MCA_pml=ob1
       ;;
     MPICH)
       # Fix to make mpich run in a sandbox
diff --git a/pkgs/build-support/setup-hooks/mpi-check-hook/topology.xml b/pkgs/build-support/setup-hooks/mpi-check-hook/topology.xml
new file mode 100644
index 000000000000..616ed839f004
--- /dev/null
+++ b/pkgs/build-support/setup-hooks/mpi-check-hook/topology.xml
@@ -0,0 +1,10 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE topology SYSTEM "hwloc2.dtd">
+<topology version="2.0">
+  <object type="Machine" os_index="0" cpuset="0x00000001" complete_cpuset="0x00000001" allowed_cpuset="0x00000001" nodeset="0x00000001" complete_nodeset="0x00000001" allowed_nodeset="0x00000001" gp_index="1">
+    <object type="Core" cpuset="0x00000001" complete_cpuset="0x00000001" nodeset="0x00000001" complete_nodeset="0x00000001" gp_index="2">
+      <object type="NUMANode" os_index="0" cpuset="0x00000001" complete_cpuset="0x00000001" nodeset="0x00000001" complete_nodeset="0x00000001" gp_index="4"/>
+      <object type="PU" os_index="0" cpuset="0x00000001" complete_cpuset="0x00000001" nodeset="0x00000001" complete_nodeset="0x00000001" gp_index="3"/>
+    </object>
+  </object>
+</topology>