#!/bin/bash
# deploy.sh — Provisions remote hosts, copies sources, and launches Docker containers
# for mscclpp CI/CD test environments.
#
# Usage: deploy.sh <test_name> [ib_environment] [platform] [container_name] [sglang_image]
#   test_name       : Test suite to deploy (e.g. single-node-test, nccltest-single-node)
#   ib_environment  : Enable InfiniBand networking (default: true)
#   platform        : Target GPU platform — "cuda" or "rocm" (default: cuda)
#   container_name  : Docker container name (default: mscclpp-test)
#   sglang_image    : Docker image used for the SGLang test container
#                     (default: lmsysorg/sglang:latest). Only used when
#                     container_name is "sglang-mscclpp-test".

set -ex

TEST_NAME=$1
IB_ENVIRONMENT="${2:-true}"
PLATFORM="${3:-cuda}"
CONTAINER_NAME="${4:-mscclpp-test}"
SGLANG_IMAGE="${5:-lmsysorg/sglang:latest}"

KeyFilePath=${SSHKEYFILE_SECUREFILEPATH}
ROOT_DIR="${SYSTEM_DEFAULTWORKINGDIRECTORY}/"
DST_DIR="/tmp/mscclpp"

if [ "${TEST_NAME}" == "nccltest-single-node" ] || [ "${TEST_NAME}" == "single-node-test" ]; then
  HOSTFILE="${SYSTEM_DEFAULTWORKINGDIRECTORY}/test/deploy/hostfile_ci"
else
  HOSTFILE="${SYSTEM_DEFAULTWORKINGDIRECTORY}/test/deploy/hostfile"
fi

SSH_OPTION="StrictHostKeyChecking=no"

chmod 400 ${KeyFilePath}
ssh-keygen -t rsa -f sshkey -P ""

while true; do
  set +e
  parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION "hostname"
  if [ $? -eq 0 ]; then
    break
  fi
  echo "Waiting for sshd to start..."
  sleep 5
done
set -e

parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION "sudo rm -rf ${DST_DIR}"
tar czf /tmp/mscclpp.tar.gz -C ${ROOT_DIR} .
parallel-scp -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION /tmp/mscclpp.tar.gz /tmp/mscclpp.tar.gz
parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION \
  "sudo mkdir -p ${DST_DIR} && sudo tar xzf /tmp/mscclpp.tar.gz -C ${DST_DIR} && sudo rm -f /tmp/mscclpp.tar.gz"
rm -f /tmp/mscclpp.tar.gz

if [ "${PLATFORM}" == "rocm" ]; then
  parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION "sudo modprobe amdgpu"
fi

# Install GDRCopy kernel module on host VMs (CUDA only)
GDRCOPY_VERSION="2.5.2"
if [ "${PLATFORM}" == "cuda" ]; then
  parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION \
    "if lsmod | grep -q gdrdrv; then
      echo 'gdrdrv module already loaded'
    else
      set -e
      sudo apt-get update -y && sudo apt-get install -y build-essential devscripts debhelper check libsubunit-dev fakeroot pkg-config dkms
      cd /tmp && wget -q https://github.com/NVIDIA/gdrcopy/archive/refs/tags/v${GDRCOPY_VERSION}.tar.gz -O gdrcopy.tar.gz
      tar xzf gdrcopy.tar.gz && cd gdrcopy-${GDRCOPY_VERSION}/packages
      CUDA=/usr/local/cuda ./build-deb-packages.sh
      sudo dpkg -i gdrdrv-dkms_${GDRCOPY_VERSION}*.deb
      sudo modprobe gdrdrv
      rm -rf /tmp/gdrcopy.tar.gz /tmp/gdrcopy-${GDRCOPY_VERSION}
    fi"
fi

if [ "${CONTAINER_NAME}" == "sglang-mscclpp-test" ]; then
  # force to pull the latest image
  parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION \
    "sudo docker pull ${SGLANG_IMAGE}"

  parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION \
    "sudo docker run --rm -itd --name=${CONTAINER_NAME} --privileged --net=host --ipc=host --gpus=all -w /root -v ${DST_DIR}:/root/mscclpp --entrypoint /bin/bash ${SGLANG_IMAGE}"
else
  # force to pull the latest image
  parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION \
    "sudo docker pull ${CONTAINERIMAGE}"

  # Set GPU passthrough flags based on platform
  LAUNCH_OPTION="--gpus=all"
  if [ "${PLATFORM}" == "rocm" ]; then
    LAUNCH_OPTION="--device=/dev/kfd --device=/dev/dri --group-add=video"
  fi

  if [ "${IB_ENVIRONMENT}" == "true" ]; then
    # InfiniBand: use --privileged for RDMA device access
    parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION \
      "sudo docker run --rm -itd --privileged --net=host --ipc=host ${LAUNCH_OPTION} \
      -w /root -v ${DST_DIR}:/root/mscclpp -v /opt/microsoft:/opt/microsoft --ulimit memlock=-1:-1 --name=${CONTAINER_NAME} \
      --entrypoint /bin/bash ${CONTAINERIMAGE}"
  else
    # Non-IB: grant SYS_ADMIN and disable seccomp instead of full --privileged
    parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION \
      "sudo docker run --rm -itd --net=host --ipc=host ${LAUNCH_OPTION} --cap-add=SYS_ADMIN --security-opt seccomp=unconfined \
      -w /root -v ${DST_DIR}:/root/mscclpp -v /opt/microsoft:/opt/microsoft --ulimit memlock=-1:-1 --name=${CONTAINER_NAME} \
      --entrypoint /bin/bash ${CONTAINERIMAGE}"
  fi
fi

parallel-ssh -i -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" -O $SSH_OPTION \
  "sudo docker exec -t --user root ${CONTAINER_NAME} bash '/root/mscclpp/test/deploy/setup.sh' ${PLATFORM}"